This PDF file contains the front matter associated with SPIE Proceedings Volume 12971, including the Title Page, Copyright information, Table of Contents, Preface, and Conference Committee information

With the progress of science and technology and the development of industrial automation, the machine vision technology is widely used in the field of industrial automation detection. Especially the size measurement and surface defects detection of some small particles, the machine vision has clear advantages in precision than human eyes. The application of machine vision is developing towards lager magnification and high resolution. Therefore, the matching lens is needed for this application. Aiming at the needs of multi-surface detection of semiconductor thermoElectric cooler (TEC) components, this paper describes a design of a telecentric lens. Using the telecentric lens, two adjacent surfaces of the TEC components can be imaged simultaneously. This lens includes two right angle prisms, a cubic beam splitter, five lenses and a cemented doublet. Its magnification is -1.5 and and the focal length is 65mm. The maximum distortion of the lens is controlled within 0.2% and the MTF is greater than 0.4 at 100lp/mm. The lens can be adapted to 2/3 "CCD, which can meet the requirements of machine vision detection. Through tolerance analysis, this lens has good tolerance and high cost performance.

A novel Mach-Zehnder-interferometer (MZI)-based external cavity diode laser (ECDL) has been proposed and theoretically investigated in this study as a light source for tunable diode laser absorption spectroscopy (TDLAS). The proposed model can modulate the dual-arm optical path difference and the round-trip optical path length of the ECDL external cavity. The results demonstrate that the wavelengths of a certain MZI pass band center and the ECDL external cavity longitudinal mode may track each other over a significant range of voltage variation leading to mode-hop-free performance. Therefore, it can be used in the TDLAS system for gas detection and characterization and is expected to replace the DFB laser.

The contradiction between the limited bandwidth of optical communication and the escalating demand for faster communication speed poses an urgent problem that needs to be addressed. To enhance the transmission bandwidth of optical communication systems and reduce transmission losses, this paper proposes a vertically structured thin film lithium niobate modulator with T-electrodes. Through simulations, the electrode's structural parameters are optimized to minimize microwave losses and increase transmission bandwidth. The results demonstrate that compared to traditional T-electrode modulators, the vertical T-electrode modulator with optimized structure parameters (t=5μm, r=45μm, s=2μm, c=5μm, h=4μm) exhibits an increased modulation bandwidth from 80GHz to 100GHz while reducing microwave loss from 0.554dB/cm/Hz^1/2 to 0.399dB/cm/Hz^1/2. This optimization approach enables broader bandwidths and lower microwave losses in electrodes and presents a novel idea for electrode optimization in modulators.

Terahertz (THz) waves are characterized by low photon energy, sensitivity to polar molecules, and different spectral responses to many biomolecules, which are of great research and application value in the biomedical field. Metamaterials are very sensitive to changes in the dielectric environment of their surfaces and have great potential in the field of sensing. Aiming at the current needs of terahertz-band biosensors whose sensitivity still needs to be improved, we propose an open-ring resonator biosensor with an asymmetric structure of discontinuous dielectric layers. Simulation results show that the refractive index sensitivity of this discontinuous dielectric layer metamaterial biosensor can reach up to 605 GHz/RIU, it exhibits a figure of merit (FOM) value of 24.2 RIU^-1 and a high quality (Q) factor of 79.8, indicating a high Q factor and an ultra-high sensitivity. These characteristics are significant in the design of terahertz ultra-sensitive biosensors, showcasing their valuable application potential.

In this paper, we propose a novel reverse design method using topology optimization for the overall structure of a rotationally symmetric concentric nanoring metalens with a continuously variable height. The optimize objective is to maximize the focal field intensity of a predetermined numerical aperture lens. Compared to nano-pillar array structures and single thickness structures, this approach offers advantages in terms of high design flexibility in phase distribution and superior imaging performance. Leveraging the rotational symmetry inherent in concentric nanorings, a combined methodology of near-field electromagnetic simulations and far-field calculations is employed to verify the two-dimensional optical field characteristics. This method significantly improves computational efficiency compared to full-wave simulations in three-dimensional optical fields, as it reduces simulation time and has lower CPU requirements without compromising accuracy. The simulation results demonstrate that the designed continuous variable height concentric nanoring metalens achieves imaging close to the diffraction limit within the visible light range (400-700 nm).

