PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.
This PDF file contains the front matter associated with SPIE Proceedings Volume 12547, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Multisensor Fusion, Multitarget Tracking, and Resource Management I
Tracking systems often provide sets of tracks rather than raw detections obtained from sensors. Integrating these track sets into other tracking systems is challenging because the usual sensor models do not apply. In this work we present a method for fusing track data from multiple sensors in a central fusion node. The algorithm exploits the covariance intersection algorithm as a pseudo-Kalman filter which is integrated into a multi-sensor multi-target tracker within a Bayesian paradigm. This makes it possible to (i) integrate the proposed fusion method seamlessly into any existing tracker; (ii) modify multi-target trackers to take a set of tracks as a set of measurements; and (iii) perform gating to enable data association between tracks. The described method is demonstrated in simulations using several target trackers within the Stone Soup tracking framework.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The recent generalized unscented transform (GenUT) is formulated into a recursive Kalman filter framework. The GenUT constrains 2n + 1 sigma points and their weights to match the first four statistical moments of a probability distribution. The GenUT integrates well into the unscented Kalman filter framework, creating what we call the generalized unscented Kalman filter (GUKF). The measurement update equations for the skewness and kurtosis are derived within. Performance of the GUKF is compared to the UKF under two studies: noise described by a Gaussian distribution and noise described by a uniform distribution. The GUKF achieves lower errors in state estimation when the UKF uses the heuristic tuning parameter κ = 3 − n. It is also stated that when the parameter κ is tuned to an optimal value, the UKF performs identically to the GUKF. The advantage here is that GUKF requires no such tuning.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This paper considers an evasion maneuver for low-altitude aircraft (A/C) in the presence of the threat of a guided missile. The missile has boost and sustain phases and its trajectory is estimated by a 7-D motion parameter vector. The maximum likelihood estimator is used to estimate the missile motion parameters based on angle measurements from an A/C-borne passive sensor. Based on the estimated closest point of approach distance between the missile and the aircraft, a warning alert is then given to execute an evasion maneuver. The aircraft can bank away from the missile by a turning maneuver in the horizontal plane. Simulation shows that the aircraft can evade the missile by using the turning maneuver. The survivability probability of the aircraft is evaluated and it can be enhanced by an early maneuver start time and a large acceleration during the maneuver.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Tracking in sensor coordinates has numerous benefits, because the filter bypasses the burden of nonlinear coordinates transformation, if consistency in the Cartesian space is guaranteed. Taking advantage of recent breakthroughs in coordinate conversion methods related to bistatic r-u-v (range and direction sines u,v ), we develop the r-u-v filter to optimize the association gate volume while maintaining Cartesian consistency. The paper focuses on materializing theoretical gains due to r-u-v tracking in a scenario with a glide vehicle subject to aerodynamic drag and gravity in a guided trajectory. Previous r-u-v filters are mixed-coordinate filters because realistic dynamics is difficult to express in r-u-v, but the present work manages to account for drag, gravity and guidance in r-u-v state equations without excessive complexity. Linear measurement model helps reduce gate volume, especially by accentuating the range accuracy of the sensor. In the presence of the “contact lens” phenomenon, gate volume reduction can be furthered by replacing the Gaussian ellipsoid gate which has a lot of empty space with a fuller and better fitted contact-lens-shaped gate region. Comparative simulations show that benefits are dependent on the sampling frequency and can be attributed to the choice of coordinates used by the filter.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The field of estimation theory is concerned with providing a system with the ability to extract relevant information about the environment, resulting in more effective interaction with the system’s surroundings through more well-informed, robust control actions. However, environments often exhibit high degrees of nonlinearity and other unwanted effects, posing a significant problem to popular techniques like the Kalman filter (KF), which yields an optimal only under specific conditions. One of these conditions is that the system and measurement noises are Gaussian, zero-mean with known covariance, a condition often hard to satisfy in practical applications. This research aims to address this issue by proposing a machine learning-based estimation approach capable of dealing with a wider range of noise types without the need for a known covariance. Harnessing the generative capabilities of machine learning techniques, we will demonstrate that the resultant model will prove to be a robust estimation strategy. Experimental simulations are carried out comparing the proposed approach with other conventional approaches on different varieties of functions corrupted by noises of varying distribution types.