Automated radar image target recognition based on segmented feature extraction using adaptive Length Estimation and Hough lines aiding optimized Neural Network classification

This research deals with the creation of an automatic target recognition system used for maritime vessels, for such applications as border patrol, friendly vs. foe id, etc. The paper encompasses robust maritime target feature extraction from 2-d real Inverse Synthetic Aperture Radar images and projects them onto a classifier friendly 1-d space. The image database in use consists of over 2500 images at 1´ range resolution. The classifiers for target designations are Neural Networks based on Gaussian Mixture Models. The automated target recognition engine performs at a 96% correct joint-classification level and at a 77% correct classification level.