Discrimination Algorithm for False Alarm Reduction in SAR Incoherent Change Detection

Abstract

This paper introduces an additional stage for incoherent change detection algorithms (CDAs) based on multilayer perceptron (MLP). Pixels the CDA initially assigned as detections are re-evaluated by the MLP based on features extracted from the processed images, according to a previously performed training. The tests considered multitemporal synthetic aperture radar (SAR) images from the CARABAS-II system, in a scenario where military vehicles were concealed under vegetation. Preliminary results show that the proposed method can reduce the false alarm rate (FAR) by up to 73% for the same probability of detection.