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Fault detection and identification strategies utilize knowledge of the systems and measurements to accurately and quickly predict faults. These strategies are important to mitigate full system failures, and are particularly important for the safe and reliable operation of aerospace systems. In this paper, a relatively new estimation method called the sliding innovation filter (SIF) is combined with the interacting multiple model (IMM) method. The corresponding method, referred to as the SIF-IMM, is applied on a magnetorheological actuator which was built for experimentation. These types of actuators are similar to hydraulic-based ones, which are commonly found in aerospace systems. The method is shown to accurately identify faults in the system. The results are compared and discussed with other popular nonlinear estimation strategies including the extended and unscented Kalman filters.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Multisensor Fusion, Multitarget Tracking, and Resource Management II
An information filter is one that propagates the inverse of the state error covariance, which is used in the state and parameter estimation process. The term ‘information’ is based on the Cramer-Rao lower bound (CRLB), which states that the mean square error of an estimator cannot be smaller than an amount based on its corresponding likelihood function. The most common information filter (IF) is derived based on the inverse of the Kalman filter (KF) covariance. This paper introduces preliminary work completed on developing the information form of the sliding innovation filter. The SIF is a relatively new type of predictor-corrector estimator based on sliding mode concepts. In this brief paper, the recursive equations used in the sliding innovation information filter (SIIF) are derived and summarized. Preliminary results of application to a target tracking problem are also studied.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Estimating the position of a unmanned ground vehicle (UGV) that is navigating a complex road is a challenging task. Numerous algorithms have been developed to estimate the maneuvering status of the UGV. In this study, a newly developed filtering technique called the sliding innovation filter (SIF) is combined with multiple model technique to improve the estimation accuracy. The SIF uses the measured states as a discontinuous hyperplane to constrain the estimates to stay close to it. By combining the benefits of both methods, the proposed filter minimizes chatter during position estimation when the UGV is maneuvering. The effectiveness of the proposed method is evaluated on a UGV navigating an S-shaped road, and the results are compared to those obtained using the standard SIF.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Diver Detection Sonars (DDS) aim to detect the diver and tracking at the specific distance. At the side of signal processing case, there are bunch of beamforming algorithms to localize the target or diver in our case in the literature. In this paper, some beamforming algorithms are combined and compared via PSNR, time each other. Some algorithms show that the effect of sidelobes and reverberation are clearly decreased. Moreover, detection and tracking algorithms are applied to artificial sonar data created with a specific scenario for this purpose.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Information Fusion Methodologies and Applications I
A modern, contested environment produces exponential amounts of data from a vast array of multimodal sensory inputs for intelligence actively updating our situational awareness (SA). The effective management and interpretation of this digital information for (near) real-time decision processes has obscured resulting in imminent costs. This data’s mathematical structures (e.g., its topology) provides a rich, alternative information space where SA could be transformed. Recent successes in topological data analysis (TDA) for a wide array of applications forecast its target recognition capability derived from a sensing grid’s multimodal data and its aggregates. This research introduces novel artificial intelligence/machine learning (AI/ML) pipeline designs invoking TDA-based feature engineering from acoustic, electro-optical (EO), and infrared (IR) data which produce efficient models with near perfect accuracy, precision, and recall in target recognition capability on a range of small unmanned aerial systems (SUASs), ground vehicles, and dismounts (or ground personnel) involving real world environments.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Advances in Convolutional Neural Networks (CNN) have demonstrated state of the art performance in the tasks of image classification and object detection over the past decade. While significant progress has been made in development of more efficient networks, the computational and memory requirements still exceed practical limits in many applications. Additionally, the pose variability in such applications requires even larger training datasets for the network to generalize to all possible scenarios. The goal of this work is to develop an architecture for fusion of multiple views of a single target to provide robust classification with a lightweight backbone network used across all agents. Motivated by approaches to ensemble learning, we demonstrate that multiple weak learners with computationally efficient networks can combine to enhance classification accuracy. Three methods of fusion are considered: decision fusion, feature fusion, and multi-scale feature fusion. A novel network architecture is developed and implemented for each approach then trained and evaluated using synthetic data. For the feature fusion models, a custom training scheme is developed to minimize classification error while maintaining a common feature extraction backbone across agents. This conforms to a distributed classification use case where each agent has no prior knowledge of its position relative to target. Finally, we discuss the requirements for shared data of each approach in the context of applications with limited communication bandwidth.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
