This PDF file contains the front matter associated with SPIE Proceedings Volume 11730, including the Title Page, Copyright information and Table of Contents

Patch-based image denoising approaches have gained popularity recently. We propose an image denoising approach using subspaces that are fit using an L₁-norm criterion. This new approach is competitive with existing approaches in terms of objective error metrics and visual fidelity, and has the added benefit that it can be implemented in parallel for large-scale applications.

There has been a sharp rise in the amount of data available for analysis in many professional fields in recent years. In the medical sector, this significant increase in data can help detect and confirm underlying symptoms in patients that would otherwise remain undetected. Machine learning techniques have been applied in the medical sector and can help diagnose irregularities when data is provided for the specific area on which the system has been trained. Leveraging the newfound amount of big data and advanced diagnostic techniques, higher dimensional data feature extraction can be better analyzed. The algorithm presented in this paper utilizes a convolutional neural network to categorize electrocardiogram (ECG) data by processing the original data implementing the fast Fourier transform (FFT) and principal component analysis (PCA) to reduce dimensionality while maintaining performance. The paper proposes three intelligent identification algorithms that can be fed into another specialized machine learning system or analyzed using traditional diagnostic procedures.

Although single-frame machine learning image restoration techniques have been shown to be effective, the proposed multi-frame approach takes advantage of both spatial and temporal information to resolve high-resolution and high-dynamic-range images. The proposed algorithm is an extension of the previously proposed algorithm DeblurGAN-C and aims to further improve the capabilities of image restoration in degraded visual environments. The main contributions of the proposed techniques include: 1) Development of an effective framework to generate a multi-frame training dataset typical of degraded visual environments; 2) Adopting a multi-frame image restoration framework that generates a single restored image as the output; 3) Conducting substantial experiments against the generated multi-frame training dataset and demonstrate the effectiveness of the proposed multi-frame image enhancement algorithm.

This paper considers human activity classification for an indoor radar system. Human motions generate nonstationary radar returns which represent Doppler and micro-Doppler signals. The time-frequency (TF) analysis of micro-Doppler signals can discern subtle variations on the motion by precisely revealing velocity components of various moving body parts. We consider radar for activity monitoring using TF-based machine learning approach exploiting both temporal and spatial degrees of freedom. The proposed approach captures different human motion representations more vividly in joint-variable data domains achieved through beamforming at the receiver. The radar data is collected using real time measurements at 77 GHz using four receive antennas, and subsequently micro-Doppler signatures are analyzed through machine learning algorithm for classifications of human walking motions. We present the performance of the proposed multi antenna approach in separating and classifying two closely walking persons moving in opposite directions.

This article presents the initial development of a biomass sensor to monitor the growth of macroalgae (seaweeds) in the Integrated Multi-Trophic Aquaculture (IMTA) at Harbor Branch Oceanographic Institute at Florida Atlantic University. The sensor utilizes a combined optical/acoustic means to quantify the seaweed biomass. Such configuration ensures the sensor providing robust coverage under different ambient conditions and biomass densities. After the biomass sensor’s performance has been validated in the lab environment, we deployed the sensor at the macroalgal seaweed cultivation raceway to quantify the seaweed density. The data processing procedures are documented, and the field test results are presented. Finally, the advantages and disadvantages of this approach and future application of the sensor to drive a machine learning-based prediction biomass model are discussed.

