Automatic target detection in forward-looking infrared imagery via probabilistic neural networks

Abstract

This paper presents a technique for automatic detection of the targets in forward-looking infrared (FLIR) imagery. Mathematical morphology is applied for the preliminary selection of possible regions of interest (ROI). An efficient clutter rejecter module based on probabilistic neural network is proposed, which is trained by using both target and background features to ensure excellent classification performance by moving the ROI in several directions with respect to the center of the detected target patch. Experimental results using real-life FLIR imagery confirm the excellent performance of the detector and the effectiveness of the proposed clutter rejecter module.