Feature Extraction And Automatic Recognition Of SAR Image

SAR image speckle noise removal, image segmentation and image feature extraction and analysis of SAR automatic target recognition of the important and key processing steps, with the continuous development of science and technology, Algorithm of SAR image processing technology, the more and more quickly and accurately, has been developing rapidly. SAR image target recognition in the military field is applied more and more widely, and this paper is based on SAR target recognition as the background, the main research introduces the speckle noise removal, image segmentation and image feature extraction and analysis of the contents of SAR image.Firstly, introduces the basic principle of SAR imaging, and speckle noise on the forming principle, forming noise SAR image noise and traditional optical image has essential distinction. SAR images can clearly show the target image detail and contour, but the speckle noise of SAR image with a relatively large impact on the formation of SAR image, then introduce several traditional filter to remove noise of SAR image to maintain smooth edge filtering, Lee filtering, Frost filtering, Lee and enhanced gamma MAP filtering and adaptive multiple visual image processing. And the experimental results proved adaptive multiple visual image processing to maintain, in image mean equivalent number of looks and relative standard deviation are other filtering methods do not have the advantage. Next, based on the SAR image segmentation research and introduction, image segmentation plays an important role in feature extraction and analysis of translation, compilation, solution of image. There are many methods of image segmentation, and to ensure the after image segmentation, also can well reserve the detail of the image, the edge and texture information is the main problem, the image segmentation method is more, there are threshold segmentation method, segmentation method, statistical segmentation and region growing method to split and merge method.Introduces the maximum threshold and two-dimensional Otsu method, and the SAR image segmentation arithmetic are analyzed in detail.Feature extraction and analysis of the images at the end of the paper, the linear discriminated analysis (LDA) and its improved projection linear discriminated analysis (L-2DLDA) and right projection linear discriminated (R-2DLDA), the main component analysis method (PCA) and its improved left projection principal component analysis method (L-2DPCA)and projection of principal component analysis method (D-2DPCA).Comparison and improvement of the two algorithms and the corresponding experiment results, experimental results show that the improved L-2DLDA, R-2DLDA and L-2DPCA, D-2DPCA, the covariance matrix of the overall image is obtained, the computation for the original feature vector is more simplified, more high rate of image recognition.