A precise relative localization system is a crucial necessity for a swarm of Unmanned Aerial Vehicles (UAVs), particularly when collaborating on a task. This paper aims to provide an alternative navigation system to enable a swarm of UAVs to conduct autonomous missions in a Global Positioning System (GPS)-denied environment. To achieve this goal, this paper proposes a relative navigation system using an Extended Kalman Filter (EKF) fusing observations from the on-board Inertial Measurement Unit (IMU) with ranging measurements obtained from the on-board ranging sensors. To ensure secure and high data communication rates, the system employs two waveforms and a low-cost beam-switching phased array. This system thus enables drone operations even in GPS-denied environments. We demonstrate the effectiveness of our approach through simulation experiments involving a swarm of six drones, which includes three fixed and three moving drones in a challenging Blue-Angel scenario. The evaluation of the statistical tests on the results of the simulations shows that this method is efficient.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Information Fusion Methodologies and Applications II
Many techniques have been developed for sensor and information fusion, machine and deep learning, as well as data and machine analytics. Currently, many groups are exploring methods for human-machine teaming using saliency and heat maps, explainable and interpretable artificial intelligence, as well as user-defined interfaces. However, there is still a need for standard metrics for test and evaluation of systems utilizing artificial intelligence (AI), such as deep learning (DL), to support the AI principles. In this paper, we explore the elements associated with the opportunities and challenges emerging from designing, testing, and evaluating such future systems. The paper highlights the MAST (multi-attribute scorecard table), and more specifically the MAST criteria ―analysis of alternatives‖ by measuring the risk associated with an evidential DL-based decision. The concept of risk includes the probability of a decision as well as the severity of the choice, from which there is also a need for an uncertainty bound on the decision choice which the paper postulates a risk bound. Notional analysis for a cyber networked system is presented to guide to interactive process for test and evaluation to support the certification of AI systems as to the decision risk for a human-machine system that includes analysis from both the DL method and a user.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Cognitive warfare is not new. Weaker parties in an asymmetric conflict have manipulated information and ideas to convince stronger opponents to not fight (e.g., the Trojan Horse). What is new is the extent to which technologies enable cognitive warfare – resulting in the delegitimization of governments by sowing discord and creating division in order to compel acceptance of political will Information sharing tools enable adversaries to interfere more directly than ever with national political processes as well as citizens minds3. Cognitive warfare is considered a new domain of warfare, along with land, maritime, air, space and cyber (technical). The goal of cognitive warfare attacks is to alter or mislead the thoughts of leaders and operators, of members of entire social or professional classes, of the men and women in an army, or on a larger scale, of an entire population in a given region, country or group of countries and impact territory, influence, service interruptions, transportation, etc. The means could be social cyber, cyber technical, electronic warfare, and broadcast, etc. Senior officers and strategists in the Chinese People’s Liberation Army (PLA) claim that AI, neuroscience, and digital applications (e.g., social media) will be able to influence enemies by affecting human cognition directly, Russia’s Gerasimov doctrine talks of the “battlespace of the mind”, Pocheptsov provides examples including creation of fake events and objects and organizing protest actions in Ukraine. Dr Giordano stated, “the brain is the battlefield of the future”. This paper will highlight current examples of cognitive warfare, touch on enabling technologies and relevant social science principles of influence (cognitive and social), highlight existing analytics, introduce the “House model” which identifies pillars of relevant fields of knowledge as well as operationally relevant aspects related to the pillars as potentially helpful framework for thinking about cognitive warfare and identifying needed research.