Based On Strong Robustness SAR Image Target Recognition Method Research

SAR(Synthetic Aperture Radar) image filtering, object segmentation, feature extraction and analysis of matching is the key technology of SAR image target recognition, paper focuses on analyzing and studying the several key technologies of SAR image target recognition.First of all, the paper aimed at resolving unsupervised approach propagate(AP, Affinity Propagation) clustering algorithm that had a problem with a larger amount of computation while processing the larger data volume of SAR image and can’t suppress effectively lots of speckle noise in gray scale segmentation. And this paper proposed an approach to SAR image segmentation combining information entropy’s with AP algorithm. First, calculate information entropy’s matrix of image by sliding window, and then taking information entropy as initial data, cluster segmentation with added weighting coefficient and optimized deviation derivative AP algorithm. Experimental result shown that in comparison with AP algorithm.the running time, clustering precision and accuracy of improved AP algorithm have been greater improved and moreover it had great effect.Secondly, the traditional SURF(speed up robust features) operator on the SAR image target detection of interest points is not ideal because of SAR image texture is rich and there is a lot of noise, which exist that interest point detection adaptability is not strong and a large number of useless feature points emerge, so that the target matching success rate decreased, this paper proposed a new SAR image matching algorithm which integrated CFAR(Constant False-Alarm Rate) with SURF, using mixed Gauss model CFAR for region of target interest detection, and using SURF operator to detect feature extraction, finally, using the improved multilayer elimination method for matching feature points. The simulation experiments show that the method has high matching rate and good robustness, with the variation of scale, rotation and noise.