Recent years have seen the emergence of novel UAV swarm methodologies being developed for numerous applications within the Department of Defense. Such applications include, but are not limited to, search and rescue missions, intelligence, surveillance, and reconnaissance activities, and rapid disaster relief assessment. Herein, this article investigates an initial implementation of learning UAV swarm behaviors using reinforcement learning (RL). Specifically, we present a study implementing a leader-follower UAV swarm using RL-learned behaviors in a search-and-rescue task. Experiments are performed through simulations on synthetic data, specifically using a cross-platform flight simulator with Unreal Engine virtual environment. Performance is assessed by measuring key objective metrics, such as time to complete the mission, redundant actions, stagnation time, and goal success. This article seeks to provide an increased understanding and assessment of current reinforcement learning strategies being developed for controlling (or at a minimum suggesting) UAV swarm behaviors.
The adoption of neural networks for optical component design has increased rapidly in recent years. In this design framework, the numerical simulation of optical wave propagation and material wave modulation are encoded directly as layers of a neural network. This direct encoding enables the optimization of physical quantities (e.g., the transmissivity values of the diffractive optical elements) by gradient descent and the backpropagation algorithm. For the body of work which uses these networks for simulation and optimization, there is a tendency to treat the training process as identical to traditional deep neural networks. However, to the best of our knowledge, there is yet an explicit evaluation of training parameters to support this intuition. This work aims to help fill this gap by providing an exploration and evaluation of data variety to help accelerate those in the community who wish to use this emerging design framework.The application of neural networks in optical component design has witnessed rapid growth in recent years. This design framework encodes the numerical simulation of optical wave propagation and material wave modulation directly within neural network layers, enabling the optimization of physical quantities, such as transmissivity values of diffractive optical elements, through gradient descent and backpropagation algorithms. Physics-informed neural networks have been employed in designing diffractive deep neural networks, optimizing holograms for near-eye displays and creating multi-objective traditional optics. However, there remains a lack of evaluation for training parameters, and discrete sampling considerations are often overlooked. To address these gaps, this study examines the impact of dataset variety on physics-informed neural networks in optimizing lenses that either satisfy or violate the Nyquist sampling criteria. Results show that increased data variety enhances optimized lens performance across all cases. Optimized lenses demonstrate improved imaging performance by reducing diffraction orders present in aliased analytical lenses. Moreover, we reveal that low data variety leads to overfit lenses that function as selective imagers, providing valuable insights for future lens design and optimization.
Optical metasurfaces enable devices to interact with light in unique ways by modulating phase, polarization, or intensity. A metasurface, composed of individual subwavelength scatterers known as meta-atoms, can be designed to provide unparalleled control of wavefronts for a variety of optical applications, yet the design of such devices is often unintuitive and challenging due to computationally expensive forward simulations and the number of free parameters. To overcome this, there is interest in developing inverse design methods as an alternative to conventional forward design. Inverse design leverages machine learning algorithms to effectively search a problem space, starting from application and resulting in solution parameters. In this work, we adopt an inverse design approach that involves targeted forward simulations of arbitrary meta-atoms. To ensure that the dataset captures all possible shapes and rotations of near field responses with second order accuracy, it is constructed using meta-atoms with varying geometries and corresponding phase shifts, including the effect of nearest neighbors. A custom deep learning system is developed to extract meaningful features from this near field response. The proposed framework provides flexibility to produce an inverse design paradigm for generalized metasurface applications without the need for repeated forward simulations. Additionally, the machine learning model is highly effective in reconstructing electric fields, irrespective of the loss function used.
KEYWORDS: Education and training, Convolution, Data modeling, Deep learning, Performance modeling, Object detection, Neural networks, Visual process modeling, Genetic algorithms, Army
With numerous technologies, seeking to utilize deep learning-based object detection algorithms, there is an increased need for an innovative approach to compare one model to another. Often, models are compared one of two over-arching ways: performance metrics or through statistical measures on the dataset. One common approach for training an object detector for a new problem is to transfer learn a model, often initially trained extensively on the ImageNet dataset; however, why one feature backbone was selected over another is overlooked at times. Additionally, while whether it was trained on ImageNet, COCO, or some other benchmark dataset is noted, it is not necessarily considered by many practitioners outside the deep learning research community seeking to implement a state-of-the-art detector for their specific problem. This article proposes new strategies for comparing deep learning models that are associated with the same task, e.g., object detection.
Vehicle maneuverability is often supported in low-light scenarios through infrared (IR) imagery. However, if the imagery contains little temperature gradient, the raw images are less applicable. In order to maximize image effectiveness, a genetic algorithm (GA) is employed to explore various contrast enhancement operators to determine an optimal sequence of contrast enhancements. We propose a new image quality evaluator that incorporates the performance of a deep learning-based object detector and considers image spatial context through cell-structured configurations. The proposed technique is assessed both qualitatively and quantitatively for the task of maneuverability hazard detection.
Object detection remains an important and ever-present component of computer vision applications. While deep learning has been the focal point for much of the research actively being conducted in this area, there still exists certain applications in which such a sophisticated and complex system is not required. For example, if a very specific object or set of objects are desired to be automatically identified, and these objects' appearances are known a priori, then a much simpler and more straightforward approach known as matched filtering, or template matching, can be a very accurate and powerful tool to employ for object detection. In our previous work, we investigated using machine learning, specifically, the improved Evolution COnstructed features framework, to identify (near-) optimal templates for matched filtering given a specific problem. Herein, we explore how different search algorithms, e.g., genetic algorithm, particle swarm optimization, gravitational search algorithm, can derive not only (near-) optimal templates, but also promote templates that are more efficient. Specifically, given a defined template for a particular object of interest, can these search algorithms identify a subset of information that enables more efficient detection algorithms while minimizing degradation of detection performance. Performance is assessed in the context of algorithm efficiency, accuracy of the object detection algorithm and its associated false alarm rate, and search algorithm performance. Experiments are conducted on handpicked images of commercial aircraft from the xView dataset | one of the largest publicly available datasets of overhead imagery.