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Sensor fusion combines data from a suite of sensors into an integrated solution that represents the target environment more accurately than that produced by individual sensors. New developments in Machine Learning (ML) algorithms are leading to increased accuracy, precision, and reliability in sensor fusion performance. However, these increases are accompanied by increases in system costs. Aircraft sensor systems have limited computing, storage, and bandwidth resources, which must balance monetary, computational, and throughput costs, sensor fusion performance, aircraft safety, data security, robustness, and modularity system objectives while meeting strict timing requirements. Performing trade studies of these system objectives should come before incorporating new ML models into the sensor fusion software. A scalable and automated solution is needed to quickly analyze the effects on the system’s objectives of providing additional resources to the new inference models. Given that model-based systems engineering (MBSE) is a focus of the majority of the aerospace industry for designing aircraft mission systems, it follows that leveraging these system models can provide scalability to the system analyses needed. This paper proposes adding empirically derived sensor fusion RNN performance and cost measurement data to machine-readable Model Cards. Furthermore, this paper proposes a scalable and automated sensor fusion system analysis process for ingesting SysML system model information and RNN Model Cards for system analyses. The value of this process is the integration of data analysis and system design that enables rapid enhancements of sensor system development.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Signal/Image Processing and Information Fusion Applications I
As the world progresses further into the digital era, we see a growing utility for combining datasets gathered on different devices and receivers as well as on varying time ranges, for use in machine learning. However, machine learning classification introduces a requirement for standardized data, which in turn hampers the ability to utilize diverse sets of data at a given timestamp. In this paper, we investigate the application of various signal pre-processing techniques (Daubecheis wavelet, discrete cosine and discrete fourier transform among others) for multi-modal, multi-class machine learning. Following the pre-processing, the multi-faceted signals are represented solely by features generated from first order statistics, eigen decomposition, and linear discriminant. Utilizing these generated features, as opposed to the signals themselves, these diverse datasets may now be combined as input to machine learning methods. Furthermore, we apply Fisher’s linear discriminant ratio and Random Forest feature importance metrics for feature ranking and feature space reduction followed by a comparison of the approaches. Our work demonstrates that dissimilar datasets with common classes may be combined using the proposed methods with a classification accuracy ≥ 95%. This paper demonstrates that the feature space may be reduced by approximately 60% with ≤ 5% loss in classification accuracy, and in some cases, a slight increase in classification accuracy.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Radio frequency (RF) communication and radar applications in the low frequency ranges (3 Hz – 300 kHz ELF-LF range and beyond) form an important subclass of RF use cases that have to utilize waveforms that are inherently limited in bandwidth and hence information throughput capability. Thus, the challenge is maximizing the performance of suitable classes of waveforms in terms of throughput, noise and interference suppression, power and practical realizability. In this paper, we summarize the limitations of the traditional approach using sinusoidal waveforms, and briefly describe alternative and modified approaches using OFDM, wavelets, filter bank-based methods and optimized prolate spheroidal waveforms.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Signal/Image Processing and Information Fusion Applications II
The sensors used for nondestructive evaluation (NDE) play a crucial role in ensuring aircraft availability. The current NDE paradigm often relies on mono-modal testing and signal-over-threshold criteria to provide robust defect or damage detection, not characterization. This approach works for go/no-go inspections of critical flaws that respond strongly to specific physical stimuli; for example, the electromagnetic method of eddy current testing (ECT) is sensitive enough to the abrupt change in conductivity of surface-breaking cracks in metals that it can be used exclusively for certain practical safety inspections. Yet, there are cases where this approach proves insufficient. Consider characterization of problematic microtexture regions (MTR) in certain titanium alloys, which exceeds the capabilities of any one NDE technique. In this work, a data fusion-based solution to MTR characterization is explored. The material problem and potential inspection methods are discussed. Registered datasets from these methods are presented and made available to the community.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Microtexture regions (MTR) are collections of grains with similar crystallographic orientation whose presence in aerospace components can be detrimental to component life. As such, a method to detect and characterize MTR is needed. In this work, we develop an algorithm to determine the boundary and average crystallographic orientation of MTR within a specimen from eddy current testing (ECT) data. The algorithm uses an extension of matching component analysis combined with a technique developed for image deblurring. We successfully apply the algorithm to simulated ECT data of a realistic titanium specimen.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
