A Feature Fusion-Based Visual Attention Method for Target Detection in SAR Images

Target detection is the front-end stage in any automatic recognition system for synthetic aperture radar (SAR) imagery (SAR-ATR). The efficiency of the detection directly impacts the succeeding stages in the SAR-ATR processing chain. This paper proposes a target detection method for SAR images based on visual attention mechanism. In the paper, a new texture feature extracting method using Local Walsh Transform (LWT) is employed and a target-saliency map is computed based on fusing the primary visual feature maps. Experiments are tested on two kinds of images with simple or complex background. The experimental results show that the detection time cost by the proposed algorithm is less than traditional visual attention methods and the detection results are visually more accurate.