This PDF file contains the front matter associated with SPIE Proceedings Volume 12604, including the Title Page, Copyright information, Table of Contents, and Conference Committee information

In order to meet the requirements of environmental design, this paper adopts a three dimensional projection oriented dioramma. In this paper, a method of minimizing matching cost is proposed to find the geometric model of the best model, so as to achieve better results. By using the geometric correlation of image elements, the coordinates of three dimensional space are reversed, and the accuracy of reconstruction is optimized. Through the simulation and data analysis, it is proved that the proposed method is much better than the existing reconstruction methods in the accuracy and complexity of multi dimensional space model, and the accuracy and complexity of reconstruction are better than the existing reconstruction methods in the same situation.

Aiming at the problem that pointer instrument detection algorithm has slow locating speed and low real time performance in edge equipment, this paper proposes a pointer instrument video detection method based on improved FCOS. The algorithm is based on FCOS model and uses lightweight network ShuffleNetV2 to extract image features. Using PAN structure to strengthen the original feature fusion network, a bidirectional feature fusion network is formed. The attention module with global context information is introduced to reduce the information attenuation in the process of feature fusion. The two parameters of pixel utilization PUR and relative time increase RIT are introduced to test the influence of images with different image pixels on the detection effect in a more intuitive form. Through experiments, when the resolution of the input image pixel is 1280×1280, compared with the baseline model, the detection time of the pointer instrument video detection method based on improved FCOS is reduced by 91.60% when the detection accuracy is similar.

An Unmanned Surface Vehicle (USV) is an autonomous marine unmanned vehicle. USV navigation control has been extensively studied as a key technology. However, for USVs in complex environments, there are many difficulties in modelling and controlling under the wind, wave, and current conditions. At present, path planning mainly focuses on the realization of obstacle avoidance function and the rapid generation of feasible paths. The main evaluation criteria depend on the calculation time and convergence of the algorithm, while the evaluation criteria of the optimal path itself and the selection of the optimal path are less involved. This paper does not use the existing path planning methods for USV. Instead, the environment is equivalently modeled in the Riemannian space, and the geodesic between two points on the surface is calculated. In this paper, two geodesic algorithms based on triangular mesh models are studied. By comparing the calculation accuracy and running time of the two, a more suitable path-planning method for the USV system is determined.

The stability of the bogie limits the maximum safe operating speed of the vehicle. The analytical methods are diverse and have their own characteristics. In this paper, the dynamics equation of the double-axis bogie model is established. The critical velocity is calculated by Matlab, and the reduced order is converted into an equation of state according to the characteristics of the mathematical model. The analog circuits of addition and subtraction, differentiation and integration, which are composed of resistive and capacitive elements and operational amplifiers, are mathematically equivalent. The equivalent circuit simulation and numerical simulation established by Multisim are compared and analyzed, and the simulation results are basically the same, which indicates that the equivalent circuit not only has the advantages of real-time parameter adjustment and fast calculation speed, but also has good equivalence and feasibility, which provides a reference to the research and analysis of mechanical system dynamics.

Insulator defects are unavoidable due to long-term exposure and harsh natural environment. With the development of the computer vision and deep learning, researchers have been drawn to automatic unmanned aerial vehicle-based insulator inspection to improve power transmission safety. However, achieving full automation of fine-grained insulator defect detection is still very challenging due to the visual complexity of defects and the high-resolution image computation complexity. This study focuses on fine-grained insulator defect detection by vision transformer based on deep learning. The proposed method is based on a Swin-transformer framework, which focuses on learning hierarchical image feature representation computed with shifted windows scheme. Experiments were carried out on high-resolution image datasets to evaluate the performance of the proposed method for fine-grained insulator defect detection tasks. The results show that the method takes advantage of vision transformer's capabilities and outperforms the state-of-the-art method in terms of mean average precision (mAP) at 94.2% when the intersection threshold over union is set to 0.5.

With the wider application of data mining technology, new media has been accepted by people for its high communication efficiency, good communication quality, and strong interaction. The city is an important factor of modernization, and the use of new media to spread the image of the city can promote the rapid development of the city. Based on this, the purpose of this article is to research and analyze the optimization path of new media city image communication strategy based on data mining technology. This article first summarizes the status quo of urban image communication, and then extends the theory of urban image communication and analyzes it in detail. Then through the analysis of new media communication methods and their advantages in urban communication. This article elaborates on the new media communication based on data mining technology, and uses questionnaire survey method, field survey method and other research methods to carry out experimental research on the theme of this article. Research shows that new media has become an important channel for city image communication.

With the continuous improvement of the electricity demand of the whole society in China and the complexity of power grid network planning and construction, how to further optimize the structure of low voltage distribution network and improve the investment efficiency and efficiency of distribution network planning has become one of the important topics facing the operation and development of power grid enterprises. This paper analyzes the current situation and problems of medium and low voltage distribution network construction in a region of GD province, combines with the relevant principles of improved genetic algorithm, optimizes the investment and construction of medium and low voltage distribution network construction in this region, and verifies the effectiveness of the model method.

The emergence of DeepFake poses serious risks to data privacy and social stability. We propose an end-to-end DeepFake video detection method based on a dense dynamic convolutional neural network (CNN) to address the poor performance of DeepFake video detection on complex compression formats and datasets of different forgery methods. In this method, extracted face images are clustered and cleaned by cosine similarity, and face images are expanded through data augmentation to improve data diversity. Dynamic dense blocks are incorporated in a CNN to address optimization difficulties in deep neural networks, and an attention mechanism further improves generalization power. Convolution kernel pruning increases processing speed by effectively reducing the computational needs due to dynamic convolution. Experiments demonstrate that this method has better results on DeepFake video detection across compression rates and datasets compared to other network models.

At present, the actual situation of spot transactions in China's power market is relatively complicated, and it is necessary to expand diversified power products to meet different market participants. Therefore, we can introduce the Nordic power market block trading products as a reference to continuously improve our own construction. Combination transactions based on block transactions and their derivatives are aimed at enabling power market participants to choose transaction types that are more in line with their own characteristics and making transactions more flexible and diverse. Based on the current situation of China's high proportion of renewable energy, combined with the Nordic day ahead power trading model, this paper first analyzes the diversified trading needs, and designs flexible and diverse power market trading varieties. Then, we constructs a power spot market clearing model which considers the flexible combination of multiple trading varieties. The calculation example shows that the flexible transaction mechanism has greatly stimulated market vitality, achieving a high transaction rate and low abandonment rate on both the supply side and the demand side. While meeting the power demand of users, it also guarantees the interests of power generation entities with different economic characteristics.

To test to evaluate the aviation brigade combat equipment support ability, based on the combat equipment safeguard drill results in recent years, with the relevant regulations standard, through consulting literature, investigation and other methods, constructed the aviation brigade combat equipment support ability evaluation index system, and through the AHP to determine evaluation index weight of every layer. The fuzzy comprehensive evaluation method is used to calculate the evaluation results, and finally a more scientific and comprehensive evaluation is made for the actual combat equipment support capability of the army aviation brigade.

This article focuses on the analysis of the performance of asphalt pavement, and establishes a prediction model for the performance of asphalt pavement in Guidu based on the data of traffic volume, climate, and road surface smoothness. After fully understanding the performance evaluation indicators of various asphalt pavements, the international roughness index is selected as the performance evaluation indicators of asphalt pavements. According to the technical performance of the asphalt mixture (high temperature stability, low temperature crack resistance, water stability, antifatigue performance, anti-aging performance, etc.), the influencing factors (temperature, rainfall days, traffic volume, etc.) of the asphalt pavement performance are derived. Collect traffic flow and asphalt pavement performance data on the spot, and process and analyze the data. Establish a gray forecast model, a moving average forecast model, as well as a multiple regression forecast model and a VAR model that consider the four variables of traffic volume, truck ratio, temperature, and rainfall days. Predict the performance of the asphalt pavement through the above model, and get the prediction result.

The effect of point cloud denoising is very important for the subsequent surface fitting and modeling design of the 3D scanning process. How to extract feature points quickly and accurately has become a research hotspot. However, the key to point cloud denoising lies in singular values and outliers. Therefore, this paper proposes a denoising model coupled with multi-feature parameters, discusses the influence degree of each feature point parameter separately, and uses the swarm intelligence algorithm to solve a set of optimal parameter weights to determine the point cloud denoising model, and to achieve the optimal denoising effect of 3D scattered point cloud. The simulation results show that the swarm intelligence algorithm used is faster and less time-consuming than the existing differential evolution algorithm. At the same time, the point cloud denoising model proposed in this paper has better performance than radius filtering and statistical filtering. denoising effect.

This paper starts from the purpose of assisting electricity market entities to make better decisions, providing guidance for the design and selection of retail packages, improving market efficiency, and guiding the healthy development of the market, and considering the future development trend of the electricity retail market, designs an optimal package decision model under the background of competitive electricity retail market. The model can be based on retail customers' electricity consumption information and risk preference analysis and select the optimal retail package that meets their needs from the retail package library. Finally, an example is used to illustrate the package decision-making process that considers the consistency of the customers' power consumption curve and the time-of-use electricity price and considers the market transaction needs of different customers in the electricity retail market.

In the real world, traffic scenes are complex and contain intricate road networks, which makes the shortest path computation on large-scale road networks a challenging task. Existing research has concentrated on small-scale urban road networks or grid maps. In practical scenarios, however, we are often faced with seeking the shortest paths on large-scale road networks. For this reason, it is imperative to develop efficient shortest path searching algorithms, as it offers significant savings in time and resources. To tackle this issue, this paper proposes two improved A* algorithms, namely the Weighted A* algorithm and the Bidirectional Weighted A* algorithm. To verify the effectiveness of our proposed algorithms, we validated the performance of our proposed algorithms against the conventional Dijkstra and A* algorithms on urban road networks of different sizes. Our results significantly demonstrate the effectiveness of our solution, as both algorithms significantly outperform Dijkstra's and A* algorithms, with little loss of accuracy.

Based on the ship data of a certain port in China within one year, this paper analyzes the distribution characteristics of ship arrival time interval of the port within one year by using mathematical statistics method, and the Matlab tool is used to fit the three probability distributions of arrival probability of each ship type to obtain the goodness of fit, sum variance and root mean square difference. Through the analysis of fitting data results of negative exponential distribution, Weibull distribution and Erlang distribution, it can be compared and concluded that: when ship types are not distinguished, the arrival laws of all ships obey the Weibull distribution; When the ship types are distinguishes, container ships, bulk carriers and general cargo ships obey the Weibull distribution, while oil tanker arrival law obeys both the Weibull distribution and the low-order Erlang distribution.

We propose a brand-new global point cloud descriptor called IM2DP that combines shape and intensity data to conduct loop closure detection. By incorporating intensity data, our method expands the multiview 2D projection (M2DP) descriptor. We calculate intensity characteristics from multiple 2D projections of the point cloud and apply them to the M2DP shape features. Then, singular value decomposition (SVD) is used to compute the compressed descriptors for reducing the dimensionality. We combine this algorithm with the LOAM and conduct experiments on the KITTI public dataset. The results show that our approach is effective in reducing trajectory error and shows competitiveness when compared to the M2DP.

The Hilbert-Schmidt Independence Criterion (HSIC) was originally design-ed to measure the statistical dependence of distribution-based Hilbert spaces embedding in statistical inference. In recent years, due to the validity and efficiency of this standard, it has been witnessed that this criterion can tackle a large number of learning problems owing to its effectiveness and high efficiency^[1]. Unlike traditional binary classifications or multi-class single label problems, one goal of multi-label problems may be related to multiple labels. The rich relationship between labels makes the analysis of multilabel problems more complex. To solve the problem of how to make use of the relationship between features and labels, a multi-label feature selection algorithm based on HSIC is presented. This method uses Lasso (the least absolute shrinkage and selection operator) to solve a non-convex HSIC problem, and converts it to solve a lasso optimization problem, which can effectively calculate the global optimal solution. Finally, experiments show that our algorithm can improve the performance of multi-label classification.

Virtual reality technology is a new multidisciplinary integrated technology, in which localization technology is the key technology to determine the user experience, and is the core of VR technology. RTK and visual slam are two common localization technologies, however they are limited by satellite conditions and rely heavily on feature point extraction and matching effects, which will affect the accuracy of localization process. So as to achieve high-precision virtual reality spatial localization, this paper proposed a novel virtual reality localization method called RS-fusion, which combined the RTK carrier phase difference technology and the visual SLAM technology by Kalman fusion. In addition, the gain matrix is used to calculate the spatial displacement state to meet the accuracy requirements of virtual reality and augmented reality localization in indoor and outdoor environment, and to obtain a more matching virtual reality fusion effect and simulation mapping. Experiment results show the robustness of the method and RS-fusion can realize the simultaneous localization of cameras in real space and virtual space.