Hobbyist electronics have greatly improved in both quality and capability over the past several years. It is now possible to solve computationally challenging problems with equipment costing less than a few hundred US Dollars. Both of the datalogger concepts presented in this work leverage this improvement by using off-the-shelf technology to replace what could previously only be done with expensive custom hardware. The processors used in these concepts, the Teensy 4.0 Audio Adapter and the CTAG BEAST, were originally designed for musicians who require the ability to manipulate multiple channels of audio simultaneously. This capability, however, also enables the construction of dataloggers capable of recording perfectly synchronized multi-channel audio – a requirement for passive phased sonar arrays. Each datalogger carries the additional benefit of low power consumption, permitting the array to be deployed for several hours before requiring recovery. Future versions of the dataloggers are expected to have mission durations comparable to existing commercial systems. This paper follows the development process of both of these concepts and compares each of their performance. The first concept, the Teensy, consists of two Teensy 4.0 control boards, each sandwiched between two Teensy Audio Adapters. This assembly is capable of recording up to eight channels of audio in near perfect sync. The second concept, the BEAST, consists of a BeagleBone Black single board computer augmented with a CTAG BEAST cape. This system is capable of recording eight channels of perfectly synchronized audio. Both of these systems were tested in the field with a four-element co-prime sonar array. The data is analyzed and the results of the comparison are presented in this work. Finally, some operational recommendations and possibilities for improvement are also discussed.

Notification services are widely used nowadays in many scenarios. For instance, they are useful if a patient needs care or an IT incident needs support. There lacks of a simple, affordable and efficient notification system that can deliver important messages across various channels to a certain number of people. Therefore, we developed a serverless 1-click notification system using the Amazon Web Services (AWS) IoT platform. The system allows a user to notify the designated personnel(s) with a simple click using configurable IoT buttons via Wi-Fi. An AWS cloud server receives the request sent by the buttons, processes the request and sends the notification such as emails, text messages or phone calls to the designated service members based on the preprogramed procedures and database. The system is user friendly and easy to implement since no hardware installation is required. The serverless system and the pay-as-you-go policy make the system very cost effective. The system is reliable as it is supported by the AWS. In this article we will describe a few key cloud functions and services that we adopted in the notification system and discuss two potential use cases in classroom technical support and home-based long-term care settings. We validated the system functionality at Texas Christian University's patient care simulation center.

This work introduces a monte carlo based technique powered with the ability to estimate the individual uncertainty contributions of each model parameter. The proposed technique utilizes the so-called parameter space analysis to identify the importance of influential Degrees of Freedom (DoFs) with respect to the uncertainty quantification problem. Once determined, these DoFs can be used to define and solve a linear system of equations based on linearizing the model of interest to determine the uncertainty contribution of each DoFs in conjunction with the monte carlo based samples.

Structured multidimensional data is often expressed in a tensor format. However, due to the large number of terms, it can be difficult to process, store, interpret, or extract patterns from data in a raw tensor format. To alleviate this, various types of tensor decompositions have been developed to reduce the number of terms used to represent multidimensional data, as well as to reveal underlying structure and relationships among the variables. This article will explore variations on two types of tensor decompositions, the CANDECOMP/PARAFAC and the Tucker decompositions, and perform baseline comparisons of their associated algorithms on common datasets with similar choices of parameters. We perform numerical experiments on synthetic and real-world data to directly compare multiple algorithms to approximate the two tensor decomposition types. We also present results comparing the two decomposition models and their algorithms for image denoising and completion examples.

This paper investigates Multi-Agent Systems (MAS) related data analytics. MAS is used for modeling complex, decentralized, and real-world tasks such as package delivery by Unmanned Aircraft Systems (UAS), environmental monitoring, precision agriculture, security, disaster management, UAS Traffic Management (UTM) among others. Fish farming is one such area, where the deployment of UAS platforms could drastically improve the current labor-intensive and resource-constraint operations. This research addresses the design of mission control and path planning strategies for UASs deployed on the fish farm. The proposed strategy enables periodic monitoring of mission-critical parameters. A control strategy is designed to address the tracking control of the UAS under wind conditions.

