Abstract:  To detect salient objects in SAR image, an adaptive detection method is proposed based on multi-scale auto-convolution variance (MSAV) saliency. With multi-scale auto-convolution operation in SAR image and by calculating MSAV, a variance saliency map is obtained. An auto-threshold-selecting detector is
constructed and salient object detection from the SAR image is achieved. Experimental results show that, by applying the proposed algorithm to a complex scene, salient objects consistent with human visual sense can be effectively detected.

Key words: target detection, multi-scale auto-convolution, variance saliency, synthetic aperture radar (SAR) image