Advances in network and storage technologies enable data centers to provide high-concurrency and low-latency processing responses. To meet user needs, a variety of low-latency technologies can be used to improve the overall concurrent response capability of the system on the premise of ensuring system stability. User mode thread pool is one of the techniques that can be used. In a multi-core processor system, the scheduling management of the user mode thread pool is much more complicated. At present, the scheduling management of the user mode thread pool mainly focuses on the resource utilization of the user mode thread pool itself and the scheduling management of many user-mode threads, such as dynamically adjusting the number and usage of user mode threads in the user mode thread pool according to the load of the system, and replacing the system thread scheduling with user mode thread scheduling, etc. The existing user mode thread scheduling methods do not distinguish the performance requirements of different data processing tasks, treat all user mode thread tasks in the same way, cannot reflect the performance requirements of different user mode thread tasks, and cannot provide more fine-grained user mode thread execution timing control; the existing user mode thread scheduling methods do not provide more effective scheduling control for the host thread, but instead use the thread scheduling mechanism of the operating system itself to control the execution of all kernel threads which is difficult to isolate the host thread and other kernel threads in the system; the existing user mode thread scheduling methods cannot give full play to the parallel execution advantages of the multi-core processor architecture, and it is difficult to implement unified resource allocation management among multiple user mode thread pools. This paper provides a two-level scheduling management method for user mode thread pools that effectively integrates the operating system scheduling mechanism and user mode thread scheduling mechanism, which can distinguish the needs of different data processing tasks and implement more targeted and fine-grained user mode threads scheduling and execution control, without affecting the scheduling mechanism of the operating system itself, realize the isolation of host threads and non-host threads, implement differentiated scheduling control. And it can realize unified resource allocation management of multi-user thread pools for multi-core processor architecture so that the overall performance of the system can be freely scaled with the change of data task processing requirements to provide a more efficient user mode thread pool management for high concurrency and low latency systems.

Prompt tuning is a parameter-efficient method that even surpasses traditional fine-tuning methods in few-shot scenarios. Nowadays, pre-trained language models are getting larger and larger, with more and more parameters, which makes the traditional fine-tuning method impractical to implement and consumes a lot of computing resources. Therefore, prompt-based methods have a broad application prospect. In the experiments, it is found that prefix tuning, a prompt-based method, has the problem of non-convergence or is quite slow to converge when the training samples are small. This paper proposes a cross-task parameter transfer method, which transfers the trained parameters from prompt tuning tasks to prefix tuning to improve the training speed and alleviate the problem of non-convergence or slow convergence in prefix tuning tasks.

Aiming at the problem that the feasibility and accuracy of image measurement and analysis are reduced due to video stabilization in automobile collision experiment, a video image stabilization decision algorithm based on block matching method based on region of interest (ROI) is proposed in this paper. Firstly, in the first frame, the user selects the ROI region containing the marker identification, uses the block matching method to automatically match the ROI region features frame by frame, and estimates the motion of the video frame. And then determines whether the feature matching fails according to the motion of the video frame. When the feature matching fails, the search range can be reduced and the ROI features can be re matched to improve the matching accuracy and speed, Finally, the video frame sequence after image stabilization is output by frame reverse motion compensation according to the amount of motion. The experimental results show that the proposed algorithm can effectively reduce the video stabilization in collision experiments in the subjective evaluation and the objective evaluation of local peak signal-to-noise ratio.

To improve the design and development efficiency of automobile project products and ensure product quality, an automobile virtual assembly platform was designed based on 3D Max software. The basic framework and development environment of the platform is built. The automobile virtual model is constructed by using the 3D Max technology. The sequence planning of automobile virtual assembly is realized by using an ant colony distributed algorithm. The functional modules of the automobile assembly platform are analyzed, and the software design of the platform is completed. In order to test the feasibility of the design platform, a simulation experiment is conducted. The results verify that the average running time of the platform’s main nodes is 0. 49 s. The function implementation cycle is short, and the platform runs smoothly, with real-time and high efficiency. For the key points of the automobile head assembly, the average deviation of the designed platform assembly’s key points is 0. 92 mm, which is within the control range of ±1.0 ~ ±3.0 mm, indicating that the assembly quality and modeling accuracy of the design platform is qualified. For the post-processing of the car head assembly, the rendering completion of the assembly is 100%. The material effect of the car head assembly is realistic. The light and shadow effect is excellent, and the realistic assembly environment and assembly effect are highly restored.

By analyzing the texture characteristics of rapeseed leaves, this paper proposes a multifractal detrended moving average analysis (MF-DMA) model to identify different nitrogen application levels. The model adopts several global generalized Hurst exponents and some other related multifractal feature parameters of rape leaf images under different λ values, and uses different feature parameter combinations to identify nitrogen nutrition in basal leaves, middle leaves and top leaves, respectively. The results show that the recognition effect is the best when λ=0, and when λ=0, the basal leaves and middle leaves are more sensitive to nitrogen than the top leaves. Through qualitative identification of three parts of rapeseed samples with moderate nitrogen application and deficient nitrogen application, the results showed that the two classification methods of support vector machine and random forest were the best, and the best identification accuracy was 96.01% and 96.83%, respectively. It shows that the model proposed in this paper has good effectiveness.

Objective: to analyze the characteristics of pulse pattern parameters of systemic lupus erythematosus (SLE) patients and the factors influencing their TCM patterns, and to provide an objective basis for their TCM pulse diagnosis and diagnosis. Methods: the SmartTCM-A1 TCM intelligent detection system was used to collect pulse diagnosis image information from 267 SLE patients (140 in the Yin deficiency internal heat evidence group, 96 in the heat toxin incandescence evidence group, and 31 in the rheumatic heat paralysis evidence group) and 130 healthy individuals, extract the image parameters of pulse diagnosis, and analyze the characteristics of pulse diagnosis parameters of SLE patients and their influencing factors of pulse diagnosis parameters of different TCM evidence types by logistic regression model. Results: ① compared with the healthy group, the pulse parameters h1 value, h4 value, and t1/t value were significantly higher in the SLE group (P<0.05) and the pulse parameter t value was significantly lower in the SLE group (P<0.05); logistic regression analysis showed that the independent influencing factors in SLE patients included: h4 value (OR=1.073; 95% CI=1.003-1.148 ;P<0.05), t-value (OR=0.003;95% CI=0.000-0.763;P<0.05) and t1/t-value (OR=0.000;95% CI=0.000- 1.199;P<0.05). ② Compared with the group with incandescent heat toxin evidence, the h4 value, h5 value, t4 value, t5 value, t5/t4 value were significantly lower in the group with Yin deficiency internal heat evidence (P<0.05), the h4/h1 value and t1/t4 value were significantly higher in the group with Yin deficiency internal heat evidence (P<0.05), the h4 value, h5 value, t5 value, h4/h1 value were significantly lower in the group with rheumatic heat paralysis evidence (P<0.05), and the t4 value, h5/h1 value, t5/t1 value were significantly lower in the group with rheumatic heat paralysis evidence (P<0.05). , h5/h1 values, and t1/t4 values were significantly higher in the rheumatism-heat paralysis group (P<0.05); logistic regression analysis. The results showed that the independent influencing factors of TCM evidence in SLE patients included: h3 values in the yin deficiency internal heat evidence group (OR=2.295; 95% CI=1.843-2.858; P<0.05) and h3 values in the rheumatic heat paralysis evidence group (OR=2.309; 95% CI=1.87-2.85; P<0.05). Conclusion: conclusion pulse diagnosis parameters h4 value, t value and t1/t value are one of the influencing factors for the diagnosis of SLE patients, and pulse diagnosis parameter h3 value is one of the influencing factors for the diagnosis of SLE Chinese medical evidence.

In the information age, virtual reference service (VRS) has gradually replaced the traditional face-to-face (F2F) reference and become an important way of reference for library users. Based on grounded theory, this research uses NVivo12 software to mine the virtual reference service data of users in the Sichuan University library, the first is high-frequency word analysis, and the second is mining the data by combining open coding, spindle coding, and selective coding. The analysis results of high-frequency words show that in the process of using various services of the library, the contents of the virtual reference service mainly focus on the following aspects: hours, literature resources, space, and facilities. The coding results show that the library virtual reference service data could be encoded into three core categories, they are "library spaces and facilities", "collections and electronic resources" and "services", where "library spaces and facilities" is the most concern aspect of the library users, which has 359 nodes, accounting for 50.71% of the total, and the proportion of "collections and electronic resources" and "Services" is not much different, 27.26% and 22.03%, respectively.

Some specific, repetitive actions made by the staffs in the refueling work scenario at the airport can be considered as a way of information transmission, so it is necessary to carry out on-site automatic identification and monitoring of these specific actions to improve the level of supervision. This paper proposes a lightweight refueling behavior recognition algorithm applicable to the field based on video sequences. The algorithm firstly uses the YOLOv3 improved target detection network for human body detection. The resulting human body detection box is tracked using the target tracking algorithm, and the tracked human body sequence maps are input into the behavior classification algorithm based on time-space feature fusion to realize the fast and intelligent analysis of the behavior. The test results of deploying the algorithm to Hi3559A embedded equipment show that the recognition accuracy of the algorithm reached 94.68%, and the inference speed reached 22FPS, which can meet the needs of real-time behavior analysis and processing at the airport refueling site.

Relation extraction refers to get the triple structure composed of semantic relation entity pairs from unstructured text, which is an important part of tasks such as knowledge graphs. At present, the joint extraction model is in common used to avoid the impact of overlapping entities, but there are the following problems. First, the dependencies between text words are not fully considered, and the recognition performance of entities with long spans is low. Insufficient utilization of information makes it difficult to fully extract implicit relationships. In order to address these issues, this text proposes an improved joint learning model, which builds text semantic representation through BERT pre-training, obtains relation type representation as an additional mapping through a multi-label classification method, and sequentially uses multi-layer BiLSTM combined with highway network to obtain semantic information, and combine The attention mechanism obtains the entity location score, and the pointer network is used to obtain the entity location. The experiments of this method on the common dataset of relation extraction task is effective.

As a miracle of bridge engineering, it is meaningful to analyze the stability of the Golden Gate Bridge. This paper focuses on the finite element method to analyze the Golden Gate Bridge’s performance on loading. In this work, ANSYS is used to build a simplified model of the Golden Gate Bridge without cables. There are six modes of the deformation of the bridge, and there are six different frequencies. The natural frequency of the bridge is 0.055Hz. Because this simplified model does not have cables to reduce the deformation, the average frequency is 0.07737 Hz which is 0.02237 Hz higher than the natural frequency. Moreover, after analyzing the harmonic response, the largest equivalent stress is 1.5487e⁷ Pa at joints of the bridge when the frequency is 0.0268 Hz and the sweep phase is -56.649 degrees. As a result, the analysis of the bridge based on the models in Ansys clearly shows the stability of the bridge.

Calligraphy is the art of writing and each writer has his own features, where calligraphy strokes carry the most important features. In order to identify the stroke and the writer, this paper proposes an approach of stroke extraction and presentation. Firstly, skeleton segments of the character are extracted according to the writing rules. Secondly, neighbouring common segment strokes are connected to build a complete stroke, followed by special short stroke backtrack. Thirdly, individual outline strokes are isolated by using curves to build the closed contour for the stroke. Finally, generate and save outline strokes for future stroke feature extraction. Our approach can extract strokes from parts of regular script and official script.

Conventional recommendation systems mostly recommend based on users’ global stable interest, while SR focuses on learning local active interest from the current session. Relevant research points out that integrating users’ global and local interest can improve model accuracy. However, most SR models are deficient in interest mining and fusion, and do not fully utilize the time interval content between sessions, result in inaccurate recommendations. In the paper, we propose a new session-based recommendation method called Global and Local Interest Integration Model (GLI-GNN). GLI-GNN introduces the concept of session time and attention mechanism to obtain global preferences from users’ previous sessions; it uses GNN to obtain local interest from current sessions; it uses attention mechanism to integrate both. The results indicate that GLI-GNN realizes new optimal performance in next-shot interactive recommendation. Experimental results show that GLI-GNN is superior to the existing SR models.

Analyzing the characteristics of climate change and clarifying the critical climate factors affecting sesame yield are significant for sesame production to cope with future climate change and ensure stable yield and income increase. In this study, trend analysis and mutation analysis were used to study the changing trends of climatic factors, yield per unit area, and climatic yield. The data included meteorological and sesame yield data in Zhumadian City of Henan Province from 1989 to 2018. The stepwise regression model was used to extract critical climate factors, and multiple regression models of sesame climate yield and critical climate factors were established. The contribution rates to sesame yield of the critical climate factors affecting sesame yield were determined. The results showed as follows: from 1989 to 2018, the temperature during the sesame growing period in Henan Province showed an increasing trend, and the minimum temperature showed the most apparent increasing trend, with an average annual increase of 0.03℃. The decreasing trend of precipitation and sunshine duration in the sesame growing period was insignificant, but the inter-annual fluctuation was significant. The coefficient of variation was 40.8% and 15.41%, respectively. The yield per unit area of sesame showed a significant increasing trend, and the annual growth rate was 34.08kg·hm^-2. The rising trend of climate yield is not adequate. The critical climate factors affecting sesame climatic yield were sunshine duration and average temperature. The contribution rate of each climate factor to climate yield in each growth period was ranked from the largest to the smallest as the average temperature at the seedling stage, the sunshine duration at the maturity stage, and the sunshine duration at the emergence stage. We have developed a future climate change adaptation strategy for the region based on the findings. The following measures should be taken to produce high-yield and high-quality sesame in south Henan Province. Local production departments should conduct training on high-yield cultivation techniques such as early sowing of sesame and risk management of waterlogging. Research institutions accelerated the selection, demonstration, and promotion of improved sesame seeds. The government has issued preferential agricultural policies.

Temporal Anti-Aliasing (TAA) is a popular method for eliminating temporal aliasing problems. However, the images simply processed by TAA become blurred and lose some details. In this paper, an improved TAA algorithm named Edge Temporal Anti-Aliasing (ETAA), is proposed. A time iterative edge detection method is designed to enhance the detection accuracy of pixels with temporal aliasing and spatial aliasing. These pixels are blurred and blended, and other pixels are replaced with the pixels of the current frame. Furthermore, an approximate minimum filter is used to eliminate the flickering phenomenon of high-energy noise. Compared to TAA, ETAA outperforms TAA in terms of image detail preservation when the rendering camera moves. Meanwhile, the time cost of proposed method also can satisfy the requirement of real-time rendering. Experiment results show that ETAA can effectively and quickly eliminate the time aliasing when the camera moves, and achieves a better trade-off in flicker, ghosting and blurring.

