Research On SAR Target Detection Based On Visual Attention Models

Synthetic Aperture Radar(SAR)has been widely used in many military and civilian fields,since it is not restricted by climate and time.Automatic Target Recognition(ATR)has become an important way for SAR image processing and analysis.As the basis of the whole SAR ATR system,the quality of target detection directly affects that of subsequent target recognition,so it is of great significance to study the target detection method from SAR image.The human visual system is able to pay more attention to objects that are prominent in one aspect of an image,which are called significant objects.Using the human eye's perception of these salient targets,researchers have proposed a number of algorithms for detecting salient targets with excellent performance.Inspired by that,this paper applies the human visual attention mechanism to the target detection of SAR image.We study some existing salient detection models and present modifications,making them applicable to the target detection task from SAR image.The main research work of this paper can be summarized as follows:(1)We study a variety of salient detection models,which is based on the attention mechanism of human vision,summarize and analyze different types of algorithms.Several classical salient detection algorithms are studied in detail,and their characteristics are analyzed.(2)When applied to SAR target detection,the Itti algorithm may suffer two deficiencies: the shape of detected target is not clear,and a large amount of original information may be lost.To solve these problems,we propose an algorithm for SAR image salient target detection based on multi-scale super-pixel.Firstly,it uses SLIC algorithm to segment SAR image into superpixels.Based on the adhesive properties of superpixels to the target edge,the original SAR image is segmented for multiple times with different superpixels sizes.Secondly,in consideration of inherent characteristics of SAR images,several feature maps are calculated on superpixels.Finally,different feature maps are integrated to obtain the final saliency map.(3)Based on a unified framework for salient structure detection by contour-guided search theory,we propose an improved salient target detection algorithm based on contour-guided search theory,for SAR image target detection.Particularly,the algorithm utilizes two parallel information processing channels according to the guided search theory.In the non-selective channel,the contour priori of image is obtained by using the mean ratio operator and the anisotropic gaussian kernel firstly,and then,the standard deviation feature is added to strengthen the target region and suppress the background region to obtain the target prior.In the selective channel,four target features that match SAR image characteristics are selected.The final target posteriori map is obtained by iterative bayesian inference,which is used for the final salient target detection result.(4)Extensive experiments are done to evaluate the performance of the proposed algorithm from both qualitative and quantitative aspects.Experimental results demonstrate its effectiveness and superiority for SAR target detection,compared with some existing algorithms.