Saliency-Based Automatic Target Detection in Remote Sensing Images

Automatic target detection in satellite images remains a challenging problem. Previous methods mainly focus on independent detection of multiple targets. In this paper, we propose a simultaneous multi-class target detection approach by using saliency computation. The advantages are twofold. First, saliency map is computed only once for all target types. This saves a large amount of computational time but does not miss any targets. Second, we use small regions, obtained from over-segmentation, to be the elementary unit of detection. This provides shape information to remove most false candidates for the final detection. Experiments show that the targets can be quickly detected and the detection rate is as high as the independent detecting methods.