One aspect of the well-being of a military unit depends on its ability to reliably detect threats and properly prepare for them. While a given sensor mounted on a ground vehicle can adequately capture threats in some scenarios, its viewpoint can be quite limiting. A potential solution to these limitations is mounting the sensor onto an unmanned aerial vehicle (UAV) to provide a more holistic view of the scene. However, this new perspective creates challenges unique to it. Herein, we investigate the performance of an RGB sensor mounted onto a UAV for object detection and classification to enable advanced situational awareness for a manned/unmanned ground vehicle trailing the UAV. To do this, we perform transfer learning with state-of-the-art deep learning models, e.g., ResNet50, Inception-v3. While object detection with machine learning has been actively researched, even on remotely sensed imagery, most of it has been through the context of scene classification. Therefore, it is worthwhile to explore the implications of this new camera perspective on the performance of object detection. Performance is assessed via route-based cross-validation collected by the U.S. Army ERDC at a test site spanning multiple days.
Object recognition is a critical component in most computer vision applications, specifically image classification tasks. Often, it is desired to design an approach that either learns from the data directly or extracts discriminative features from the imagery that can be used for object classification. Most active research in the field of computer vision is concerned with machine learning at some level, whether it be a completely automated process from start to finish via deep learning strategies, or the extraction of human-derived features from the imagery that is subjected to a machine learning-based classifier. However, there are numerous applications in which a particular known object is of interest. In such a setting where a relatively specific object and scene are known a priori, one can develop an extremely robust automatic target recognition (ATR) system using matched filtering. Herein, we consider the use of machine learning to help identify a near-optimal template for matched filtering for a given problem. Specifically, the improved Evolution Constructed (iECO) framework is employed to learn the discriminative target signature(s) to define the template that leads to improved ATR performance in terms of accuracy and a reduced false alarm rate (FAR). Experiments are conducted on ideal synthetic midwave infrared imagery, and results are reported via receiver operating characteristic curves.
Applications seeking to exploit electromagnetic scattering characteristics of an imaging or detection problem typically require a large number of electromagnetic simulations in order to understand relevant object phenomena. It has been shown in a previous work that deep learning may be used to increase the efficiency of creating such datasets by providing estimations comparable to simulation results. In this work, we further investigate the utility of deep learning for electromagnetic simulation prediction by adding to the existing training and testing dataset while also incorporating additional material properties. Specifically, we explore using artificial neural networks to learn the connection between a generic object and its resulting bistatic radar cross section, thereby removing the need to repeatedly perform timely simulations. While deep learning can be seen as a computationally expensive technique, this cost is only experienced during the training of the system and not subsequently in the acquisition of results. The goal of this work is to further investigate the applicability of deep learning for electromagnetic simulation prediction as well as its potential limitations. Additionally, performance is compared for different data pre-processing techniques focused on data reduction.
Applications seeking to exploit electromagnetic scattering characteristics of an imaging or detection problem typically require a large number of electromagnetic simulations. Because these simulations are often computationally intensive, valuable resources are required to perform the simulations in an efficient and timely manner, which is not always freely available or accessible. In this work, we investigate the utility of deep learning for electromagnetic simulation prediction. Specifically, we explore using artificial neural networks to learn the connection between a generic object and its resulting bistatic radar cross section, thereby removing the need to repeatedly perform timely simulations. Such a system would be trained in an offline setting and consequently enable rapid bistatic radar cross section predictions for new objects in the future. While deep learning can be seen as a computationally expensive technique, this cost is only experienced during the training of the system and not subsequently in the acquisition of results. The goal of this work is to learn the applicability of deep learning for electromagnetic simulation prediction as well as its potential limitations. Several simple objects are investigated and a thorough statistical analysis will be used to assess the performance of our proposed method.
Synthetic aperture radar (SAR) benefits from persistent imaging capabilities that are not reliant on factors such as weather or time of day. One area that may benefit from readily available imaging capabilities is road damage detection and assessment occurring from disasters such as earthquakes, sinkholes, or mudslides. This work investigates the performance of a pre-screener for an automatic detection system used to identify locations and quantify the severity of road damage present in SAR imagery. The proposed pre-screener is comprised of two components: advanced image processing and classification. Image processing is used to condition the data, removing non-pertinent information from the imagery which helps the classifier achieve better performance. Specifically, we utilize shearlets; these are powerful filters that capture anisotropic features with good localization and high directional sensitivity. Classification is achieved through the use of a convolutional neural network, and performance is reported as classification accuracy. Experiments are conducted on satellite SAR imagery. Specifically, we investigate Sentinel-1 imagery containing both damaged and non-damaged roads.
One promising technique for improving tunnel detection is the use of spotlight synthetic aperture radar (SL-SAR) in conjunction with focusing techniques. Still, clutter arises from surface variations while severe attenuation of the target signal occurs due to the dielectric properties of the soil. To combat these ill-effects, this work aims to improve imaging and detection of underground tunnels by examining the feasibility of matched illumination waveform design for tunnel detection applications. The tunnel impulse response is incorporated in an optimum waveform derivation scheme which aims to maximize the signal-to-interference and noise ratio (SINR) at the receiver output. Numerical electromagnetic simulations are used to consider wave propagation in realistic soil scenarios which include uniform and non-uniform moisture profiles. It is demonstrated that by considering matched illumination waveforms for transmission in SL-SAR systems, an improvement in the detection and imaging capabilities is achieved through enhanced SINR.
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.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
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.