Laser decoy is one of the commonly used technologies to counter the threat of laser guidance. In order to address the problem of mutual interference when multiple optoelectronic countermeasures systems engage in autonomous decoy interference, this paper proposes a decoy interference laser encryption technology based on pseudo random sequences, which encrypts the copied interference laser encoded pulse sequence twice, Embed identity information into the original encoded pulse to form a new interference laser encoded pulse sequence. When the interference laser is emitted, the own optoelectronic countermeasure system uses the known encrypted sequence to determine the source of the current received laser, thereby solving the problem of self interference and mutual interference, and improving the countermeasure performance. This article uses pseudo random encryption based identification of friend or foe technology to conduct simulation experiments on encryption and decryption of existing guided laser coding methods. The results show that this technology can effectively improve the accuracy and reliability of the laser warning module, and has certain practical value.

Polar codes with large kernel matrix have larger polarization rates than polar codes with the original 2×2 kernel matrix, resulting in better decoding error correction performance for the same block length. In this paper, the piecewise Gaussian approximation (PGA) design method based on polar codes with the 2×2 kernel matrix is extended to polar codes with the 3×3 kernel matrix. A recursive formula for the mean of the log-likelihood ratio of the 3×3 kernel matrix polar codes is derived, and the PGA design algorithm for the 3×3 kernel matrix polar codes is given. The simulation results show that the frame error rate performance of polar codes with the 3×3 kernel matrix constructed by the PGA method is better than that the polar codes constructed by the exact Gaussian approximation (EGA) and the approximate Gaussian approximation (AGA) methods.

OFDM is widely used in wireless communication with the development of the communication technology. Timing synchronization is the important part of the receiver in OFDM system. The performance of timing synchronization may have a great influence on signal reception. So far, many researchers have made a lot of studies on timing synchronization. However, classical algorithms always focus on AWGN channel and Rayleigh channel. However, the synchronization performance of classical ones will sharply decrease in Rician channel. This paper proposes a new timing synchronization algorithm based on Rician channel. Simulation results show that the performance is much better than the classical ones.

Balance homodyne detector plays an important role in quantum information processing, which is used to achieve high precision quantum measurement. In the actual measurement, the symmetry of the two arms of the detector will become worse due to the influence of objective factors such as environment and temperature, which will lead to the performance decline and the wrong quantum information. In this paper, a high precious automatic balanced homodyne detector based on Proportional-Integral-Derivative is proposed, in order to realize the automatic balancing of the detector, which can improve the common mode rejection ratio of 22.4 dB, so as to complete the stable and high precision quantum information processing.

An underwater image enhancement model based on Swin Transformer is designed to solve the problems of color shift, low contrast, detail loss and noise in underwater images caused by absorption and scattering effects during the propagation of light underwater. The model consists of two parts: feature extraction and image reconstruction. First, image features are extracted using a set of improved residual Swin Transformer modules, which allows the model to process images with fewer parameters and establish long-term dependencies between image features. Further, a parallel attention mechanism is introduced into the model to focus on both semantic and spatial information contained in underwater images to improve image enhancement. Experimental results show that compared with existing popular underwater image enhancement methods, the proposed Underwater Image Enhancement Method in this paper has advantages in both the number of network parameters and image enhancement accuracy and has made significant progress in correcting color bias, improving contrast, enhancing details and eliminating noise.