This work examines how a forced-attention technique can be applied to the task of Video Activity Recognition. The Look & Learn system performs early fusion of critical detected areas of attention with the original raw image data for training a system for video activity recognition, specifically the task of Squat “Quality” Detection. Look & Learn is compared to previous work, USquat, and achieved a 98.96% accuracy on average compared to the USquat system which achieved 93.75% accuracy demonstrating the improvement that can be gained by Look & Learn’s forced-attention technique. Look & Learn is deployed in an Android Application for proof of concept and results presented.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Quality assessment of digital images plays an important role in modeling, implementation and optimization of image and video processing applications. One of the most popular methods in image quality assessment (IQA) is feature based IQA techniques. These feature based image quality assessment (IQA) techniques, which consist of feature extraction and feature pooling phases, extracts features from the images in order to generate objective scores. Various hand-crafted features have been used in the feature extraction phase of the feature based IQA methods. In this work, instead of implementing a hand-crafted feature extraction scheme, automatic feature extraction is utilized by using a pre-trained deep neural network (DNN) inference structure. Feature pooling, which provides mapping between the proposed features and the subjective scores, is carried out by utilizing a fully-connected layer at the end of the network architecture. Experimental results show that the proposed technique obtains promising results for the IQA problem by making use of the generalization capability of deep learning architectures.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Deep learning can identify different signals and extract a range of useful features or track a signal source. Semi and selfsupervised learning techniques can be used to teach networks the underlying dynamics of a problem and broaden generalizability. We demonstrate preliminary results on machine learning software capable of identifying the source of a target and extracting key pieces of information to help resolve or identify the source including angle of arrival. A Ushaped convolutional network may be trained to classify signals based on IQ samples according to modulations or other select features while reconstructing the clean signal. Use of semi-supervised learning training schedule including Barlow Twins on the generated latent space was demonstrated on combinations of real and synthetic radiofrequency (RF) signals. These signals were augmented under various common signal obfuscations such as Raleigh fading, reflections, varying noise and background signals. Group structure of the signals may be displayed through latent space visualizations. Classification accuracy on unseen test sets was used as the primary measurement of performance under varying levels of obfuscation. From this base, we attempted to combine this network with directional sensitivity in order to enable beam steering or identifying the source. A similar augmentation route enhanced by similar semi and selfsupervised techniques was deployed to improve tracking accuracy under realistic conditions. Statistical techniques may be used to identify frequency regions of interest during the prototyping of this signal identification network. This Deep network framework may be applied across a variety of domains and regimes for sensing and tracking.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Signal/Image Processing and Information Fusion Applications III
According to studies, today, we have seen an exponential increase in forest fires, on the island, in the United States and other places in the world. Due to climate change and extreme weather conditions, wildfires have started to appear in parts of the world that have never experienced this before. Many solutions are already being used to monitor forest fires such as: GPS, weather balloons, aerial drones, and many others. Since all these approaches use imaging sensors, they must wait until wildfires are large enough to be detected. This situation becomes a problem because forest fires would be too large to be easily controlled. This research proposes a solution that aims to attack this problem by using a microphone in the device located on the ground. Additional sensors will be incorporated, such as a digital thermometer to register humidity and temperature, a gas sensor to detect different types of gases, and long-range communications that would help our device to communicate with a network of other similar devices. Also, Internet of Things (IoT) will be implemented to send live sensor data to a central command. By focusing on the detection of forest fires, we can not only detect their occurrence in a timelier manner, but also the proposed system will have the ability to predict fires by monitoring meteorological data through smart networks.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
