Preprocessing Methods For Synthetic Aperture Radar Images

SAR is a kind of high resolution imaging radar. To be a kind of new radartechnique, with its special advantage and compensation to the traditional radar system,SAR is more and more important to the areas such as solid geosciences, ecosystemscience, hydrology science and the ocean science etc. The research and application ofthe SAR image is become very important since 90's with the space technique andinformation technical development. SAR information processing is very difficultbecause the information expression method of SAR image is different from that of theoptical image. Not only the speckle noise in the SAR image, but also the differentgeography characteristics could present some special phenomenon in the SAR image.SAR image acquisition techniques obviously precede that of the post-processing. So itis an open problem that how to explain SAR image data quickly and accurately.Therefore image processing and analysis teclmiques for SAR images face a hugechallenge.In our research, the traditional despeckling and edge extraction method togetherwith SAR image segmentation methods are studied and improved. The main workwhich has been done in the thesis is as follows:1) SAR image speckle noise modeling and SAR image despeckling are studied.We developed a despeckling algorithm for SAR images basing on the SWTdenoising method. Experiments are conducted on real SAR images.Comparisons have been done with the conventional LEE filtering, Kuanfiltering, hard wavelet filtering and soft wavelet filtering in quality andquantity manner.2) We developed a new bi-threshold edge detection method based on theimproved Ratio detector and the Canny detector. The algorithm puts thecharacteristics of the speckle noise in SAR image into consideration. Theedge direction of the Canny detector reflect gradient direction of the denoised image. The twelve-directional Ratio detector could guarantee thatthe edge calculated is CFAR. The algorithm prevents the line from breakingand has a good edge detection performance and high accuracy of edgeposition.3) A simple and efficient target segmentation method is proposed which isapplied to the SAR target recognition. We construct a framework of greedyEM algorithm for multivariate t-mixture models. First we use multivariatet-mixture models on SAR images, and then the parameters of t-mixture areestimated and the SAR image is segmented by greedy the EM algorithm.Because the stable feature of the t-mixture model, the image segmentationmethod could combine image features such as gray level and texture into themodeling procedure to get satisfying target segmentation results in SARimages.