No abstract

Ground penetrating radar (GPR) target detection and classification is a challenging task. Here, we consider various online dictionary learning (DL) methods to obtain sparse representation (SR) of the GPR data to enhance feature extraction for target classification via support vector machine. Online methods are preferred because traditional batch DL like K-SVD is not scalable to high-dimensional training sets and infeasible for real-time operation. We also develop Drop-Off MINibatch Online Dictionary Learning (DOMINODL) which exploits the fact that a lot of the training data may be correlated. The DOMINODL algorithm iteratively considers elements of the training set in small batches and drops off samples which become less relevant. For the case of abandoned anti-personnel landmines classification, we compare the performance of K-SVD with three online algorithms: classical Online Dictionary Learning, its correlation-based variant and DOMINODL. Our experiments with real data from L-band GPR show that online DL methods reduce learning time by 36-93% and increase mine detection by 4-28% over K-SVD. Our DOMINODL is the fastest and retains similar classification performance as the other two online DL approaches. We use a Kolmogorov-Smirnoff test distance and the Dvoretzky-Kiefer-Wolfowitz inequality for the selection of DL input parameters leading to enhanced classification results. To further compare with state-of-the art classification approaches, we evaluate a convolutional neural network (CNN) classifier which performs worse than the proposed approach. Moreover, when the acquired samples are randomly reduced by 25%, 50% and 75%, sparse decomposition based classification with DL remains robust while the CNN accuracy is drastically compromised.2 Furthermore, online DL 2 methods have been studied more generally in GPR. Only one other previous study has employed DL (K-SVD) using GPR signals [24], although for the application of identifying bedrock features. We employ online DL methods and use the coefficients of the resulting sparse vectors as input to a SVM classifier to distinguish mines from clutter. Our comparison of K-SVD and online DL using real data from L-band GPR shows that online DL algorithms present distinct advantages in speed and low false-alarm rates.2) A new, faster online DL algorithm. We propose a new Drop-Off MINi-batch Online Dictionary Learning (DOMINODL) which processes the training data in mini-batches and avoids unnecessary update of the irrelevant atoms in order to reduce the computational complexity. The intuition for the dropoff step comes from the fact that some training samples are uncorrelated and, therefore, in the interest of processing time, they can be dropped during training without significantly affecting performance.3) Better statistical metrics for improved classification. Contrary to previous studies [24] which determine DL parameters (number of iterations, atoms, etc.) based on bulk statistics such as normalized root-mean-square-error (NRMSE), we consider statistical inference for...

No abstract

This work investigates the performance of a sparse decomposition based approach applied to measured Ground Penetrating Radar (GPR) datasets for landmine recognition. The decomposition of the datasets is achieved via the solution of a constraint-relaxed convex optimization problem known as Basis Pursuit Denoise (BPDN). We demonstrate that it is crucial to appropriately construct a database of known scattering responses from mines and clutter, which will form the so-called dictionary. The robustness and accuracy of the methodology are evaluated against different parameters such as the size of the dictionary, the number and selection of time samples and the regularization parameter (noise estimate). Achieved performances are then assessed using the probability of detection and false alarm rate. As figure of merit for the classification accuracy, we use an additional measure, the so called Sparsity Concentration Index (SCI). For validation purposes, we finally compare the classification performance of the presented strategy with another sparse reconstruction based technique (Orthogonal Matching Pursuit, OMP) and an algorithm based on Support Vector Machines (SVM). The obtained results evidence that the proposed method is not only able to discriminate between targets and clutter, but also to recognize the particular type of mine simulants present in the evaluated surveys

We present adaptive strategies for antenna selection for Direction of Arrival (DoA) estimation of a far-field source using TDM MIMO radar with linear arrays. Our treatment is formulated within a general adaptive sensing framework that uses one-step ahead predictions of the Bayesian MSE using a parametric family of Weiss-Weinstein bounds that depend on previous measurements. We compare in simulations our strategy with adaptive policies that optimize the Bobrovsky-Zakaï bound and the Expected Cramér-Rao bound, and show the performance for different levels of measurement noise.