CFAR-guided Convolution Neural Network for Large Scale Scene SAR Ship Detection

Abstract

Ship target detection in large scene synthetic aperture radar (SAR) image is a very challenging work. Compared with traditional constant false alarm rate (CFAR) detector, detectors based on convolution neural networks (CNNs) perform better. However, there are still two defects ‐1) Small ship targets make it hard to extract ship features, and 2) Totally abandon traditional methods leads to the increasement of positioning-risk. In order to solve these problems, we propose a SAR ship detection network which combines CFAR and CNN, called CFAR-guided Convolution Neural Network (CG-CNN). CG-CNN realizes the fusion of CFAR and CNN at the original image level and feature level, and enhances the guiding role of CFAR detection for CNN detection. Detection results on Large-Scale SAR Ship Detection Dataset-v1.0 show that CG-CNN has the best detection performance.