In this paper, a novel approach combining image enhancement and improved Canny-based edge detection for the thermal wave radar imaging (TWRI) was proposed to identify the internal delamination defect in carbon fiber reinforced polymer (CFRP) components. Experimental testing on artificial CFRP delamination defects was carried out. The fused method is used for defect identification. An image enhancement algorithm was firstly used to improve the contrast of the detection image. An adaptive Gaussian filter was subsequently selected to smooth the detection image for suppressing the effect of noise, and a Sobel operator with eight-directional convolution template was utilized instead of the Canny operator to calculate the gradients. A double-threshold concatenation was lastly employed to establish the contour of the defect areas. The traditional image edge detection methods and the combined method were compared and investigated. The experimental results showed that the internal delamination defects can be recognized with better sharpness and edge details by the proposed method based on image enhancement and improved Canny edge detection

Single-frame infrared small target detection plays a crucial role in various applications, such as monitoring critical areas at airports, ensuring flight safety, and surveillance of unmanned aerial vehicles (UAVs). However, detecting small targets in infrared images is challenging task due to factors like imaging distance, image quality, target environment, and diverse target types. To improve the detection of small targets within complex backgrounds, we redesign the backbone network and transform the detection task into an image binary segmentation problem. Firstly, we introduced Ublock module to capture contextual information surrounding small targets. Additionally, we employ deconvolution to enhance the resolution of downsampled images. Furthermore, we introduce multiple top-down and bottom-up pathways for feature fusion. Lastly, we incorporate attention modules after each Ublock module to highlight the small targets more effectively. Experimental results demonstrate the effectiveness of our proposed method in enhancing infrared small target detection.

Infrared small target detection has extensive applications in military reconnaissance, precision guidance, and urban security. To address the issues of limited pixel coverage, feature obliteration caused by the lack of texture and color, as well as poor generalization, we propose a dual dynamic convolution U-Net model: DDCU-Net. DDCU-Net not only incorporates the perception and downsampling of traditional dynamic convolution in the network to enhance target feature extraction capabilities, achieving more accurate shape segmentation, but also introduces a novel dynamic parameter convolutional kernel module at the network's skip connections and cross-layer fusion. This convolutional kernel parameter adapts locally to input instances, yielding a better fit for non-stationary infrared small targets and effectively improving network generalization. Experimental results on the two public datasets demonstrate the effectiveness of DDCU-Net's modules, outperforming other advanced algorithms in terms of detection accuracy and shape segmentation performance.

To solve the problem that the traditional deconvolution beamforming algorithm is too computative when processing wideband random signals, an accelerated method of wideband deconvolution beamforming is proposed. According to the characteristics of the array beam pattern, the wideband random signal is divided into multiple narrow bands with non-equal spacing in frequency dimension, and deconvolution beamforming is performed in each narrow band to reduce the deconvolution computation. By means of boundary extension on the beam power spectrum, the boundary ambiguity problem is solved and the computational cost of deconvolution beamforming is further reduced. The simulation and sea test data verify that the proposed method has higher resolution and processing gain than conventional beamforming. The speed of deconvolution beamforming is more than 50% higher than that of conventional deconvolution beamforming.

Aiming at the problem of low calibration accuracy of the telephoto camera, a telephoto camera calibration method based on a highly robust homography matrix is proposed. In the long-distance and long-focus calibration scenario, this paper expounded and analyzed the reasons for the low calibration accuracy of the telephoto camera, and simulated and compared the influence of optimizing the distortion coefficient on the intrinsic parameters calculation in Zhang's calibration method and staged calibration method , to illustrate the importance of the precision of the homography matrix. To improve the accuracy of the homography matrix, a new optimization loss function under multiple constraints is proposed. Experiments show that the proposed telephoto camera calibration method is reasonable and effective. Compared with the traditional two-dimensional plane calibration method, our method has better calibration accuracy and robustness. When the image is slightly distorted and the noise level = 0.5pixel, the average reprojection error reaches 0.2488 pixel, and the error is reduced by 3.86%.