With the rapid development of the modern manufacturing industry, the requirements for milling machine products are getting higher and higher, and then the working accuracy of milling machines has higher requirements. The cross beam of the milling machine is a key component of the milling machine. The cross beam is connected to the top of the machine tool bed by bolts, which mainly play the role of connecting and supporting the saddle and the spindle sleeve. The beam of the milling machine bears most of the working load of the milling machine, so the rigidity and strength of the beam of the milling machine directly affect the working accuracy and service life of the milling machine. In this paper, three dimensional software is used to establish the geometric model of the milling machine beam and establish a finite element model. The stiffness and strength of the saddle are analyzed by the finite element method according to the load condition of the saddle in the middle of the beam. Through analysis, it can be seen that the sliding saddle of the milling machine is in the worst working condition in the middle of the beam. According to the strength theory, it can be checked that its strength meets the requirements. This static analysis can provide reliable data for the design optimization of the milling machine beam.

Lost circulation during drilling wells is very detrimental since it greatly increases the non-productive time and operational cost, also seriously lead to wellbore instability, pipe sticking, blow out, etc.. However, in the process of drilling wells, geological characteristics and operational drilling parameters all may have impacts to the lost circulation. This makes the establishment of the relations between the lost circulation and drilling factors very challenging. In this paper, we tested five different kernel function (linear, quadratic, cubic, medium Gaussian and fine Gaussian) derived support vector regression (SVR) models and four-layer artificial neural network (ANN). By combining their accuracy and time efficiency, the ANN is regarded as the optimal predictor of lost circulation. By training ANN using different combination of drilling features, we concluded that depth, torque, hanging weight, displacement, entrance density and export density are the key factors to accurate predict the lost circulation. The corresponding trained ANN network can achieve 99.2% accuracy and evaluate whether a drilling feature vector corresponds to lost circulation or not in milliseconds.

Through the analysis of power big data and according to the regional power grid facilities development level, load margin, power supply quality and other aspects of research, evaluate the level of power supply capacity level of each region, for the quantitative power supply capacity level of power supply companies. And through the power data comprehensive analysis of each industry of each region, to determine the most suitable industry recommendation.

Identifying power transformer faults accurately is critical to maintaining the stable operation of power system. Intelligent fault diagnosis algorithms based on dissolved gases have been extensively researched and implemented. However, in practice, collecting labeled data is time-consuming and costly. Therefore, it is necessary to establish a valid diagnostic model with limited labeled data. To solve this problem, a novel semi-supervised learning method for power transformer fault diagnosis is proposed in this paper. First, all the dissolved gas samples are constructed as a weighted K-nearest neighbor (KNN) graph to initially describe association among all samples. Then, a semi-supervised random multireceptive field propagation graph convolutional network (SSRMFPGCN) is designed for fault feature extraction and classification. Finally, the collected power transformer fault data are used to validate the proposed method. The experimental results show that the method proposed in this paper can still achieve 94.06% accuracy with only 20% of labeled training samples, which is significantly superior to the traditional intelligent diagnosis methods.

Machine translation is an important research field in natural language processing and artificial intelligence, which studies how to use computers to automatically convert languages. We experimented with attentional neural network machine translation for Chinese Kazakh language pairs and used data enhancement methods to alleviate the serious shortage of parallel corpus. The experimental results show that the proposed method is simple and effective, realize the mutual translation between Chinese and Kazakh with extremely low resources. At the same time, our proposed the method of two-way iterative back translation improves +4.47 BLEUs on Chinese to Kazakh and +5.97 BLEU on Kazakh to Chinese respectively.

In marine ranching, as an important part of marine ranching, the force of netting is extremely complex. The calculation of hydrodynamic load of netting is particularly important for the design and construction of marine ranching. In this paper, the theory of calculating the force of netting at home and abroad is sorted out, the MATLAB software is used to program, and the Screen calculation model is used to study the applicability of the Aarsnes formula used in the current Chinese ' Guidelines for the Inspection of Marine Fishery Facilities'. The applicability of the Aarsnes formula formula between PE material netting, metal netting and nylon netting and the influence of the presence or absence of nodules on the numerical calculation results of the netting are judged respectively, and the applicable conditions of the Aarsnes formula are summarized.

When the monocular visual measurement method is adopted, the initial alignment precision of the moving base tends to be easily interrupted by the surrounding environment. In order to solve this problem, in this study, the authors suggest an initial attitude reference transfer method for moving base based on correlated frames of images. This method utilizes the advantage of inertial reference, namely, high precision of relative attitude measurement within a short time period, transforms the multi-frame sequence images of the planar target to the coordinate system of the initial moment, and averagely reduces noise interruption through more than one image, so that the attitude-calculating precision could be promoted. Simulation research shows that: the attitude reference method based on correlated frames, which is put forward in this study, could effectively reduce random noise interruptions. Within the rotation scope of -1° to 1°, the precision of attitude alignment angles in three directions is better than that of 45''.

At 2:37: 34 seconds on April 14, 2017, a large area of Qinglan Zone suffered a power outage. During the troubleshooting of the power supply bureau, it was found that there was a suspicious self-unloading vessel "HXX 338" anchored in the anchor forbidden area of submarine cable and weighed anchor. Hainan Power Grid sued the suspicious ship and won the lawsuit, but the shipowner insisted on the ship not damage the submarine cable. Power grid entrusts the author as maritime expert, make a professional judgment with respect to the suspicious of this accident. The author makes use of navigation common sense: after anchoring, the position of the ship must be within the radius of the arc with the anchor position as the center of the circle, the length of the chain thrown out plus the sum of the distance between bridge and anchor chain hole; put in another way, if the ship's position is known after anchoring, the anchor position must be in a circle with the ship's position as the center of the circle, the length of the chain thrown out plus the distance between bridge and the bow's chain hole and as the radius. The trajectory of the ship was analyzed, and the suspicious ship was identified as the ship resulting in the accident. Investigate and analyze the facts of the ship in question moved in the prohibited anchor zone and hooked up submarine cables, find out the causes and distinguish the responsibilities. This article has explored an effective analysis for maritime investigation.

In order to accurately simulate the dynamic response process of ceramic composite target under the penetration of 7.62 mm steel core projectile in ballistic test., this paper selects FEM (Finite Element Method), SPH (Smoothed Particle Hydrodynamics), SPH-FEM fixed coupling algorithm, SPH-FEM adaptive coupling algorithm as the simulation scheme, and uses LS-DYNA finite element software to analyze the penetration of small-caliber bullets. The research shows that: by analyzing the dynamic response process of ceramic failure and the residual velocity of the projectile, the FEM-SPH adaptive coupling algorithm can simulate the failure modes of ceramics at each stage, and quantify the simulation accuracy with the residual velocity of the projectile. The error of the adaptive coupling algorithm is 1.17%. The FEM-SPH adaptive coupling simulation model is reasonably applied to simulate the impact process of the bullet penetrating the ceramic target.

Machine vision technology has shown great potential for development and application in the coal heat utilization and coal chemical production process It is important to carry out image analysis-based stability adaptation analysis of coal heat utilization and coal chemical equipment. To further expand its application, this work reviewed the characteristics of different vision sensor systems and image processing algorithms in the process of image acquisition and processing, conducted applicability analysis for rapid prediction of coal quality, presented the current problems and future development trends of image analysis technology in the coal heat utilization and coal chemical production process. It provided theoretical guidance and reference for promoting automation, intelligence, and mechanization in the coal industry.

In the civil aircraft certification process, theoretical or empirical data provided by equipment suppliers are used to demonstrate compliance with the safety requirements related to CCAR 25.1309. During the aircraft operation phase, operational safety monitoring is required to continuously evaluate the operational data to confirm or correct the theoretical data given in the pre-certification system safety assessment process and to indicate the actual safety level of the aircraft. Through the operational safety data analysis method, the actual level of the aircraft is compared and analyzed with the deviation of the pre-certification safety target value, and corrective measures are taken to improve the aircraft operational safety level if necessary.

With the continuous maturity of big data, artificial intelligence, Internet of Things and other technologies, the rapid development of smart grid has been helped, but at the same time, the increasing line loss power has also attracted widespread attention. In the process of building a smart grid, each link of the grid operation generates a large number of multi-source heterogeneous data, including line loss data and line loss cause related data, which constitutes the big line loss data. First of all, considering the mining efficiency in big data, FP growth algorithm in association rule learning is selected to search the frequent item set of line loss features. Support, confidence and lift are used as evaluation indicators to analyze the association relationship between the causes of line loss; Secondly, a line loss prediction model based on deep learning is established. By eliminating the influence of line loss characteristics in turn, the correlation contribution of line loss causes to line loss is calculated to quantify the line loss caused by line loss causes. After verification, the depth confidence network and BP depth neural network as the prediction model of the depth learning method are superior to the shallow artificial neural network model in the prediction effect, and the prediction accuracy means the reliability of the contribution calculation. Finally, combined with the above two aspects of analysis, the causes of line loss in the substation area are comprehensively evaluated, and guidance suggestions are given to assist power enterprises in decision-making.

In order to determine the factors affecting emergency vehicle traffic and find out the mitigating measures, the accident tree method was used to qualitatively analyze the four main factors and 20 sub-factors affecting emergency vehicle traffic. Combined with analytic hierarchy process for quantitative analysis, using MATLAB to determine the weight of factors, the aim is to reduce the impact of subjective factors. The results show that among the four main factors, the weight value of traffic factor and road factor is as high as 0.834, which is the primary focus to determine the solution measures. Among the 19 sub-factors, the comprehensive weight sum of traffic accidents caused by intersection signal priority equipment, social vehicles not allowed, few lanes, complex terrain and emergency vehicles passing through the intersection is 0.543, which is more than half of the total weight. Therefore, it is necessary to solve the problem targeted.

Aiming at the prediction of wheel-rail relationship of high-speed trains, a prediction method combining maximum information coefficient (MIC) and LSTM was proposed. Firstly, the dynamic model of high-speed train was established by SIMPACK and the original dataset was obtained. Secondly, the maximum information coefficient is used to preprocess the data. Then, a prediction model of wheel-rail relationship based on long short-term memory neural network was built. Adam optimizer was used to optimize the learning rate and network structure. Finally, the optimized long short-term memory neural network is used to predict the wheel-rail relationship. The prediction results of wheel-rail relationship show that when the number of model iterations is 45, and the number of hidden layers is 110, the average absolute error percentage of model prediction is the smallest, and the value is 0.0247. Under these conditions, the predicted result is very close to the real value, that is, the data pretreated by the maximum information coefficient can make the model accurately predict the change trend of wheel-rail relationship, which can provide support for further research.

In this paper, we propose the concept of local-oriented freight volume of inland water transport, which can reflect the inland waterway transportation level in a province or a city more truly and objectively. Based on the characteristics of China’s inland water transport, the local-oriented freight transportation can be broken down into two parts: short-distance transportation and long-distance transportation. The freight volume of short-distance transportation can be captured in existing statistical reports, the freight volume of long-distance transportation can be calculated through the basic data of report system for ships entering and leaving ports. We take Sichuan province as an example, and investigate the correlation between local-oriented freight volume and GDP. The results show that the correlation between local-oriented freight volume and GDP is higher than that between published freight volume and GDP, this indicates that the localoriented freight volume and its statistical method proposed in this paper is feasible and applicable.

In order to better meet the development needs of interconnection and further improve the core operation capability of hospitals, the research of operation and maintenance interactive system based on artificial intelligence and big data information system is proposed. In the process of building the hospital information operation and maintenance platform, there are still some problems such as insufficient support in technology and operation and maintenance resources. It is necessary to effectively optimize the operation and maintenance system with limited human resources to ensure the stable operation and development of the hospital information system. Under this background, this paper analyzes the problems existing in hospital information management and the current situation of operation and maintenance of information technology, and discusses the mode construction of operation and maintenance platform based on big data artificial intelligence architecture. In the construction of hospital information operation and maintenance platform, we should focus on improving the service level and meeting the needs of patients, so as to continuously enhance the security, integration and interactivity of the system platform, thus promoting the sustainable development of China's medical reform.

This paper proposes a continuous analysis method of transmission line channel based on historical early warning data. It calculates the feature data, the real-time alarm data and the corresponding feature data, and passes the threshold value R₀. Determine whether it is a continuous alarm. In this paper, through the extraction of sample data extraction and calculation of Pearson moment correlation coefficient, solved the fireworks, foreign body alarm for low frequency, seasonal periodic, conventional single comparison and mechanical have good dimension reduction analysis cannot effectively identify the problem, for subsequent application scenarios such as alarm level intelligent annotation, AI image recognition model suspected false alarm and omission of sample identification model support, and improve the intelligent level of transmission line shipment inspection.

The current traditional human pose detection methods mainly acquire the pose data collected by sensors and realize the pose discrimination by edge computing decision unit, which leads to poor detection effect due to the lack of fusion of sensor data. In this regard, a high difficulty yoga body pose detection method based on data fusion is proposed. Complementary filtering, Kalman filtering and adaptive Kalman filtering methods are used to fuse sensor data for pose prediction, and a skeleton extraction model is constructed to build a multilayer perceptron network architecture to achieve the detection of difficult yoga asana poses. In the experiments, the proposed human posture detection method is validated. The analysis of the experimental results shows that the proposed method has a high comprehensive evaluation index and excellent detection performance when used for the detection of difficult yoga asana postures.

