Data fusion analysis for maritime automatic Target Recognition with designation confidence metrics

This research encompasses robust data fusion methodology for maritime target feature extraction and combinatorial classifiers (Vector Quantizer, Neural Networks based on both Gaussian Mixture Models and Radial Basis Function), which also includes confidence metrics for designations. Feature extraction techniques act on 2-d Inverse Synthetic Aperture Radar (ISAR) images and novel 1-d range profiles. Designations are made with an ISAR database of over 2500 images and 8 basic classes. The Fusion classifier confusion-matrix results indicate the correct classification probability of 80.1% and perfect joint-classification designation in the rarer instances where 2 similar classes are too hard to distinguish from.