Aiming at the problems of low detection efficiency and high labor cost for manual aerospace electronic components surface, this paper proposes a surface defect detection method based on machine vision. Firstly, based on the design of the components surface defect detection device, the characteristics of the surface image are analyzed. Then, the image is preprocessed by bilateral filtering algorithm and morphological operation, and the edge is removed while the noise is removed. Finally, the fast adaptive threshold is adopted. Segmentation to achieve defect detection on the components surface. The experimental results show that the detection method has a good detection effect on the surface of the components. The detection rate and false detection rate are 3.2% and3.8%, respectively, which provides a feasible solution for the detection of defects on the surface of aerospace electronic products. Has a certain practical value.

Accuracy of the detection algorithm is very important for the safety of autonomous driving in complex road target detection and recognition. Therefore, this study proposes a novel lightweight YOLOv5-CGα target identification algorithm that focuses on optimizing the algorithm's lightness and accuracy. First, the algorithm adds the CBAM (Convolutional Block Attention Module) to improve the network's capacity to extract target characteristics and detection accuracy; secondly, by replacing the ordinary convolution module with the Ghost Bottleneck module in order to compress the parameters of the network and reduce the overall amount of model computation; finally, the original C-IoU (Complete-Intersection Over Union) model's loss function is changed by the α-IoU (α-Intersection Over Union) loss function, leading the model training process to converge faster. The mAP of the YOLOv5-CGα algorithm for the detection of road targets can reach 97.4%, according to experimental results on the BDD100K dataset. The simulation results show that compared with the performance of YOLOv3 and YOLOv5 algorithms, the YOLOv5-CGα algorithm improves the mAP of detecting road targets by 2.7% and by 1.4%. This algorithm only needs half of the parameters in order to enhance on-road target recognition accuracy, which may balance the algorithm's small weight with the issue of accurate road target detection.

This paper studies the multi-channel digital Optical module based on PLCC packaging, and designs and manufactures a small 4-channel parallel receiving and emitting module. The problem of 10Gbps rate signal transmission on substrate with stamp holes is solved through high-speed Signal integrity simulation design. The hybrid integration of optoelectronic chips is completed using chips-on-board (COB) packaging process, and the simultaneous coupling alignment of four receiving and four emitting chips is realized through inclined fiber array. The optical module can operate normally under the ambient temperature of -55 ℃~+85 ℃, which provides a feasible scheme for high-speed data parallel transmission in very short distance under harsh environment.

Due to the complex body shape and higher aesthetic requirements of modern cars, the requirement for antenna size and radiation performance become increasingly high. The form of radio antennas has gradually shifted from pole shaped antennas to conformal antennas represented by glass antenna. Due to the coupling effect of vehicle metal structure and other on-board electronic and electrical components on antenna performance, quantitative testing and evaluation of vehicle-level radio antennas are urgently needed. This article first analyzes domestic and foreign radio antenna standards, compares environmental requirements for laboratory, and proposes a signal uniformity verification method; Secondly, the testing methods for vehicle-level radio antennas were elaborated in detail, with a focus on the precautions based on the reference antenna testing method; Finally, introduces the evaluation indicators and calculation methods for radio antenna performance. For vehicle-level testing, it is generally the frequency response of antenna gain and antenna gain, and recommended limits were proposed. This article provides a basis and a theoretical foundation for promoting the performance testing and evaluation of vehicle-level radio antennas in the trade.

This paper elaborated the cybersecurity problems brought by new technology to intelligent and connected vehicles, summarized the development status of quantum communication technology as well as the technology standards and regulations of automotive cybersecurity developed by leading automotive industrial counties and international organizations such as UN/WP.29 and ISO/TC22. Discussed the application value of quantum communication technology to empower intelligent and connected vehicle with cybersecurity protection, and analyzed the application advantages and limitations of quantum communication technology and common key technology. The feasibility of the application of quantum communication technology in the field of intelligent and connected vehicle cybersecurity is demonstrated through typical cases, suggestions on the development of intelligent and connected vehicle cybersecurity in the new stage are also given.