In this paper, an improved clustering algorithm is proposed and a heterogeneous model based on this model is developed. A new data extraction technology, such as data classification, network platform anomaly detection, distributed maximum frequent sequence extraction, comparison and mining of maximum frequent sequence data, is adopted. Through the comparison experiment, it is found that the algorithm can better reflect the correlation with the corresponding abnormal data types, and can better reflect the actual use of the algorithm.

As an infectious disease, pneumonia can cause great harm to human health. If pneumonia can be detected and treated early, its harm will be greatly reduced. Previously, hospitals relied on specialized doctors to diagnose diseases, but with advances in computer technology, deep learning is widely used in the medical field. In recent years, many excellent pneumonia classification methods have been proposed. They can judge whether a patient is infected with pneumonia based on their chest x-ray image, which effectively solves the shortage of professional doctors. In this paper, a convolutional neural network was proposed for pneumonia classification, and the pneumonia classification model was trained based on 1211 real chest x-ray image provided by Third Military Medical University. Experimental results on the test set show that the convolutional neural network proposed in this paper is not dominant, and its classification accuracy is only 72.0%, which is lower than the other three pneumonia classification models compared. Therefore, this paper integrates data enhancement, data preprocessing, transfer learning technology, then proposes a pneumonia classification accuracy enhancement scheme. This scheme improves the classification accuracy of the proposed pneumonia classification model from 72.0% to 82.0%, and the classification effect exceeds the other three pneumonia classification models compared.

With the popularization of information technology, the proportion of universities in the field of scientific research investment has increased year by year, which is also mainly used for laboratory investment. With the continuous maturity of cloud service technology, intelligent laboratory management has become the main mode of university management, which will gradually replace the traditional manual management. Through various sensing devices, the intelligent laboratory can realize the intelligent management of equipment, personnel, safety and technical data, which is the main embodiment of intelligence. Research method: through the intelligent laboratory management and control platform, managers can directly intelligentized the management system and control the laboratory, which is the necessity of the laboratory intelligent security management mode. Through the intelligent system, managers can realize the standardization, process and informatization of technical data, which will improve the utilization and sharing rate of experimental equipment. This paper mainly refers to the literature survey and empirical research. Through the investigation of literature to obtain information, this paper comprehensively and correctly understand a method to master the problem to be studied. The literature research method is widely used in various disciplines. Through empirical research, this paper proposes a design mode using scientific instruments and equipment, which can be used to observe, record and measure the intelligent management mode of laboratories step by step. Conclusion: the development of laboratory intelligent security management mode based on big data and cloud services has become the development trend of university laboratory management, which is of great significance to improve the scientific research level of universities. Based on big data analysis algorithm, this paper constructs an experimental intelligent security management mode.

The distributed hydrological model's high-resolution representation of watershed heterogeneity is especially suitable for hydrological simulation in watersheds with small areas, and the hydrological model needs to use flow observation data for parameter calibration to obtain adapted parameters, but the lack of observation basins is the core problem we face in China, where there are few hydrological stations and they are concentrated in important rivers. This work attempts to optimize the parameters of the distributed hydrological model using limited flow observation data in a small watershed where flow observation is relatively scarce and to explore the value of combining multiple sources of data to optimize the parameters of the distributed hydrological model. Based on the combination of meteorological observations and remote sensing data collected, a distributed hydrological model is constructed based on the WetSpa model for the Jianshan River basin, and a multi-objective genetic algorithm NSGA-II is applied to rate the model to predict the simulated basin runoff process. The results show that the hydrological model optimized by multi-objective parameters has good adaptability in the study area, and the simulation has certain accuracy, which can provide basic support for the simulation of the water environment in the basin and also provide a reference for hydrological simulation.

Parallel allocation is a decentralized mechanism for allocating indivisible objects to agents, in which agents are allowed to report their favorite objects among the remaining goods parallelly. How to maximize the interests of self-interested agents while taking into account fairness has always been a hot research topic. In this paper, we study the loser-reporting policy which ensures that the difference between the number of items taken by the agents does not exceed one. In order to make the expected benefit obtained by the agent equal to the expected benefit obtained by the agent in the sub-game perfect Nash equilibrium (SPNE) in parallel allocation, we come up with a concept of shelving disputes, and the SPNE strategy can be computed in polynomial time with respect to the number of objects.

With the upgrade of transmission line maintenance technology, the visual remote inspection of transmission line channel is widely used. However, the application of these marked data is currently in the stage of statistical analysis and report form, a deeper data mining work has not been carried out, such as the identification of areas with high incidence of alarm. This paper presents a method of analysis of high risk area of transmission line channel based on historical early warning data, which can be applied to the field of transmission line maintenance. By preprocessing the transmission line visual alarm data, using the improved k-means algorithm, analyze the clustering results, and finding out the series of modeling and data analysis. In addition, when determining the initial points, we propose the maximum and minimum longitude and latitude coordinates of the alarm dataset, divide a certain number of longitude and latitude grid, obtain the candidate data within each grid, and the method of obtaining the initial point set after screening. Based on the initial point set, the alarm area distribution is reasonable, and the regional stability does not drift. This paper can identify the visual alarm data of transmission lines with high alarm incidence areas, and provide effective data support for maintenance personnel to guide the deployment of human resources and ensure the safe operation of transmission lines.

The study investigated the mental health status, learning and living behaviors of college students, and found that there were significant differences in their mental health levels among college students with different learning and living behaviors. Taking sleep, late return, communication with classmates, study, absenteeism and sudden illness as information collection indicators, a psychological crisis monitoring and early warning system based on big data technology is established to conduct two-level monitoring and early warning, realize dynamic tracking management and accurate monitoring and early warning of college students' mental health status, and improve the level of mental health education in colleges and universities.

With the rapid development of sensors and computer vision, simultaneous localization and mapping plays an important role in the field of intelligent robots. However, traditional odometer from ORB-SLAM2 methods based on a RGB-D camera still face the problem of low accuracy and tracking failure in complex environments with obvious irregular motion, which lead to lower stability and reliability. Aiming at solving the above problems, an odometer method is proposed by introducing three-view information to improve the localization precision of current frame. The method establishes a framework for RGB-D odometer. The co-view relationships were developed between three frames based on feature matching method. Thus, the BA optimization model is constructed based on the tracked relationship among three frames. In this way, the problems of accumulated errors can be decreased by constraints of three-view information. Experiments on the public TUM data set show that the trajectories from the proposed method is more accurate and robust than those of tradition ORB-SLAM2.

In recent years, with the continuous development of China's electric power spot market construction, the market power problem brought by the structure of China's power generation-side has gradually become severe, and has evolved as the focus issue in the process of power market reform. The injection of renewable sources and the participation of distributed generation resources have brought profound influence on the power system and the power market, and it is urgent to establish and complete the market power monitoring and mitigation mechanism of China's electricity spot market. This paper primarily discusses the market power monitoring methods and mitigation mechanisms, then introduces the mature experience of typical international power markets in market power monitoring and mitigation mechanisms. Secondly, this paper summarizes the existing power monitoring and mitigation mechanisms in China, and analyzes the development and challenges of market power monitoring and mitigation mechanism construction under the new power system. We aim to provide some reference for the evaluation, monitoring and mitigation of market power in a series of practical aspects.

The Z-FFR (Z-Pinch Driven Fusion Fission Hybrid Reactor) is an important innovative design concept. The high uncertainty of the operating process of the pulsed power unit and the physical process of fusion and the absence of some theoretical and experimental conditions make it difficult to establish a high-precision mechanistic model, and it is difficult to obtain an accurate mathematical model of a complex, dynamic system. A data-driven physical modelling approach is urgently needed to replace the mechanistic models obtained with the aid of extensive simulations and experiments. The approach includes the creation of functional modules, the packaging of sub-modules, the configuration of module interfaces and the configuration of analytical models. Based on the actual needs of Z-FFR design and operation monitoring, the online analysis can be autonomously configured to accommodate different experimental data through machine learning, enabling anomaly detection, trend prediction, model design evaluation and operation assessment during the experimental process.

Wireless Body Area Network (WBAN) is radio frequency-based wireless network technology. It is currently widely used in sports, medical health and other fields. WBAN often transmits the user's body monitoring data, such as heart rate, pulse, blood pressure and other personal privacy data. These sensitive data are bound to face a serious risk of data leakage or malicious tampering. In order to solve the privacy security of user data in WBAN environment, this paper proposes an improved privacy protection strategy. This strategy selects a portion of the region so that the phantom nodes are evenly distributed around the source nodes, and the adjacent packet space has a certain angle, in addition to increasing the diversity of the path from the source node to the base station by selecting the peer node. The simulation results show that the strategy can provide better privacy data protection without adding too many child nodes, and extend the security time.

Cross-sea bridges have the characteristics of long strips, and the geological conditions along the route are complex. The geological data information collected by the survey is related to the quality of the entire bridge project. The geological information in the current survey stage is mostly recorded in paper logs. The design unit will design the structure after obtaining the paper geological information, and make design adjustments based on the actual geological information during the actual construction process. This puts forward higher requirements for the convenient, efficient, accurate collection and sharing of geological information, and the software based on mobile terminals can just meet these requirements. In view of the current problems in the collection and sharing of bridge engineering geological information, this paper designs an information solution for the collection and sharing of cross-sea bridge geological data, designs the system architecture and application functions in detail, and develops a cross-sea bridge geological information collection and sharing solution. The Android and IOS terminal apps build a bridge engineering geological system with B/S architecture through the front-end and back-end separation mode, and establish a standard geological database. Through the application of system functions in practical engineering projects, the digital collection and transmission management of geological data is realized, which greatly improves the transmission and sharing efficiency of geological data in the process of survey, design and construction, and provides a reference for the realization of digital management of bridge geological information.

The improvement and upgrade of the electric power enterprise integrated management system are faster and faster. In the selection of the system architecture, the selected architecture is often analyzed according to the actual needs, which is to establish the advantages of the system architecture and reduce the cost of subsequent development and upgrade. With the increasing workload of operation and maintenance of more and more terminal equipment and network management day by day, the defects of an insufficient number of operation and maintenance personnel and backward management means will become more prominent. This paper applies the B/S interactive structure and the MVC three-tier architecture. The development platform is based on J2EE. SQL Server is applied as the database server, and UML modeling tools are adopted in the modeling process. The system before design was investigated by a detailed survey of the user’s needs. The implementation of the system effectively improves the efficiency of its infrastructure project management, making the company's infrastructure management more scientific and reducing the business operating costs.

The analysis and study of students' classroom behavior can help develop students' abilities and improve teachers' teaching, and has been one of the key issues closely followed by the education community. In recent years, graph convolutional neural networks (GCN) have been widely used in various fields with outstanding success. Therefore, this paper proposes a GCN-based approach for modeling and analyzing student classroom behavior data. The experimental results show that the method helps to develop students' ability and improve teachers' teaching level.

In the internet environment, an electronic contract is a necessary transaction condition for business cooperation between enterprises. To avoid the risk factors in enterprise electronic contracts and promote the smooth progress of business cooperation, this paper researches the prediction method of enterprise electronic contract performance risk based on the decision tree model. On the basis of the decision tree model, the key nodes are pruned, and then the validity parameters are combined to complete the collection and analysis of enterprise electronic contract risk data. According to the distribution characteristics of risk factors, the performance reliability conditions are improved. The prediction of electronic contract performance risk is realized by solving the value range of prediction parameters. The experimental results indicate that under the decision tree model, the maximum accuracy of risk vector extraction in the process of electronic contract performance reaches 97.5%, which has outstanding application value in avoiding risk factors and promoting business cooperation between enterprises.

Rice yield prediction is a vital problem in the national agriculture and economy. The development of deep learning overcomes the obstacles of traditional machine learning and shows superior performance in solving complicated problems. Especially for natural language processing (NLP) models such as LSTM and GRU, these models outperform the time series data, thus having great potential for complex agricultural spatiotemporal data with high dimensionality and nonlinearity. However, there is little discussion about performance of these two models in rice yield prediction. In this article, we adopted two popular NLP models to build and test 12 different model frameworks based on optimal hyperparameter configurations. And we compared model depth as well as bidirectional setting on the rice yield prediction by observing the performance of MSE losses throughout the training process. The results illustrated that both simple and complex models had outstanding fitting for small-sample training, and the depth and direction of the models did not significantly impact the performance of the experiment. But the complex model notably increases the training cost and decreases the convergence rate, implying that it’s not necessarily suitable for time-series problems with small-sample data. Further, the results could provide insights into a deep learning framework construction and hyperparameter selection for subsequent studies with comparable characteristics.

Text-to-image(T2I) generation methods aim to synthesize a high-quality image which is semantically consistent with the given text descriptions. Previous (T2I) generative adversarial networks generally first create a low-resolution image with rough shapes and colors, and then refine the initial image into a high-resolution image. Most stacked architecture still remains two main problems. (1) The final images generated by these methods depend heavily on the quality of the initial image. If the initial one is not initialized correctly, the resulted image seems like a simple combination of visual features from several images scales. (2) The cross-modal fusion methods about text and image that previous works widely adopted is limited in the text-image fusion process. In the paper, we propose a novel generation model, which introduce a one-stage backbone directly generate high-quality images without multi generators and a novel semantic layout deep fusion network to sufficiently fuse text features and image features. Experiments on the challenging CUB and COCO-Stuff datasets demonstrates the ability of our model in generating images, regarding both semantic consistency with input text description and visual fidelity.