In this work, we consider compressed sensing based pooled testing, where k out of n items are defective (with a non-zero state). Each time, a subset of items are mixed together, and a real-valued quantitative measurement is obtained, where the measurement equals a random linear combination of the states of the mixed items. Our objective is to detect the k defective items based on the quantitative measurements. This problem arises in a variety of applications, including viral infection diagnosis, network state inference, etc. We assume that each item can be mixed in a limited number of tests, and the mixing coefficients are drawn independently according to a standard Gaussian distribution. We obtain sufficient conditions on the number of tests required for the exact detection of the defective items using an exhaustive search decoder. Our result indicates that the sample complexity scales in the order of O( k^_ log(nk )), where ^_ is approximately the minimum of the n proportions of tests that include individual items. Our result recovers the optimal sample complexity in compressive sensing when ^_ = 1. The performance of the exhaustive search decoder is evaluated numerically under various assumptions on the mixing constraints, signal to noise ratio, and sparsity level.

Clustering is an unsupervised machine learning technique that serves to extract patterns in unlabeled datasets by grouping their elements based on a similarity measure. A priori knowledge of the number of clusters is needed in most of the clustering techniques, which is both difficult and necessary for an effective and accurate pattern recognition and latent (not directly observable) feature analysis. Recently, graph based Symmetric Non-negative Matrix factorization (SymmNMF) has been demonstrated to perform better than k-means and spectral clustering. Here, we present a consensus clustering based on robust resampling technique which in conjunction with SymmNMF and Proportion of Ambiguous Clustering (PAC) criterion performs a robust graphical clustering and accurate identification of the number of clusters in several non-convex benchmark datasets.

The automatic detection of road network from satellite and aerial images is highly significant in many actual applications, for instance, urban traffic measurement, military emergency response, and vehicle target tracking. Compared with other high-resolution satellite remote sensing images, high-resolution synthetic aperture radar (SAR) has become a popular research perspective for road detection owing to its insensitivity to the atmosphere and sun-illumination. However, the method of road network detection is still lagging due to the strong multiplicative speckle noise and complex background interference, causing the loss and break in the road segment extraction results. Aiming to solve this problem, a three-step road network detection framework is proposed. In the first step, the road segment candidates are extracted by the Fuzzy Local Information C-Means (FLICM) algorithm based on the gray-level co-occurrence matrix(GLCM) with Markov Random Fields (MRF), and it contains an adaptive parameter selection procedure which is presented for adjusting joint clustering parameters. In order to reduce false segments, we perform the local processing which combines the morphological operation, linearity index, and local Hough transform in the second step. Finally, as for the global road segment connection, we propose an improved region growing algorithm which fully considering the rationality of road elements to gain the road network. Compared with the traditional region growing algorithm, the proposed method can effectively promote the improvement of the integrity of the road network detection. Moreover, the performance of the proposed method is evaluated by comparing the results with the ground truth road map and the evaluation index including the completeness, correctness, and quality factor. In experiments, the algorithm has been verified with the SAR images from the different resolutions of the GF-3 satellite SAR image. The results of the various real images demonstrate that the proposed algorithm has improved considerably the adaptability and efficiency of road detection compared with other methods.

Video coding is the process of reducing the huge volume of video data to a small number of bits. High coding efficiency reduces the bandwidth required for video streaming, and the space required to store the video data on electronic devices, while maintaining the fidelity of the decompressed video signal. In recent years, deep learning has been extensively applied in the field of video coding. However, it remains challenging how to explore the intra- and inter-frame correlations in deep learning-based video coding systems to improve the coding efficiency. In this work, we propose a hierarchical motion estimation and compensation network for video compression. The video frames are tagged as intra-frames and inter-frames. While intra-frames are compressed independently, the inter-frames are hierarchically predicted by adjacent frames using a bi-directional motion prediction network, which results in highly sparse and compressible residue. The residue frames are then compressed via separately trained residue coding networks. Experimental results demonstrate that the proposed hierarchical deep video compression network offers significantly higher coding efficiencey and superior visual quality compared to prior arts.