In the billions of faces that are shaped by thousands of different cultures and ethnicities, one thing remains universal: the way emotions are expressed. To take the next step in human-machine interactions, a machine must be able to clarify facial emotions. Allowing machines to recognize micro-expressions gives them a deeper dive into a person’s true feelings at an instant which allows designers to create more empathetic machines that will take human emotion into account while making optimal decisions; e.g., these machines will be potentially able to detect dangerous situations, alert caregivers to challenges, and provide appropriate responses. Micro-expressions are involuntary and transient facial expressions capable of revealing genuine emotions. We propose to design and train a set of neural network (NN) models capable of micro-expression recognition in real-time applications. Different NN models are explored and compared in this study to design a hybrid deep learning model by combining a convolutional neural network (CNN), a recurrent neural network (RNN, e.g., long short-term memory [LSTM]), and a vision transformer. The CNN can extract spatial features (of a neighborhood within an image) whereas the LSTM can summarize temporal features. In addition, a transformer with an attention mechanism can capture sparse spatial relations residing an image or between frames in a video clip. The inputs of the model are short facial videos, while the outputs are the micro-expressions gleaned from the videos. The deep learning models are trained and tested with publicly available facial micro-expression datasets to recognize different micro-expressions (e.g., happiness, fear, anger, surprise, disgust, sadness). The results of our proposed models are compared with that of literature-reported methods tested on the same datasets. The proposed hybrid models perform the best.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
We propose a tensor network that can learn to perform multiple tasks by adjusting the factors of each layer. Most of the existing methods for multi-task learning train a single network to extract task-specific features and subsequent prediction. We propose to use a single network with task-specific transformations that can extract task-specific features and perform task inference with small memory overhead. In particular, we transform features using low-rank updates in the convolution kernels. We present experiments on different datasets for multi-task and multi-domain learning and demonstrate that our method achieves state-of-the-art performance with minimal memory overhead compared to existing methods.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The heart is an important organ in the human body. The heart pumps blood throughout the body. This mechanical action is generated by electrical signals which can be measured. Measuring this electrical signal, we can perform a series of diagnostics to examine different functions of the heart. The electrocardiogram is a tool to record this activity with the purpose of examining the condition of the conductive system in terms of the timing of the activity of the cardiac muscle. The activity recorded by the ECG is the net electric activity between different points around the body. Using the ECG and the radial pulse we can find the cardiac rhythm in a given situation. Variations in the parameters of these signals could mean possible malfunctions of the conduction system. Most of these variations are known and have been related to heart diseases. Arrhythmias can also be determined using the ECG. In this paper, we will record the ECG of each person in the group, and we will determine a variety of parameters of the cardiac system, including the cardiac vector, the cardiac rate, and the P-R interval. Using the ECG recordings, we will train a neural network, in an unsupervised way, to learn the different types of signals for different individuals. We will also investigate the possible sources of distortion of the signal as well as the effect of inspiration and expiration on the recording of the ECG. In preliminary data obtained, the ECG signals have frequency averages between 0.38, 0.39 and 0.4 Hertz, for individuals between 20 and 25 years old.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Classification of one-dimensional (1D) data is important for a variety of complex problems. From the finance industry to audio processing to the medical field, there are many industries that utilize 1D data. Machine learning techniques have excelled at solving these classification problems, but there is still room for improvement because the techniques have not been perfected. This paper proposes a novel architecture called Multi-Head Augmented Temporal Transformer (MHATT) for 1D classification of time-series data. Highly modified vision transformers were used to improve performance while keeping the network exceptionally efficient. To showcase its efficacy, the network is applied to heartbeat classification using the MIT-BIH OSCAR dataset. This dataset was ethically-split to ensure a fair and intensive test for networks. The novel architecture is 94.6% more efficient and had a peak accuracy of 91.79%, which was a 13.6% reduction in error over a recent state-of-the-art network. The impressive performance and efficiency of the MHATT architecture can be exploited by edge devices for unmatched performance and flexibility of deployment.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Extraction of spectrum features for target molecules from measured spectra, for purpose of their detection, can be achieved by comparison to template spectra within a spectrum database, which are sufficient approximations of dominant spectral features. This study continues presentation of the concept of using Density Functional Theory (DFT) to calculate template spectra for practical detection of target substances, by comparison with spectra within databases. DFT-calculated spectra are well posed for comparison to measured spectra, as template spectra, to the extent of their scalability to larger space-time scales.