In order to improve the accuracy of YOLO series algorithm in detecting small ship targets in SAR images, a target detection algorithm based on improved yolov5 is proposed in this paper. In this paper, The Multi-Scale Channel Attention Module (MS_CAM) is added to the network structure to aggregate local and global feature information in the way of channel attention, which can alleviate the problem of large semantic gap between different scales to a certain extent. In addition, the PANet fusion structure in YOLOv5 was replaced by BiFPN structure to make the network better weight of learning features. The experiment on the open RSDD-SAR dataset shows that compared with the traditional method, the AP value and recall rate of the whole dataset are improved.

Gastric cancer (GC) is the fifth most prevalent malignancy in humans and one of the common causes of cancer-related deaths. In 2020, there will be 1.8 million new cases of GC and 770,000 deaths worldwide, which is the fifth and fourth highest of all malignancies, respectively. The dataset GSE103236, GSE13911, and GSE79973 was downloaded from the GEO database to comparing differentially expressed genes (DEGs) in paired carcinoma tissue and paraneoplastic tissue clinical samples by LASSO and SVM-RFE machine learning methods. Finally, we analyzed the relationship between differential genes and 22 types of immune cells by using the CIBERSORT method, and then provided new ideas for immunotherapy of gastric cancer.

This paper proposes an electronic search algorithm-based transient overvoltage suppression method for the transient overvoltage problem caused by DC faults during the outgoing transmission of new energy clusters from high-voltage DC. Firstly, the relationship between the control parameters of the rotor-side converter and the transient voltage rise of the doubly-fed wind turbine is determined by combining the transient overvoltage principle; secondly, the control parameters are optimized and searched based on the electronic search algorithm. Finally, by simulation, the electronic search algorithm is proved to be effective in terms of convergence and parameter optimization as well as the applications of transient overvoltage suppression.

The existence of bilingual advantage has been under debate because of the inconsistent findings across different studies. Two important factors contributing to this controversy are traditional measurements of bilingualism and the granularity of behavioral data in cognitive tasks. In this study, we address this issue by operationalizing bilingual experience as a multifaceted spectrum and using the Cox proportional hazard model for analyzing behavioral data. Unlike the traditional analysis, the method enabled the inclusion of both correct and incorrect responses. And the results showed that behavioral performance got better with the increased overall bilingual level in the flanker task. Additionally, we found that the active L2 usage time in the immersiona environment was a more reliable predictor of bilingual advantage than the overall bilingual level of traditional analysis.

Aiming at the problems of high leakage risk and serious information loss that may exist in the encryption and decryption of digital images, an image encryption and decryption algorithm based on chaos theory is proposed. The algorithm introduces both one-dimensional logistic mapping and two-dimensional Hénon mapping. In the encryption process, the digital image is dislocation encrypted by logistic mapping and diffusion encrypted by Hénon mapping, and finally the cipher image is obtained. Compared with other similar algorithms, this algorithm has the advantages of low algorithm complexity and high operation efficiency. The simulation experiment is carried out on the MATLAB platform, and the results show that the algorithm can meet the test standard in the field of digital image encryption.

Investment estimation is an essential part of hydropower projects. This paper proposes a learning rate control-enabled deep learning neural network model that can be optimized for different data sizes, especially when small. Then, a DNN model with learning rate optimization is constructed based on the existing hydropower project data in China; finally, the practicality and reliability of the learning rate control enabled example calculations to verify the DNN model. According to the results, the learning rate control-enabled DNN model accurately predicts outcomes. Therefore, it can achieve accurate, fast, and adequate investment estimation for large-scale and middle-scale hydropower projects.

In order to improve the effectiveness of the medium and long-term market power purchase and sale strategy and enhance the trading efficiency of the electricity selling company, a medium- and long-term market power purchase and sale strategy considering the time-sharing power consumption deviation assessment is proposed. In the medium- and long-term market model of electricity, it will suffer from bias assessment when the actual electricity consumption of electricity users deviates from the amount of electricity traded in the medium- and long-term market. The purchasing cost of the power selling company is modeled and the assessment cost of time-sharing deviation of the power users is considered. On this basis, a model for evaluating the utility of electricity sales in the medium- and long-term market is proposed. If the benefit of purchasing and selling electricity exceeds its expected income, the purchasing and selling electricity strategy is effective; otherwise, you need to adjust your trading strategy. Finally, a numerical example is constructed based on the actual data of a provincial power grid to verify the effectiveness of the proposed strategy. The results of the example show that if the time-sharing deviation assessment is not considered, the electricity purchasing and selling strategy of the electricity selling company may lead to higher transaction risk.

Up to now, PLC output power is getting higher and higher, creating more favorable conditions for PLC control coil. With the maturity of sensing technology, the level of manipulator assembly operation is enhanced. The application of electromechanical intelligent(EI) control technology based on PLC technology is studied and analyzed in this paper. It takes YL-235A optical EI equipment as the research object, introduces its basic working process, introduces the hardware composition, electrical control composition, software required for work and use requirements of YL-235A optical EI training equipment in detail, and finally discusses the overall working process and requirements of the equipment in detail, and discusses the application of EI intelligent control technology of PLC technology.

In the blast furnace (BF) ironmaking process, to solve the problem of uncertain thermal hysteresis in pulverized coal combustion system, the uncertain hysteresis is converted into the uncertainty of the coefficient matrix of the system state equation. Based on Lyapunov stability theory and Linear Matrix Inequalities (LMI) method, it is proved that the sufficient condition for stable control of pulverized coal combustion system through state feedback control is equivalent to the solution of a set of LMI. Compared with the traditional PID control mode, the simulation results show that this method can effectively realize the stable control of coal injection in blast furnace.

Internet public opinion sentiment analysis is significant for managing and controlling food safety events. Since emotions can play a decisive role in behavior, netizens’ emotions towards the food safety events will influence their expressions of opinions on the Internet, thereby influencing the development of public opinion on the events. However, few scholars have analyzed the sentiment of Internet public opinion regarding food safety. We employ XLNet, a dynamic text representation method, to build context-dependent word vectors for the distributed representation of Internet public opinion in order to better analyze Internet public opinion on food safety events according to its characteristics. Then, we input the word vectors into Convolutional Neural Networks (CNN) and Bi-directional Long Short-Term Memory (BiLSTM) layers for local semantic features and contextual semantic extraction. Additionally, we introduce an attention mechanism to assign different weights to the features based on their importance before conducting sentiment tendency analysis. The experimental results showed that the average accuracy and Fl values of the sentiment analysis model proposed in this study for Internet public opinion regarding food safety reached 94.12% and 94.61%, respectively, which achieved better results than comparable sentiment analysis models.

In order to further investigate the influence of different structural parameters of the armature on the performance of the electromagnetic rail gun, the Ansys Maxwell simulation software is used to simulate and obtain the results of different diversion arc angle, width and center circle radius of the armature. In order to further investigate the influence of different structural parameters of the armature on the performance of the electromagnetic rail gun, the Ansys Maxwell simulation software was used to obtain the magnetic induction strength and maximum electromagnetic thrust values for different diversion arc angles, widths and radii of the center circle. The simulation results show that for the design of the rail gun structure, the armature width should be as small as possible within a reasonable range, the diversion arc angle should be large enough and the center circle radius should be at a suitable position, which can effectively improve the electromagnetic thrust performance of the rail gun and obtain a larger electromagnetic shielding range.

With the development of Internet of Things, cloud computing, big data and other information technologies, the design objects and design methods have changed greatly. In order to meet the development direction of the times, this paper studies and analyzes the product design method of artificial intelligence based on product semantics. Firstly, this paper analyzes the design method and theoretical basis of product semantics, and then makes an in-depth research and analysis on the formation of artificial intelligence technology, including detailed discussion and planning of behaviorism, connectionism and symbolism of AI. Finally, it expounds the method of using artificial intelligence technology in product semantics design. By making full use of network technology, big data and artificial intelligence technology, a user model which is more in line with the user's thinking is established, and the application and development of artificial intelligence in product semantics are promoted.

In recent years, the vehicle safety detection system is a common information management method of vehicle safety detection in various ports and forts. However, it is generally unable to accurately identify by the system, and still needs manual re-inspection. In this paper, the image algorithm is comprehensively utilized after the functions of automatic identification, display storage and foreign object comparison. A large number of image processing, target recognition, depth learning and other technologies are used to realize the accurate identification of abnormal objects. According to the difference between the normal image and the current image in the database, the vehicle bottom image will be detected for the presence of abnormal objects. From the intelligent point of view, the processing level and accuracy of the detection system can be greatly improved to more than 99.1% and the human intervention and foreign object investment can be reduced to the greatest extent. From the perspective of image processing methods, this paper introduces a new safety detection method of vehicle bottom image data identification.

Multiple object tracking (MOT) is an important subject in applications of computer vision As a subtask of object detection and tracking, vehicle tracking has important research significance. This paper proposes a vehicle tracking and detection technology which is based on improved YOLOv5 and Strong SORT. The YOLOv5 combined with the CBAM attention mechanism work as the detector of Strong SORT in the tracking process, this arrangement decreases computational time. Experiments proved that this proposed algorithm can effectively deal with the problems of object occlusion, target loss, and ID switch. The trained model is easy to deploy for an embedded device, which makes it a very good candidate for a real-time surveillance system.

In order to further improve the linkage mechanism between employment and enrollment plan, talent training, and specialty adjustment, this paper conducts a quantitative analysis and deepening of the Report on the employment quality of undergraduate graduates of Wuhan Business University. This paper uses the analytic hierarchy process to build the evaluation index system of the employment quality of college graduates, focusing on the evaluation criteria, evaluation indicators and calculation methods to calculate the index weight. At the same time, we train the model of BP neural network method, which improved the accuracy of the model, and obtained a more comprehensive, objective, independent and measurable employment quality monitoring and evaluation results.

Engine valve is the core component of the engine, and its quality determines the performance of the engine. In industrial production quality inspection, it is necessary to detect the size of the valve and whether there are defects on the surface. Usually, the quality of the valve is determined by comparing the image of the valve surface with the standard image. However, the existing surface defect detection technology cannot detect the curved surface device. In order to solve this problem, this paper designs a valve size and defect detection method based on computer vision. The experimental results show that the method can quickly and accurately detect the rod diameter, groove radius and surface defects of the valve. The method is practical, robust and real-time.

In order to ensure the effective use of funds, the government must carry out reasonable traffic planning, in which traffic network design is one of the core contents of traffic planning. The traditional transportation planning has unreasonable factors due to deterministic assumptions. This paper assumes that the demand is a random variable, and then considers the time factor. With the cost recovery and link update as the constraint conditions, the real option is used to solve the problem of the flexibility value of the optimization strategy. The optimization model under uncertainty considering the time factor is given, and the real option is solved by using the LSM method. Genetic algorithm with Monte Carlo is used for network design. The example analysis shows that time factor has a significant impact on network construction decision-making, and real options can effectively describe the flexibility of network construction decision-making.

This paper designs a smart home control system based on ZigBee technology. The author analyzes the system design architecture, software and hardware implementation methods, software system construction, key processing flow and other aspects. This design uses home gateway (PC), pan coordinator (network coordinator), home appliance sensor to build ZigBee smart home control system, and realizes communication with remote mobile phone terminal through GPRS wireless transceiver system. On the premise of ensuring energy saving, the system can also make the ZigBee wireless network stable during long-distance transmission and meet the remote monitoring needs of smart homes.

Due to the new crown and other epidemic diseases that make people wear masks to travel, the accuracy of the original face recognition system is affected. To address this challenge, a mask-wearing face recognition system based on an improved attention mechanism is proposed. First, Adding a maximum pooling operation to the CA (Coordinate Attention) attention module, then, placing attention module in the residual unit to form a feature extraction network. LResNet18E-IR is selected as the backbone network. Finally, the ArcFace loss and occlusion probability loss are combined to establish a multi-task network, which further promotes the accuracy of occluded face recognition. The results demonstrate that the system effectively increases the recognition accuracy of masked face and maintains almost the same accuracy as the original model on the unmasked dataset.

Accurate prediction of future car ownership is of great importance for road traffic planning, automobile industry development planning and the formulation of related policies. Therefore, this paper constructs a machine learning-based analysis and prediction model of the factors influencing private car ownership. First, the XGBoost method is used to identify the factors affecting private car ownership based on the data published by the National Bureau of Statistics. Then, comparing the prediction effects of three methods, XGBoost, random forest and neutral network, we found that neural network has better prediction accuracy in the private car ownership prediction model. Finally, based on the neural network method, the future private car ownership in China is predicted. The results of the study showed that GDP per capita and urbanization rate are the two most important factors affecting private car ownership; by 2030, China's private car ownership is expected to reach 438.3 million, 452.56 million and 469.42 million under the low, medium and high development scenarios, respectively.

High precision medium and short term load forecasting is a reliable guarantee for optimizing power grid operation strategy and improving power grid operation efficiency. This paper introduces the machine learning algorithms commonly used in power load forecasting. In order to improve the prediction accuracy of these algorithms, this paper applies parameter optimization to these models. The experimental results show that the prediction accuracy of the optimized model is better. In addition, the prediction accuracy of LSTM model outperforms than XGBoost and LightGBM models greatly.