In the era of intelligence, data traffic has exploded, and optical modules, as one of the components of optical communication, play a crucial role, which is used in data centers, metropolitan area network optical transmission networks and 5G bearer networks. In optical communication system, high sensitivity optical module is one of the key factors that affect the normal operation of the system. Aiming at the influence of high sensitivity on the performance of optical module, a sensitivity optimization scheme of optical module is proposed. This paper introduces the influence of TIA on the sensitivity of ROSA component of 25G SR SFP28 optical module. Theoretical analysis and experimental comparison are used to analyze PIN-TIA and AGC circuits and overload characteristics, and the problem of interval error of optical module is analyzed by focusing on the wiring length of TIA-PD and testing. The influence of TIA-PD wire length and TIA 𝑉_𝐴𝐺𝐶 parameter on the sensitivity of optical module is realized, and the problem of sensitivity optimization of optical module is solved. The results show that the low light sensitivity of the optical module can be optimized by shortening the length of the TIA-PD line within the allowable range of the chip equipment. If the TIA 𝑉_𝐴𝐺𝐶 parameter is too low, the sensitivity of the optical module in the large optical area will be abnormal.

Massive MIMO technology has been used in wireless communication systems, due to its high spectral efficiency. However, existing baseband processing methods for uplink massive MIMO systems mainly rely on centralized baseband processing architectures, which results in high data volume between base station antennas and baseband processing unit, and leads to high computational complexity in the single baseband processing unit. In response to these problems, some decentralized baseband processing (DBP) architectures have been presented recently. This paper investigates the performance of two kinds of DBP architectures, i.e., the Star architecture with a fusion unit and Daisy-Chain architecture without a fusion unit. For the Star architecture, a first-order approximate message passing based detection method is used in the decentralized-processing units (DUs) to reduce the computational complexity while keeping a comparable detection performance, compared to the expectation propagation based DBP detection method. Furthermore, for the Daisy-Chain architecture, a sequential detection scheme and a local fusion method is utilized to improve the detection performance, compared to the Star architecture. Simulation results show that the proposed methods exhibit desirable tradeoff between detection accuracy and computational complexity.

A trellis decoding scheme based on Massey trellis for polar codes with a ternary kernel is proposed. First, we construct trellises for kernel computations of the successive cancellation (SC) decoder based on the Massey trellis. Then, the scheme calculates the bit-channel transition probability through these trellises, which provides an appropriate framework to take advantage of the distributive law and can effectively reduce the number of operations. Simulation results show that for a G₅^⊗3 polar code of length 243 and rate 0.5, our scheme can save 14.2% of the computational cost with no sacrifice in error performance under the SC decoder.

Spaceborne SAR has a vast scope of operation that is not restricted by national borders or airspace, so it has been extensively utilized in diverse fields. Due to its non-cooperation, it is difficult for the reconnaissance antenna to obtain the main lobe signal for parameter analysis and working mode identification of the SAR signal, especially when the spaceborne SAR imaging area is far from the location of the reconnaissance antenna. The reconnaissance receiving antenna can only receive a weak SAR signal with a low signal-to-noise ratio, resulting in a significant error in signal parameter measurement and making it unable to fulfill the analysis of its intended operation. Aiming to address the issue of significant measurement errors in spaceborne SAR signal bandwidth parameters under low signal-to-noise ratio conditions, a high-precision measurement method for spaceborne SAR signal bandwidth is proposed. This method is based on a channelized waterfall diagram, which involves accumulating the channelized signal twice to create the chart. The chirp signal on the waterfall chart is then extracted using an image processing technique to accurately measure the signal bandwidth. Additionally, scene simulation under low signal-to-noise ratio conditions is conducted. The simulation results show that even when the signal-to-noise ratio is as low as -10dB, the error in signal bandwidth measurement does not exceed 17%, indicating stable measurement accuracy.