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Acoustic signals in complex real-world environments are randomized by processes such as multipath reflections, surface scattering, and volume scattering by turbulence and vegetation. Traditional machine learning classification algorithms require large datasets with substantial variation to mitigate the challenge posed by randomized signals. In this paper, a Bayesian classifier is introduced that incorporates the underlying physics of the scattered signal into a realistic likelihood distribution for the random signal variations. Performance of the Bayesian classifier is compared to machine learning classification algorithms based on two convolutional neural networks (CNN), one for images and one for sound. Given the varying architectures of the three classification algorithms each approach requires a different feature set. The image-based CNN uses a spectrogram as the input, the soundbased CNN uses the normalized waveform, and the Bayesian classifier is developed on features given by the power in the one-third octave bands. For the comparison, we employ the acoustic seismic classification identification data set (ACIDS) and the environmental sound classification data set (ESC-50). To account for the small size of the considered data sets, we use transfer learning for the machine learning classification algorithms. For the ACIDS and ESC-50 data sets, we show that the Bayesian classifier method outperforms the image-based CNN and the sound-based CNN when measured by accuracy on a test set of data. Unlike the physics-based Bayesian classifier, the machine learning algorithms do not take advantage of the underlying physics of the acoustic signal, which impacts the classification accuracy of traditional machine learning approaches.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
The patented DIRECT® process offers a unique combination of resolution, detectability and interpretability. DIRECT® (Dual Infra-Red Effusivity Computed Tomography) provides a way to visualize contrasting subsurface object temperature spreads to within 1/10th to 1/5th degree. A singular degree is more commonly seen by industry. The novel process removes clutter from unseen weather changes. Field tests of shallow and deep object gaps are discussed in detail. Shallow gaps were imaged for anti-tank landmines, roadside stashes and paved bridge decks. Deep object empty spaces were imaged at an ancient drainage channel, a geothermal resource aquifer and a slanted pipeline. DIRECT® detected unseen gaps and empty spaces not detected by aerial photography. Field tests uncovered 15 cm deep sand-covered mines at Yuma, AZ, Proving Grounds. They detected a 15 m deep soil-covered, 1-m diameter, pipeline at Dugway, UT, Proving Grounds. The novel DIRECT® process Proof of Concept, Phase 1 calibrations were successful. They remotely sensed a 10 m deep tunnel at the Israel border. Geo-Temp’s field team located a 1 m deep unknown smuggler’s hidden roadside stash. This was at the Brownsville, TX south-eastern US-Mexican border. A 2-15 m deep descending rock-covered drain was imaged at 3 km above the “Dome of the Rock” platform at Jerusalem, Israel. The deepest object detected was a 6-60 m deep dry soil covered geothermal aquifer at Long Valley, CA.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
Field programmable gate arrays (FPGAs) are increasingly popular due to their customizability, which enables them to be tailored to specific applications, resulting in minimal resource usage that saves energy and space. In this work, we used an FPGA with a Z-board from Xilinx to simulate the application of the sliding innovation filter (SIF) to a robotic arm. SIF is a predictor-corrector filter used for both linear and nonlinear systems to estimate states and/or parameters. It shares similar principles with sliding mode observer and smooth variable structure filter (SVSF) and uses a correction gain derived to satisfy Lyapunov stability, keeping the estimates near the measurements. We tested SIF on a manipulator with two joints (rotational and prismatic), using FPGA to run the simulation while tracking resource utilization. We compared the results with those of SVSF.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
To article proposes an approach to improve the quality of data used in various processes based on machine vision systems. The paper proposes a combined approach to applying the method of multi-criteria processing based on the use of a combined criterion in order to implement an edge detector, smoothing and separation areas of the background / object in the image. The application of the method allows eliminating the noise caused by external factors (such as dust and water suspension on the lens or space). The generated data make it possible to form an adaptive criterion for changing the correction parameters for a non-linear change in color balance in areas of increased detail or selected masks of changes blocks. The proposed algorithms make it possible to increase the visibility of small elements, reduce the noise component, while maintaining the boundaries of objects, increase the accuracy of selecting the boundaries of objects and the visual quality of data. As test data used to evaluate the effectiveness, nature data and expert evaluation results for test images obtained by a machine vision system with a sensor with a resolution of 1024x768 (8-bit, color image, visible range) are used. Images of simple shapes are used as analyzed objects.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.