The internet of things (IOT) is a novel concept. With the rapid development of wireless communication technology, it has quickly become a hot topic. Its basic concept is to communicate, communicate and cooperate with each other around all kinds of things and objects around us through the unique address mode in the world, so as to achieve the common goal. Through in-depth research and investigation, it is found that at present, most banks in China generally adopt the scoring system for personal credit risk assessment (RBI), and the assessment indicators have not kept pace with the times for many years, and basically there is no big change. Obviously, this evaluation method is highly subjective, and the evaluation indicators are relatively backward, slightly rigid and single. In this regard, this paper studies the relatively mature evaluation criteria and methods in the field of information security, and analyzes the information security status of small IOT systems. Integrate the information security features of the IOT to construct a RBI index system for small IOT systems.

In order to solve the problem of "poor effect of spreading fruit knowledge through paper books, pictures and videos", this paper adopts augmented reality technology to recognize fruit and spread relevant knowledge. The specific operation process is as follows: camera shooting, Internet uploading of fruit pictures, server recognition of the relevant fruit, the relevant data of the fruit will be transferred to the user's display device, superimposed on the real fruit. With the help of the augmented reality software designed in this paper, fruit knowledge can be displayed in three dimensions, enhance the interaction and immersion of learning, improve learners' interest, and expand their life knowledge.

Recent advances in the field of speaker recognition (SR) have shown very accurate and high-performance algorithms. When the amount of data is small, its performance drops dramatically [1]. Today, when testing and training involve only small amounts of speech data, identifying the speaker remains a key consideration, as real-life applications can only access speech data for a limited period. Speech recognition-based security systems are one of the main areas of research. In this paper, we will use a graphical CNN algorithm for speaker recognition and compare the effect of different times of speech on the accuracy of the model. The results show that better results can be obtained by using the speech feature data with a deep learning CNN model, with an average accuracy of 86.3%. In addition, the accuracy of speech recognition can be better improved by comparing multiple sets of speech segments with smaller results in this study.

In recent years, with the emergence of the concept of smart campus, it is more and more urgent to build a smart dormitory energy management system in colleges and universities. In this paper, LoRaWAN wireless communication technology architecture is adopted to develop a smart dormitory energy management system, which realizes automatic collection of electricity meter information and intelligent control of air-conditioning equipment, and solves the problems of inefficiency and error in previous manual operation. Through the field operation, all parameters meet the design requirements, and the actual needs of colleges and universities can be well met.

In recent years, deep learning technology has developed rapidly, and the application of deep neural networks in the medical image processing field has become the focus of the spotlight. This paper aims to achieve needle position detection in medical retinal surgery by adopting the target detection algorithm based on YOLOv5 as the basic deep neural network model. The state-of-the-art needle detection approaches for medical surgery mainly focus on needle structure segmentation. Instead of the needle segmentation, the proposed method in this paper contains the angle examination during the needle detection process. This approach also adopts a novel classification method based on the different positions of the needle to improve the model. The experiments demonstrate that the proposed network can accurately detect the needle position and measure the needle angle. The performance test of the proposed method achieves 4.80 for the average Euclidean distance between the detected tip position and the actual tip position. It also obtains an average error of 0.85 degrees for the tip angle across all test sets.

Aiming at the problems of low level of informationization, high risk of construction safety and waste of resources due to lack of information management in traditional construction industry, this paper proposes to design and implement a set of smart site management platform based on IoT and QR code. The platform realizes the monitoring of load-bearing tower cranes and other equipment on construction sites with the help of IOT devices, and receives real-time alarm signals of over-limit or interference collision of tower cranes, which reduces the safety risk; realizes the remote control of over-limit alarm and switch of electricity equipment in office and living areas, which avoids energy waste. At the same time, the system combines QR code technology to realize the information tracking of maintenance orders by supervisors, simplifying their process of declaring faults and dealing with safety hazards, and improving construction efficiency and project quality.

With the development of computer vision technology, object tracking is widely used in various fields of social life. The traditional object tracking method mainly characterizes the object by considering the designed features, and then realizes the object tracking. With the proposal of the deep learning algorithm, the object tracking field has ushered in a new development. The object tracking algorithm based on the Siamese neural network calculates the target position of the current frame by calculating the response between the template image and the feature map of the image to be searched. It is concise and efficient, but it is prone to tracking drift when facing complex backgrounds. The visual attention mechanism is beneficial to highlight the background and suppress the background. Therefore, this paper introduces an attention mechanism based on the Siamese network to improve the performance of object tracking. In addition, a key point for object tracking based on Siamese network is object and background classification. Compared with high-level features, low-level features have stronger discriminative ability. Therefore, a feature fusion mechanism is also introduced and combined with the attention mechanism to optimize the object tracking algorithm based on Siamese network. Compared with the OTB2013 benchmark evaluation, the algorithm in this paper has better tracking performance than the original algorithm.

Point clouds stand out among other 3D data representations for their effectiveness and adaptability. Thus, point cloud analysis has been a popular research topic in recent years. However, due to complex features, numerous scene categories, occlusion, and noise, reliable semantic segmentation in huge scenes is difficult. Existing point cloud segmentation techniques have two drawbacks: 1) since different nearby points are frequently considered similarly, it is impossible to accurately characterize the relationship between the center point and its vicinity. 2) How to simultaneously address the issues of point cloud semantic segmentation efficiency and data balance in huge scenes. A network paradigm based on the Gated Graph Attention Network (GGAN) is suggested in order to get around these restrictions. To improve local feature extraction, GGAN can highlight not only the significance of various neighbor points but also the significance of various representation Spaces. Recall Loss is employed in the meantime to address the issue of data imbalance brought on by random sampling in order to increase effectiveness. The experimental findings on the Semantic3D dataset revealed that the average intersection ratio of GGAN reached 77.1%, which could ensure the segmentation accuracy of large categories (natural landscape, buildings), as well as small categories (cars), simultaneously. This segmentation performance was superior to that of the existing segmentation methods.

In order to solve the problem of mechanism design of new energy participating in day-ahead spot market and promote the transaction bidding of new energy participating in spot market to improve its production efficiency, a new energy participating day-ahead spot market clearing method based on two-layer architecture was proposed. The upper structure is a conventional power unit transaction clearing model based on new energy generation prediction, which aims to maximize the consumption of new energy through conventional power unit transaction clearing and evaluate the expected wind and light curtailment power. The lower structure is a new energy transaction clearing model based on expected wind and light power abandonment, which aims to share the expected wind and light power abandonment in a market-based way and promote new energy to improve its production efficiency. Through the orderly connection of two layers of architecture, the generation plan of conventional power units and new energy power stations can be optimized and cleared. Finally, a numerical example is constructed based on the actual data of a provincial power grid to verify the effectiveness of the proposed method. Compared with the traditional clearing method, the proposed clearing method can improve the efficiency of new energy power generation by more than 10%, which is helpful to further improve the efficiency of power grid operation.

With the continuous development of maritime transportation, the flow of ships and the density of navigation are growing continuously. The traffic of ships is getting busier. As a result, the number of maritime traffic accidents is more than before, and the navigation environment is becoming more complex. Therefore, it is necessary to make a scientific and reasonable assessment. This paper identifies the factors that affect the safe navigation of channels from the perspective of navigation environment in a traditional system engineering way. In addition, this paper pursues establishing a model based on extension matter-element theory for these factors, combined with AHP method, to evaluate the safety of the traffic navigation environment of channels. 3 channels of the Ningbo Zhoushan Port are selected as an example to verify the accuracy of the model. The safety assessment results of Ningbo Zhoushan port show that they are all of low risk. Finally, it is concluded that the matter-element model is accurate for navigation environment safety assessment.

There are many more accidents of gas explosion in heading face than in other working faces. For this reason, the safety measures of the underground heading face have become particularly important. With the trend of the coal mining face becoming longer and longer, the requirements for long-distance excavation ventilation have been increasing. How to conveniently and efficiently solve the energy-saving problem of long-distance ventilation has become a hot topic. In this paper, the ventilation design scheme of the 7323 upper heading face in Panxi coal mine is determined based on the 7323 upper heading face in Panxi coal mine. In addition, through VB software, the function of accurately calculating the required air volume is realized to ensure that the local fan can provide sufficient air volume, which provides the current conditions for the analysis of safety technology and long-distance ventilation.

Hubei is the second batch of power spot market pilot areas, and the construction of the spot market is imminent. At present, there are many difficulties and challenges in the construction of the spot market in Hubei. First, the price of thermal coal has fluctuated significantly in recent years, and many thermal generators have incurred losses due to the rising fuel prices, disrupting the stable operation of the market. Secondly, Hubei has a high installed capacity of hydropower, and there are many large-capacity hydropower generators with storage capacity. How to participate in the spot market of hydropower generators is a difficult problem that needs to be solved urgently at present. Based on the actual data of Hubei Power Grid, the electricity spot market is simulated. The market price and the profitability of various generators under different coal prices and different periods of hydropower dispatch are compared. The results show that in the spot market, the market price can correctly reflect the real price of electricity and ensure the profits of various generators. For hydropower generators, compared with the daily dispatch, the monthly dispatch of hydropower can better stimulate the storage capacity adjustment ability of hydropower and reduce the total cost of electricity consumption in the whole society.

With the popularization of vehicles in the world, the problem of driving safety has also received more and more attention. Among them, understeer and oversteer are the main causes of many traffic accidents. In order to explore the causes of understeer and oversteer and propose solutions, this paper builds models of two real cars, BMW M3 and Chevrolet Cavalier, in python, and simulates two turns of lane change and U-turn, as well as the side of the vehicle under different parameters. The simulation of the variation of the slip angle leads to the conclusion that the variation of the slip angle is the main cause of understeer and oversteer. Subsequently, this paper investigates and discusses other causes of understeer and oversteer, and finds that vehicle weight and driving speed are also the causes of understeer and oversteer. The solution is discussed, and specific solutions are given, which is intended to ensure the safety of drivers who encounter similar emergencies.

Since the construction of the domestic electricity spot market in 2017, the problem of market power abuse caused by high market concentration and poor demand-side flexibility has gradually become prominent, and the electricity market operators have gradually established a market power regulation system, with bidding price replacement and clearing price replacement as two common market power regulation measures. This paper constructs a market power regulation model based on the principal-agent mechanism and analyzes the market equilibrium and the decision interval of market participants of the above two market power regulation measures. The results show that both market power regulation measures can make the market equilibrium price be the regulated price, but under bidding price replacement regulation the market clearing quantity distorts the market resource allocation and the decision interval of market participants is narrower. Therefore, the clearing price replacement is recommended as market power regulation measures. Finally, this paper solves the optimal strategies of market participants and market operators, and concludes that the optimal strategies of market participants under both market power regulation measures are to make the market regulation ineffective, which verifies the effectiveness of market power regulation measures.

Carbon market and electricity market will play an important role as market mechanisms in implementing the double-carbon goal and building a new type of power system. At present, the national carbon market has been traded online, and the pilot construction of power spot, as the core of electricity market reform, is accelerating. The two markets are closely connected and affect each other, and have a far-reaching impact on the development of the power industry. Starting with the analysis of the relationship between the carbon market and the electricity market, this study introduces the construction progress of China's carbon market, focusing on the impact of China's carbon market on the electricity spot market, including the impact of the carbon market on the trading behavior of spot market participant, market clearing and market operation and puts forward mechanism suggestions to promote the coordinated development of the two market.

According to the actual internal force and internal force calculation by engineering space ideal hinged truss model for different problems of UHV tower design,1000 kV in UHV Engineering double circuit lines on the same tower SZ304 as an example, by using SAP2000 finite element program, set up truss element, beam-bar element and the beam element of three different mechanics model according to the angle steel and steel pipe two component types. Research shows that: the principal axis force and the natural vibration period of the three models are similar to each other. The beam element and the beam-bar element model of principal end bending moment calculation results are similar. The lower part of the tower body and steel tube tower variable slope at principal moment of secondary stress is greater than 20%, the design should be considered. The steel tube tower, under the same conditions, the increase of material length diameter ratio, can significantly reduce the secondary stress influence.

This paper first analyzes three major problems that can be encountered in video behavior recognition tasks: sampled blocks cannot be focused on motion regions, global motion affects recognition results, and the network's spatio-temporal modeling capability is weak. To address these three problems, we propose the SlowFast behavior recognition algorithm (MASlowFast) that incorporates motion saliency as an application scenario for mine personnel safety behavior recognition. The specific solutions are the sampling method based on motion saliency, the extraction of motion boundary features, and the spatio-temporal segmentation strategy of fast and slow channels. Finally, we validated the effectiveness and accuracy of the algorithm in this paper by ablation experiments on UCF101 dataset and HMDB51 dataset.

The question answering method based on knowledge atlas has become a hot research field in natural language processing, and has been gradually applied in the field of electric power. In order to solve the problems such as the difficulty of using unstructured text data and the shallow application depth of equipment knowledge, which exist in the process of power operation and maintenance personnel carrying out transformer equipment inspection, this paper puts forward the technical framework of transformer equipment knowledge map, adopts the intelligent identification and extraction method of structured text, and establishes the calculation model of equipment semantic similarity. On this basis, this paper verifies the functions of automatic audit of transformer equipment status evaluation report, auxiliary diagnosis and recognition of equipment fault knowledge. The experimental results show that the recognition method based on knowledge atlas improves the accuracy of fault defect text recognition, and provides a new idea for improving the operation and maintenance efficiency of field equipment.

This article introduces the monitoring method of ship navigation under the situation of existing mission. Mission process monitoring and monitoring tips feedback can be achieved through data structuring, current mission accessibility, and final mission accessibility calculation. When the mission can’t be performed as required, calculate the navigation adjustment strategy and mission adjustment strategy, give the best adjustment suggestion to help commanders to make corrections. The algorithm process allows man-in-loop, when the commanders adjust the mission based on strategy, the subsequent calculations will be updated synchronously to meet the real-time and effectiveness of calculating.

