The focus of this article is deep learning on small, class imbalanced data sets in support of explosive hazard detection (EHD) and automatic target recognition (ATR). To this end, we explore artificial neural networks that are driven by similarity versus classification or regression. Similarity can be emphasized via network design, e.g., siamese networks, and/or underlying metric, e.g., contrastive or triple loss. The general goal of a similarity neural network (SNN) is discriminative training via focusing on similarity between tuples of like (and unlike) inputs. As such, SNNs have the potential to learn improved solutions on small data; aka do more with less". Herein, we explore different avenues and we show that SNNs are essentially neural feature extractors followed by k-nearest neighbor classification. Instead of experimenting on a government data set that cannot be shared, we instead focus on benchmark community data sets for sake of reproducible research. Preliminary findings are encouraging as they suggest that SNNs have great potential for tasks like EHD and ATR.
The focus of this article is extending classifiers from N classes to N+1 classes without retraining for tasks like explosive hazard detection (EHD) and automatic target recognition (ATR). In recent years, deep learning has become state-of-the-art across domains. However, algorithms like convolutional neural networks (CNNs) suffer from the assumption of a closed-world model. That is, once a model is learned, a new class cannot usually be added without changes in the architecture and retraining. Herein, we put forth a way to extend a number of deep learning algorithms while keeping their features in a locked state; i.e., features are not retrained for the new N+1 class. Different feature transformations, metrics, and classifiers are explored to assess the degree to which a new sample belongs to one of the N classes and a decision rule is used for classification. Whereas this extends a deep learner, it does not tell us if a network with locked features has the potential to be extended. Therefore, we put forth a new method based on visually assessing cluster tendency to assess the degree to which a deep learner can be extended (or not). Lastly, while we are primarily focused on tasks like aerial EHD and ATR, experiments herein are for benchmark community data sets for sake of reproducible research.
The detection of side-attack explosive hazards (SAEHs) is a challenging task especially if the SAEHs are camouflaged. Three-Dimensional Radar is one of the most prominent sensors that has shown a great capacity for detecting concealed SAEHs. This system produces high-resolution volumetric images where each voxel’s intensity represents the radar signal return at a specific point in the three-dimensional space. It has the capability to enhance the signal response from a SAEH nested in camouflage materials and suppress the interference from the surroundings. Nevertheless, processing the radar data in the spatial domain has some limitations in differentiating SAEH from clutter objects. In this paper, we propose the use of the discrete Fourier transform (DFT) to analyze the voxel data and capture the spatial frequency characteristics of the radar signal from SAEHs. Through a machine learning approach, our proposed algorithm is able to identify the frequency signatures of SAEHs and to differentiate them from anomalies caused by the background or clutter. This approach yields a confidence value indicating the likelihood of a SAEH at a particular location. The detection ability of the proposed algorithm is demonstrated by the receiver operating characteristic (ROC) curves generated using a dataset collected from a U.S. Army test site.
Recently, the Stalker system has been developed as a high-resolution three-dimensional radar imaging system for the detection of concealed roadside explosive hazards. This system has shown considerable capability in distinguishing between true targets and false alarms using conventional processing techniques such as RX filtering on 2D projections of the data. In this paper, we develop an extension of these methods for use with 3D radar imagery. We show several different prescreening approaches for automatically marking potential target locations and describe an evaluation program called the Tiger scorer. We tested our approach on data collected at an arid U.S. Army test site.
Three-dimensional point clouds generated by LIDAR offer the potential to build a more complete understanding of the
environment in front of a moving vehicle. In particular, LIDAR data facilitates the development of a non-parametric
ground plane model that can filter target predictions from other sensors into above-ground and below-ground sets. This
allows for improved detection performance when, for example, a system designed to locate above-ground targets
considers only the set of above-ground predictions. In this paper, we apply LIDAR-based ground plane filtering to a
forward looking ground penetrating radar (FLGPR) sensor system and a side looking synthetic aperture acoustic (SAA)
sensor system designed to detect explosive hazards along the side of a road. Additionally, we consider the value of the
visual magnitude of the LIDAR return as a feature for identifying anomalies. The predictions from these sensors are
evaluated independently with and without ground plane filtering and then fused to produce a combined prediction
confidence. Sensor fusion is accomplished by interpolating the confidence scores of each sensor along the ground plane
model to create a combined confidence vector at specified points in the environment. The methods are tested along an
unpaved desert road at an arid U.S. Army test site.