The fusion of multispectral sensor data techniques for sets containing complementary information about the subject of observation leads to the visualization of data into a form more easily interpreted by both humans and algorithms. Many applications of feature-level fusion seek to combine edges and textures, across the bandwidth of the sensory spectrum. Visualization techniques can be skewed by the introduction of corruption and redundancies induced by harmonics. A majority of image fusion techniques rely on intensity hue saturation (IHS) transforms, principal component analysis (PCA), and Gram Schmidt. PCA’s ability to remove the redundancy from a set of correlated data while preserving the variance and its resistance to color distortion lends itself to this application. PCA also has a lower spectral distortion as compared to IHS and has been found to create superior image fusion. The application of neural network control techniques has been shown to more accurately recreate results similar to those found by human inference. Over the years, increased computation power has given rise to the spread of neural networks into roles previously carried out by humans. Select advanced image processing techniques have benefited greatly from their implementation. We propose a novel method of utilizing PCA in conjunction with a neural network to achieve a higher quality of image fusion. Implementation of an autoencoder neural network to fuse this information creates a higher level of data visualization when compared to traditional weighted fusion techniques.

Swallowing dysfunction, or dysphagia, occurs secondary to many underlying etiologies such as stroke and can lead to pneumonia. The upper esophageal sphincter (UES) is a major anatomical landmark that allows the passage of swallowed materials into the esophagus during swallowing. Delayed UES opening or reduced duration of opening can lead to the accumulation of pharyngeal residue, which can increase risk of aspiration. UES opening is observed through the inspection of radiographic exams, known as videofluoroscopy swallow studies (VFSSs), which expose patients to ionizing radiation and depend on subjective clinician interpretations. High resolution cervical auscultation (HRCA) is a non-invasive sensor-based technology that has been recently investigated to depict swallowing physiology. HRCA has been proposed for detecting UES opening duration through a deep learning framework. However, the proposed framework was only validated over swallows from patients. For such an algorithm to be robust, it has to be proven equally reliable for the detection of UES opening duration in swallows from both patients and healthy subjects. In this study, we intend to investigate the robustness of the HRCA-based framework to detect the UES opening in signals collected from a diverse population. The framework showed comparable performance regarding the UES opening detection with an average area under the ROC curve of 95%. The results indicate that the HRCA-based UES opening detection can provide superior performance on swallows from diverse populations which demonstrates the clinical potential of HRCA as a non-invasive swallowing assessment tool.

Acquiring field data for machine learning-based image enhancement techniques can be challenging. These techniques often use a supervised learning architecture that requires pairs of degraded and target images that would be unfeasible to acquire in the field. One alternative approach is to employ simulation models that can accurately capture the unique characteristics of a degraded visual environment (DVE). The fidelity or accuracy of these models determines the effectiveness of the fully trained image enhancement algorithm. This paper explores the benefits and deficits of utilizing simulation software to properly portray underwater LiDAR capture in DVEs. This is accomplished by employing 3DSMAX Studios to generate 3D renderings of underwater targets. The Image Systems Engineering Toolbox for Cameras (ISETCAM) is then used to synthetically generate a training dataset. Subjective and objective metrics are devised to measure the effectiveness of these approaches in training a GAN-based underwater LiDAR image enhancement algorithm.

Recent advances in neuroscience have revealed many principles about neural processing. In particular, many biological systems were found to reconfigure/recruit single neurons to generate multiple kinds of decisions. Such findings have the potential to advance our understanding of the design and optimization process of artificial neural networks. Previous work demonstrated that dense neural networks are needed to shape complex decision surfaces required for AI-level recognition tasks. We investigate the ability to model high dimensional recognition problems using single or several neurons networks that are relatively easier to train. By employing three datasets, we test the use of a population of single neuron networks in performing multi-class recognition tasks. Surprisingly, we find that sparse networks can be as efficient as dense networks in both binary and multi-class tasks. Moreover, single neuron networks demonstrate superior performance in binary classification scheme and competing results when combined for multi-class recognition.