Based on the current situation of less research on the risk assessment of overboard-man at sea and a large number of search and rescue experiences in making decisions based on qualitative judgments, a comprehensive risk assessment model was established through the combination of analytic hierarchy process and gray fuzzy theory, and verified by real cases, and the results calculated by the model were basically consistent with the actual situation. This lays the foundation for further development of a Web-based risk assessment system for overboard-man.

To establish a detection network appropriate for buildings in remote sensing images and lessen the issues including poor detection effects, missing detection and false detection due to the deficiency of detailed features, this paper conducted the design on the basis of Segformer network to solve the problem, coupled the transposed convolutional networks at the decoder stage, and addressed the issue of missing feature semantics via adding holes and fillings. Multiple normalization layers and activation layers were cascaded after the convolution layer to avert overfitting regularization expression and guarantee the classification of stable feature parameters, so as to further advance inter-class differentiated extraction. Ablation experiments and comparison experiments were conducted on AISD, MBD and WHU remote sensing image datasets: The robustness and effectiveness of the improved mechanism were demonstrated by control groups of ablation experiments; in comparison experiments with Hrnet, PSPNet, UNet, Deeplabv3+ and the original detection algorithm, the mIoU of AISD, MBD and WHU was improved by up to 12.83%, 28.82% and 14.26%, respectively. The experimental results indicated that the improved method was better than the comparative methods such as UNet, and had better effects on integrity detection of building edge as well as the reduction of missing detection and false detection.

With the increasing number of monitoring equipment, it is necessary to study whether there is a collision between objects in the image. In this paper, taking the monitoring of whether the transmission wire and the plastic bag collide as an example, aiming at the shortage of collision between objects in the image, especially the inaccurate detection results and long calculation time, we propose a collision detection algorithm. Firstly, detect the target wire and plastic bag in the picture, and judge whether there is a coincidence between the wire and plastic bag in two-dimensional space. If there are no overlapping spots in two dimensions, there is no collision between the wire and plastic bag. On the contrary, measure the position coordinate information in its three-dimensional coordinate system by binocular vision digital measurement method, and surround the wire and plastic bag by bounding boxes respectively. Finally, judge whether there is collision between them by calculation. Compared with the collision detection algorithm based on two-dimensional space, the accuracy of detection is improved. Compared with the complete use of binocular distance measurement for collision detection, the detection speed is significantly improved and the detection efficiency is effectively improved.

Sign language recognition is a popular and important problem in computer vision. This paper uses MediaPipe as a method of feature extraction to extract coordinates of the joint points of the human body. The sign language dataset used in this paper is 64 Argentine sign languages (LSA64). Then the coordinate data is input to Long-Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) neural networks. After testing two models in 640 videos, the result has shown that the accuracy of the trained LSTM and GRU model reaches 94.0625% and 94.5312%, respectively. It is able to recognize 64 Argentinian sign language words or phrases. As a convenient method of sign language recognition, it is feasible in reality.

Most of the jade on the market now comes from Myanmar, Guatemala, and a few from Russia. The gemological properties of jadeite from different producing areas are consistent. However, in the middle-end jade market, under the same quality, the prices of Guatemalan jade and Russian jade are generally lower than those of Myanmar jade, so some illegal merchants will use Guatemalan jade to impersonate Myanmar jade. Due to the continuous improvement of jade counterfeiting technology, traditional identification methods can no longer meet the demand. In order to protect the rights and interests of consumers need to establish a rapid and effective jade origin traceability method. In this paper, through the (LA-ICP-MS) trace element dataset and the method based on weighted extreme learning machine, AdaBoost and incremental learning fusion, the jadeite discrimination model of different producing areas is established to realize the intelligent discrimination of jadeite producing areas. The recognition accuracy of integrated learning algorithm is more than 80%. Compared with the basic extreme learning machine and weighted extreme learning machine, it can be found that the classification accuracy of integrated learning algorithm is higher and more stable.

Aiming at the disadvantages of slow convergence speed and unable to reduce recognition errors quickly, BP neural network is used for intelligent recognition of computer images. On this basis, the image recognition model is established by using BP neural network, and its modeling and modeling are carried out. Under the same experimental conditions, different recognition algorithms are compared. The results show that this algorithm has a better convergence speed, can effectively reduce errors, and can effectively improve the recognition rate of images.

Neural network accelerators provide deep learning algorithms with a faster and more energy-efficient way of computing. Among those hardware accelerators, the bit-level sparsity accelerator seeks the “0” bits in weights and activations to skip the ineffectual computation. Currently, bit-level sparsity accelerators have well explored the possible sparsity in various CNN models with different parameter precision. However, the large number of parameters in CNN occupies many memory storages and troubles designers to exploit the sparsity. Therefore, this paper introduces the Binary Neural Network (BNN) to shorten the computation time and energy consumption with acceptable losses of accuracy. In this paper, the research approach is based on both bit-level sparsity accelerators and BNN algorithms. Some current studies on bit-level sparsity accelerators are shown first, then BNN is introduced. The result provides a framework of bit-level sparsity deep learning accelerator embedded with the binary neural network.

Aiming at the problems of ghosting and distortion in the process of UAV low-altitude aerial image Mosaic, an improved algorithm for UAV low-altitude aerial image Mosaic was proposed. The algorithm firstly through the SIFT features of UAV image image mosaicing, and then based on the SIFT algorithm in unmanned aerial vehicle (UAV) image matching problems that exist in the process, to improve the image matching in the process of feature extraction methods, effectively reduced the number of feature points extraction, and uses the direct average fusion method for low pass filter, get spliced images. The simulation results show that the algorithm can effectively solve the problems of ghost and distortion in the process of UAV image Mosaic, and the algorithm runs fast, which has certain military application value.

In order to solve the problem that manual discrimination of gastric cancer pathological images is time-consuming and laborious and requires high relevant medical expertise, this paper proposes a method based on improved convolutional neural network to assist recognition of gastric cancer pathological images. In this paper, residual network (Resnet) was used as the basis for optimization and improvement, and the network structure was reasonably adjusted to improve the adaptability of the recognition task. The two-dimension attention mechanism of Convolutional Block Attention Module (CBAM) is introduced to improve the accuracy of the model. The knowledge of transfer learning is used to accelerate the training process and the fitting of network training. In order to increase the generalization of the network model, the experimental images were data enhanced. The experimental results show that in the fifth layer network structure CONV5, replacing the first convolution layer with the attention mechanism module is better than replacing the second convolution layer and replacing both convolution layers at the same time, and the attention mechanism module is a serial arrangement of CAM+SAM. Compared with VGG16, Alextnet and Googlenet, the improved network structure has the highest accuracy of 96.58%.

Although deep CNN networks have excellent image classification performance, they do not provide interpretability, and furthermore existing work reveals that these models have complex internals, for example, mysterious polysemantic neurons activate to multiple features. In this work, we analyze the intermediate data of the network dissection paper made by Bau et al. to understand to what extent polysemantic neurons exist. We divide the polysemantic neurons into five types and calculate the percentage of each type by sampling. We find that above 50% neurons identify one concept but there are a quite proportion of neurons that recognize two or more features. This can explain the high classification accuracy and some capacity saving of a deep CNN. By case studies, we draw some conclusions and hypotheses: First, unlike the human visual system, a CNN cannot distinguish detailed features (metaphor: a CNN is like a nearsighted eye). Second, the reason that the CNN is prone to adversarial attacks may be partially due to the polysemantic neurons. Third, polysemantic neurons may partially explain why people wrongly visualize one thing as another in neuroscience.

Because the image has the characteristics of large amount of information and high redundancy, the traditional AES, DES and other encryption methods applied to text and numbers can not be effective. Chaotic sequence algorithm has natural similarity with cryptography, such as extreme sensitivity to initial value, ergodicity, non-periodicity, pseudo-randomness and very high iterative rate. as a result, the encryption of chaotic sequence algorithm in the field of electronic conference image will be better than the traditional encryption method, and it has become a new direction to solve the problem of electronic conference image encryption. However, the limited computer accuracy will also reduce the randomness of chaotic sequences. in order to make up for this defect, this paper uses the idea of hybrid encryption to design a more secure and efficient hybrid chaotic sequence encryption algorithm.

Underwater images are easily affected by external factors such as light and water impurities, resulting in low segmentation accuracy of underwater images. Aiming to solve the problem, this paper proposes a semantic segmentation method of underwater images by introducing a new branch to improve the structure of the DeepLab v3+ model. It exploits the low-semantic features of the input image to improve the performance of the network. In the new branch, down-sampling is used to adjust the resolution of the input image, and the dimension of the feature map is adjusted by 1×1 convolution. The attention mechanism is used to focus on the channel of the feature map, and finally, the feature map of the existing branch is merged. In the experiments on the SUIM underwater dataset, the average intersection ratio of the model is 72.88%, and the average pixel accuracy is 84.03%. In the experiments on the PASCAL VOC dataset, the average intersection ratio of the model is 75.85%, and the average pixel accuracy is 84.5%. Compared with existing mainstream algorithms, the proposed algorithm achieves better results.

Technological transformation investment is an important part of the investment in power grid enterprises. With the passage of time, the investment scale of technological transformation of power grid system has gradually improved, and the contradiction between high-intensity investment, rigid cost growth, slow power growth and difficult growth of efficiency is becoming increasingly prominent. It is more difficult to maintain steady operation and complete the profit target, which puts forward higher requirements for the investment decision of power grid technological transformation. Therefore, strengthening the input-output evaluation of technological transformation investment projects of power grid enterprises can further improve the efficiency of technological transformation investment of enterprises, so as to improve the level of operating efficiency of enterprises.

High frequency zoom CCD camera, magnification adjustable microscope and other devices were used to visualize the carbon deposition of injector. The image processing system and image comprehensive analysis system were constructed based on MATLAB. The Canny algorithm edge detection. Regionprops function and Viscircles function are used to get the area of the hole circle in the image processing system. The area proportion and thickness of carbon deposit are used as indicators in the image comprehensive analysis system. When the area of carbon deposit is greater than 20%, the injector needs to be cleaned. The four groups of fuel injectors were photographed and measured. It was found that the area proportion of carbon deposition was between 16 ~ 30%, and the thickness of carbon deposition was between 0.02 ~ 0.04 after working for a period of time.

Based on MATLAB, a processing method which can accurately measure the spray characteristic parameters in the image was proposed, including background difference, image enhancement, threshold segmentation, edge detection, morphological operation and other algorithms. The macroscopic characteristic parameters of spray were extracted, and the variation rules of penetration distance and spray cone angle under different factors were analyzed. It is found that under the conditions of 140 MPa and 160 MPa injection pressure, 1 MPa and 2 MPa back pressure, and 0.315 mm and 0.366 mm nozzle diameter, with the increase of spray pressure, penetration distance increases and spray cone angle decreases. The penetration distance and cone angle decrease with the increase of hole diameter. The cone angle increased with the increase of back pressure, and the penetration distance showed the opposite trend.

The fittings detection of transmission lines plays a vital role in ensuring the safe and stable operation of transmission lines. The fitting detection method based on deep learning only scales the original image to a smaller size. However, it ignores the high-definition resolution of the aerial image in the transmission line, resulting in the loss of rich features in the high-resolution aerial image. In order to solve this problem, we observe that aerial images of fittings are concentrated in a particular area of the aerial image. Therefore, we propose a cascading YOLOx model, including Dense Target Regions YOLOx (DTR-YOLOx), which can detect dense target areas, and Multi-Fitting YOLOx (MF-YOLOx), which can detect multiple categories of fittings. In addition, an algorithm based on the connected regions is proposed to automatically generate the dense target region for DTR-YOLOx training, reducing manual labeling costs. Furthermore, the EIOU loss function is introduced to improve the precision of the model's coordinate regression. Experiments show that the AP_0.50:0.95 value of our proposed model is 18.4% higher than that of the YOLOx model.

The increasing number of car ownership aggravates the traffic problem, so the intelligent transportation system is proposed. As a component of intelligent transportation system, vehicle identification system has been studied by researchers all over the world. In this paper, the theoretical knowledge of deep learning is introduced at first, and then the migration training is carried out by using VGG-16, RESNET-50 and the improved neural network based on these two kinds of networks on the self-made model dataset. In the experiment, a verification set was used to verify the training effect, and the training results of the four models were compared in terms of the number of parameters, training time and accuracy. The training results show that the improved RESNET-50 has the best performance, and its accuracy rate reaches 93%.

Image stitching is widely used in computer vision where it is often necessary to first manually specify the “left to right” or “right to left” transformation. To solve the problem of evaluating which stitching mode is better, on the basis of studying of SIFT algorithm’s image stitching, this paper calculates the polygons after perspective transformation by analyzing two methods of homography matrices and proposes an measurement that refers to the degree of blurring caused by the oblique angle on the long side of the transformed polygon rectangle as stretching degree. It was used as one of the evaluation index for the quality of image stitching. The experimental results show that this measure can more automatic and robust to choose the better mode of image stitching.