Rapid, accurate extraction of rural residential areas is of great significance to rural planning and urbanization. On the basis of the improved YOLOv8 object detection algorithm, this paper puts forward a technical method for accurately extracting rural residential areas from multi-scale remote sensing images. The rapid extraction of rural blocks comes true via improving the retrieval mechanism of YOLOv8 algorithm: First, the feature extraction module based on ECA local crosschannel interaction attention mechanism is designed to deeply dig the detailed features with inconsistent scales in the detection of residential areas. Efficient channel interaction pays more attention to the positive sample feature information in the feature map, and meanwhile, it reduces the complexity of the model. Second, Swish activation function is proposed to avoid gradient disappearance and poor activation effect caused by over-fitting. Third, DIoU loss is introduced to accurately show the real distance error between two predicted residential areas and enhance the performance of multitarget detection. In the end, ablation experiments and comparative experiments are conducted on CBDV1.0 building data set. The experimental results show that this method can extract rural residential areas from multi-scale remote sensing images, which provides support for large-scale remote sensing image mapping of rural residential areas.

In this paper, the reciprocal channel approximation (RCA) method is used to design polar codes with large kernels. Firstly, based on the given large kernels, we obtain the encoder graph. Secondly, the bit-channel SNR parameters are successively updated according to the encoder graph by the RCA. Finally, the frozen bits are selected by the SNR of the last bit-channels. The RCA method can avoid the distortion caused by the kernel matrix being too large. Furthermore, in the calculation process, the closed-form expression of the channel capacity is used to avoid excessive use of transcendental functions, which can greatly reduce the calculation cost. The experimental results show that for the G₃^⊗5 and G₅^⊗3 with lengths of 243 and 125, and bit rates of 1/2, when the channel SNR is more than 3dB and 2.5dB, respectively, polar codes with large kernels designed by our scheme have better frame error rate than the Gaussian approximation (GA) method.

This paper presents a generative adversarial network (GAN)-based algorithm for generating Qin bamboo slips character images, specifically addressing the issues of limited samples and a high occurrence of fragmented characters. To mitigate the interference caused by the image background on the network, a global thresholding binary segmentation method is employed to separate the foreground and background of Qin bamboo slips character images. Additionally, we propose the SwinGAN network model based on the DCGAN architecture. The SwinGAN generator network incorporates a windowed multi-head attention mechanism and a Qin Transformer module that combines convolutional neural networks. To prevent gradient varnish, spectral normalization is applied to the convolutional layers of the discriminator, constraining the weight variations. Furthermore, to ensure stable model training, the Wasserstein distance is adopted as the objective function to measure the difference between the generated data distribution and the real data distribution.

This article proposes a no reference video quality assessment method based on deep learning, aiming to simulate human perception of video quality and evaluate videos. This method evaluates the quality of videos by learning effective feature representations in the spatiotemporal domain. First, in the spatial domain, 2D-CNN is used to extract the spatial quality of video frames. Then, in the temporal domain, Recurrent neural network (RNN) and pyramid feature aggregation (PFA) module are used to model the temporal domain and aggregate the frame level feature quality. The experiment shows that the method proposed in this paper has good performance on the KoNViD-1k and CVD2014 datasets, and also indicates that the method has high generalization ability.

Because of the problem that the large amount of remote sensing data and the difficulty of feature selection lead to inaccurate land classification, we proposed a land classification algorithm based on attention u2net using hyperspectral technology. To solve the problem of a large amount of hyperspectral image data and high dimensionality, we adopted the LDA method for dimensionality reduction. To solve the problem that the traditional deep learning network does not focus enough on key areas, an attention u2net algorithm model is proposed, which uses an attention mechanism to strengthen the network’s learning on key areas to obtain better classification accuracy. We conducted experiments based on the existing three mainstream databases, and the results showed that the algorithm achieved an accuracy of 86.6% on the Indian Pines dataset, 95.2% on the Urban dataset, and 82.7% on the Fanglu dataset. Compared with other deep learning algorithms, the average improvement was 2.5%.

