Fusion of Radar and Electro-Optical/Infrared (EO/IR) for Ship Classification and Identification

Current state-of-the-art ISAR based classification tools utilize ISAR image enhancement tools, automatic image selection, adaptive clutter suppression and segmentation to facilitate dimensional feature extraction necessary for classification using a comprehensive database of combatants, non-combatants and selected non-naval vessels. However, its classification performance suffers when classes of vessels have similar physical dimensions to each other. In such cases additional features are needed to improve classification performance. The additional salient features not available in ISAR may be available in EO/IR and LIDAR imagery. In this report, Sigma-3 demonstrates that dramatic classification performance gain can be achieved using a two-prong machine learning approach that simultaneously uses two complementary techniques, deep learning CNN and manifold learning, to exploit the automatic feature and regularities discovery of deep learning to fuse the multiple sensor data and the sparsity representation of the data in manifold learning to fuse the raw sensor data as represented by their highly compressed lower dimensional manifolds.