With the rapid development of social economy, lifelong learning has become one of people’s main tasks. How to recommend appropriate learning resources according to learners’ preferences has become a concern by more and more researchers. This paper takes the personalized recommendation of learning resources as the research object. Firstly, the semantic relationship of learning resources is extracted by knowledge map to realize the vectorization representation of learning resources. Based on this training convolutional neural network model, learning resources recommendation list is generated. Secondly, user-based recommendation algorithm is adopted in user behavior analysis with the purpose of generating learning resource recommendation list. Finally, the above recommendation list is gathered to realize the personality of learning resources, which can realize information filtering to the greatest extent, reduce the user information load. The results show that when the fusion ratio is close to 0.5, the algorithm recommendation accuracy and recall rate are higher, and the user’s preferences and interests have a higher degree of coincidence; the deviation between the recommended learning resources download amount and the user’s actual learning resources download amount is smaller; the proportion of recommended learning resources in the total learning resources is increased; and some unexpected learning resources can be found and recommended to the customers. When the sparsity of experimental data set is smaller, the accuracy, recall rate, coverage rate and MAE index of the personalized learning resource recommendation algorithm proposed in this paper are significantly higher.

An algorithm for identifying the action of quantitative refueling that can be deployed to edge equipment is proposed for the problem that refueling' action in production scenarios is not subject to real-time supervision. The algorithm firstly uses a YOLOv5s-improved object detection network for rapid human detection, then uses a tracking algorithm that combines IOU and histogram similarity to track the detected human. The traced sequence images are used to predict the skeletal key-point sequence of the human body through a quantitative pose estimation network, and finally, the skeletal key-point sequence data is input into the fully-connected network classifier on the sixth floor for action classification, to determine whether the refueling's actions are normally completed. Experimental data show that the algorithm greatly reduces the network weight and calculation amount. The human body detection speed on the BITMAIN Sophon SE5 terminal can reach 18 ms, and the action detection accuracy can reach 95.92% on the actual scene dataset.

Virtual-real occlusion is a key step in augmented reality systems. The traditional virtual-real occlusion method has some problems, such as low accuracy and poor adaptation to the scene. On this basis, this paper proposes a virtual real occlusion method based on monocular camera. Firstly, AKAZE algorithm was combined with Tanimoto similarity estimation to find the optimal pair of feature points, and the relative depth of feature points on adjacent frames was calculated by using the triangulation method in SFM algorithm. Then, the real depth value was obtained by combining the translation scale. Finally, the depth value of virtual and real objects was compared to realize the virtual and real occlusion of monocular camera. The experimental results show that the virtual-real occlusion based on monocular camera is realized and the occlusion accuracy is high.

Team role research has been a hot research issue in recent years, and a good team background is very important for team role research. A good team is not a simple collection of multiple people but many people with different roles and functions (leadership, group leader, staff), they have close cooperation and strict relationship between superiors and subordinates. Team role research helps us to understand the function and behavior of the members of the team. Compared with the traditional walking method, Mjwalk can walk on multiple independent team networks, showing advantages in multinetwork joint learning. The conv_att method proposed in this paper solves the downstream role recognition task and improves the recognition effect. We evaluate the performance of the Mjwalk and conv_att framework with real team data, and improve f1_score and accuracy.

As an important part of the transmission channel, the intelligent monitoring of the construction operation status of the transmission pole tower is an important research direction of the intelligent inspection of the transmission channel. The development of artificial intelligence technologies such as deep learning provides good conditions for the intelligent monitoring of the construction status of transmission poles and towers. However, the application of deep learning models requires high-quality sample data. At present, there is still a small sample library of the construction status of transmission poles and towers. The construction of a set of transmission pole tower construction operation status sample database that can be used for deep learning model is of positive significance for intelligent inspection of transmission channels. This study focuses on the characteristics of satellite remote sensing images under different operating conditions, compares the remote sensing images with real images, and gives a detailed process for constructing a sample library for monitoring construction operation status. It expands the innovative application of artificial intelligence in the intelligent inspection of transmission channels, and further improves the modernization level of construction operation monitoring and supervision.

Traditional examination rooms rely on invigilators to monitor examinees in real time and use cameras to help invigilate the exam, which is prone to problems such as incomplete monitoring, inadequate response to cheating. This paper builds an abnormal behavior recognition model in examination room based on LOLOv5 and the cascading attention mechanism. The model effectively improves the backbone network of YOLOv5, and combines the cascading attention mechanism to enhance the features. Finally, the model is tested on the self-created dataset. The results show that the examinees abnormal behavior detection results of the proposed model are P (92.53%), mAP (93.52%), fps (0.547). Compared with several classical abnormal behavior detection algorithms, the proposed algorithm has higher accuracy and recognition speed.

It is well known that animals have a special meaning for humans, and in biology humans have coexisted with animals for a long time. From the ancestors of human beings there has been an inseparable relationship with animals. And this relationship also makes the human visual system seem to have a more special visual processing mechanism for animals than other targets. To find out whether this mechanism exists only in animals, we performed an experiment. The experiment was a two forced choice (2-AFC) task. Since the scene has an effect on object recognition, we will use animal stimuli without background for comparison experiments with non-animal stimuli. We mixed animal stimulus images and non-animal stimulus images in a disordered manner to form stimulus image sets (50 images each), all of which were without background. Our results showed that subjects had faster reaction times for the animal stimulus pictures than for the non-animal stimulus images, with 524 ms for the animal stimulus pictures and 547 ms for the nonanimal stimulus pictures. The subjects' correct judgment rate for animal stimulus images was higher than that for nonanimal stimulus images, with 96.1% for animal stimulus images and 91.6% for non-animal stimulus images.

Recent years have witnessed a spurt of progress in different marketing strategies, resulting in an increasing number of research in how promotion could be more effective with the combination of the knowledge from other domains. Nowadays, advertising campaigns for different consumer goods in promotion and marketing have become a truistic way to better catch consumer`s eyeball as well as to increase companies` sales. In this paper we have uncovered the “irrationality” behind the consumer behavior that each choice consumers made, that is, we analyzed the consumer preference for decision making in maximizing their utility of products through unconscious processes, and this is so called “neuromarketing” So this paper gives a brief analysis along with the results of basic waves extracted from different EEG signals by comparing the variation of the waves when consumer facing different choices, using LSTM model to machine learning and to predict consumer`s choice, plus the Savitzky-Golay filter and wavelet transform as the methods used for data preprocessing and feature extraction respectively. To make it more critical and comparable, we also introduce other ways that lies in the preprocessing and feature extraction stages. However, by repeating the experiment with different variables, that are, the different methods, we have discovered that there is a kernel of limitations arise by using other methods. So, in this paper we mainly discussed the model we built based on LSTM, with informative discussion on a range of various methods and methods we used in data preprocessing and feature extraction.

Four channels of surface EMG signals were collected from six gesture movements, and their time-domain and time-frequency domain features were extracted. Principal component analysis is used to reduce the dimension of features. For the dimension-reduced features, the decision tree is used to build a gesture recognition model. The experimental results show that the overall accuracy of the model for classifying six kinds of actions is more than 97%, which has good application value.

The image shaping of tourist attractions is one of the important factors to attract tourists. This paper explores the perceived impression of Fuzhou in tourists' mind and the perceived dimensions of tourists' attention by studying the review texts of tourists when they visit Fuzhou. By collecting reviews about Fuzhou on Tongcheng.com, Ctrip.com, and Donkey Mama, the destination image of Fuzhou is obtained through LDA (Latent Dirichlet Allocation) theme modeling model and SVM (Support Vector Machine) sentiment analysis model text data mining, so as to objectively extract the perceptual dimensions of tourists when visiting Fuzhou. The study found that the perceptual dimensions that tourists focus on during their visit to Fuzhou are mainly, namely, "tourism services," "entrance fee," "human landscape," "natural landscape;" "natural scenery," "history and culture," and "activity experience". The results of the study have some reference significance for Fuzhou in construction and promotion.

Arteriosclerosis is an independent predictor of cardiovascular disease which is one of the most important causes of human death. Retinal vascular features parameters are important for intelligent diagnosis of arteriosclerosis. These parameters depend on accurate artery and vein (A/V) classification result, but the convolutional neural networks (CNN) used for retinal A/V classification all have some A/V misclassification. In this paper, an optimization method is proposed to optimize the A/V classification of CNN basing on vascular topology. A multi-level lightweight algorithm is proposed to obtain the lightweight information of the vascular skeleton. Using the knowledge of graph theory, the multi-lightweight algorithm preserves the continuous and complete center line structure information. And then, we use multilevel Dijkstra algorithm to estimate the vascular topology on the lightweight vessel skeleton graph, and use the direction algorithm to obtain the vascular directed topology. Lastly, the topology labels algorithm is found on the type of branch nodes to deliver the A/V information along the vascular directed topology to optimize the A/V classification. Experiments clearly show that our optimization method can correct effectively the A/V misclassification.

Aiming at the problem of small scale and little characteristic information of vehicles in remote sensing images, this paper proposes a multi-scale attention detection network. Firstly, VGG16 backbone network is used to obtain multi-level image features, and the cascade feature pyramid module and the dilated attention module are added to enhance the multiscale features of the target. Secondly, the focal loss function is added to the classification task to improve the fitting ability of the network to samples. Finally, the comparative experiments are conducted on the dataset. The results show that the multi-scale attention detection network has a high detection accuracy and its mAP reaches 94.56%.

At present, apple disease control mainly relies on manual recognition, which causes a waste of human resources. In order to improve the efficiency of disease detection, this study focuses on apple leaf disease images in natural scenes. A lightweight network RepSSD consisting of feature extraction module, multi-scale feature recognition, and structural reparameterization is designed to detect apple leaf diseases. Replace the backbone part of the network with RepVGG-A0 without identity branch, and the auxiliary convolutional layer is replaced with a re-parameterized RepConv module without identity branch on top of the SSD. The CBAM attention mechanism module is also added to ensure adequate extraction of global features of apple leaf disease images in natural scenes. The experimental results show that the improved SSD has better detection accuracy and real-time performance in apple leaf disease recognition with an accuracy of 93.05%, which is 2.56% better than the original SSD, and the overall number of parameters decreases by 47.8%, the computation volume reduces by 86.7%, so the overall effect is better than the current common object detection models. Therefore, the algorithm can be better applied to apple disease control. Therefore, the model can be better used for apple disease control.

Human action recognition is one of the core tasks in the field of computer aided driving. Considering that the auxiliary driving system requires high real-time, and the hardware requirements can not be too high, it is proposed to identify human behavior in a single image. Considering that the night illumination is insufficient, and the infrared camera receives the infrared radiation of the object, it can work at night without the influence of visible light. Therefore, we focus on the human behavior recognition in infrared images. According to the scale of the problem, we first use AlexNet with moderate network depth as the backbone network, then improve the network, modify the classification output layer of the network according to the classification number. After preprocessing the dataset to adapt to improve AlexNet, we trained and tested the network. The experimental results quantify the classification performance of the network. Experimental results show that the proposed algorithm mean average precision, average recall and F1 score are better than traditional methods.

Fire distribution simulation is an extensive research topic, which belongs to the important research content of system of systems confrontation simulation. This paper presents a model of air defense fire distribution loop, which is relatively efficient, convenient and reliable, and can be used to guide the implementation of related algorithms. The computational fire distribution loop model includes the calculation of target distribution line and target distribution loop. This model can realize the effective distribution of air defense fire to targets. This model can guide the analysis of fire superiority and the calculation of automatic fire distribution. Combined with the actual project, good results have been achieved.

Catastrophic forgetting is an undesirable phenomenon in convolution neural networks and iCarl is an effective algorithm for preventing catastrophic forgetting. However, a potential defect that more similar tasks result in severer catastrophic forgetting (STSF) in iCarl is explored in this work. The reason of STSF is that similar tasks are prone to occupy similar feature channels and similar feature representations, thus they can be replaced by each other easily. Based on these findings, a novel method Similar Margin Loss (SML) is proposed. SML aims to make feature representations of samples from the same task compact while making feature representations from the different tasks differentiable in the feature space. Experiment results show that SML is effective in alleviating STSF.

In order to better achieve the continuous jamming ability of chaff and effectively evade the enemy radar tracking in space, this paper proposes a chaff combination detonation mode. A physical model is established based on LS-DYNA, and it is numerically simulated by fluid solid coupling method. The diffusion vector diagram of six groups of chaff bombs detonated through leads under the action of micro explosion is simulated. Finally, three groups of chaff bombs are tested and verified, The farthest diffusion distance obtained from the experiment was recorded and the radar cross-sectional area was calculated. It can be concluded from the results that the chaff combination detonating method is feasible, and has good continuous diffusion characteristics and large-area scattering effect, which provides guidance for practical engineering application.

This paper aims to explore the design of sports training system based on wireless sensor network. According to the management of electronic information technology used in physical education at present, it can adapt to the needs of the development of the times and analyze and design the demands of physical education in the WSN environment. The research shows that the use of wireless sensor technology can improve the performance of the system, improve the interactive ability of the sports training system by 16.0%, and improve by 8.0%. The research results show that the design of physical education system under the environment of wireless sensor network can meet the needs of practical applications, and can produce good results.

Road pollution identification is a very important link in keeping the road clean and the safe running of vehicles. In the process of mine operation, it is especially easy to produce sediment and other garbage to cause road pollution. However, currently widely used pollution detection methods are not very accurate segmentation of road pollution. In this paper, an efficient semantic segmentation algorithm EEDNet based on codecing and decoding structures is proposed to solve the detection and segmentation problem of mine road pollution. The designed pollution segmentation model is trained through the data set collected by the cameras arranged in the mine monitoring area, and a better pollution segmentation effect is obtained. Especially compared with the most advanced pollution segmentation methods, the model designed by us greatly improves the accuracy of road pollution segmentation, and provides road information support and safety guarantee for the safe driving of road vehicles in mine working areas.