The airborne Photoelectric pod has an infrared image and a visible camera and has been widely used in a variety of fields. To more effectively fuse the infrared and viewable pictures gathered by Photoelectric pod, a latent low-rank representation (LatLRR) and NSCT-based image fusion technique is provided. The visible image is preprocessed using the bicubic interpolation method, and its resolution is unified with that of the infrared image. The initial image is decomposed employing a potential low-rank representation to yield low-rank and sparse components, and the low-rank elements are subsequently further deconstructed into low-frequency and high-frequency coefficients through the NSCT. The low-frequency coefficients are fused using adaptive weighted fusion rules, while the high frequency coefficients are fused using average gradient fusion rules. To enhance the performance of the fused mage, the information is fused using the average gradient of the image and the fused low-rank component is generated by means of the NSCT inverse transformation. The experimental results demonstrate that the algorithm suggested in this study can successfully fuse the infrared and visible images acquired by the Photoelectric pod, that the resulting image has better contrast and overall information, as well as that both subjective and objective evaluation is superior to that of other commonly employed algorithms.

In this research work, an improved underwater image enhancement algorithm is presented that is based on physical modeling techniques. It aims to improve the quality of underwater images by addressing color differences and blur. The proposed algorithm performs better than other advanced and classical underwater image enhancement methods. The method is composed of a two-step process. First, a coarse transfer map is estimated, which optimizes the contrast and minimizes the loss of information during the image mapping process. Dark channel preprocessing and guided filter refinement are used to improve transmission map accuracy. Second, the occluded light is estimated by considering the differential attenuation of the red, green, and blue light underwater, thereby mitigating the color differences caused by the attenuation effects and solving the blur problem based on the image models. In addition, the gray world method is used for color correction, resulting in a deblurred and color corrected underwater image. In particular, it addresses color differences and blurriness in underwater imagery caused by light attenuation at different wavelengths, visible light diffusion, and diffusion from plankton and debris. The ability to significantly improve the quality of underwater images is the significance of this enhanced algorithm. This enhancement gives researchers a more apparent and reliable basis for understanding and analyzing underwater environments and phenomena. It has far-reaching implications for various fields, including marine biology, underwater archaeology, and oceanography, as it improves the accuracy of research and applications in these domains.

Aiming at the characteristics of clustering, multi-scale changes, low contrast, and complex background in remote sensing images, this paper proposes an improved rotation frame algorithm based on yolov5s network, which reduces the traditional horizontal detection frame in remote sensing images by rotating frames. Missing detection of small targets due to aggregation and rotation problems; by using the attention mechanism module CBAMC3 that combines space and channels in the Backbone part to replace the original C3 module, the feature representation ability of targets of different scales in remote sensing images is improved; In the loss function, the Focal Loss function is used to reduce the complex background of the remote sensing image and the certain influence of the imbalance of positive and negative samples, and strengthen the training weight of the positive samples; at the same time, the Copy Paste data enhancement is used to enrich the number of samples on the data set. Enhance the generalization ability of the model.The result on the DOTA dataset reaches map86.4%, which is 3.1% higher than the 83.3% of the original yolov5s, which proves that the model has high detection accuracy in remote sensing detection and has certain reference value for detecting images under remote sensing.

Automatic and accurate polyp segmentation is critically involved in the early diagnosis of colorectal cancer.Despite its importance, achieving accurate polyp segmentation remains a challenging task due to the diverse range of appearances exhibited by the polyps in terms of size, shape, and brightness and the blurred boundary between polyps as well as the blurred boundary that often exists between polyps and their surrounding mucosa.This study proposes an encoder-decoder framework-based depth model for polyp segmentation to overcome these challenges. In order to reweight the encoder features, a Hybrid Spatial-Channel Attention Module (HSCAM) was developed. The segmentation network was induced to place increased emphasis on polyp boundaries while enhancing the key feature channels critical to the segmentation task. Subsequently, a Global-Regional Context Module (GRCM) was designed to extract contextual information applicable to the polyp segmentation of different sizes. Finally, a feature aggregation module was designed to efficiently aggregate features from HSCAM, GRCM, and the previous layer of decoder blocks for better polyp segmentation.The proposed method was rigorously assessed through extensive experiments on three publicly available polyp datasets. The results of the present study showed the superior performance of the proposed method as compared to the other state-of-the-art methods.