Region-based SAR Image Segmentation And Its Application To SAR Image Classification

Synthetic Aperture Radar (SAR) can operate all day and all weather for its unique imaging mechanism. High resolution SAR image processing becomes hot area of research as the developing of the SAR sensor in recent years. Image segmentation and image classification are key step for the image processing, while the inherent echo noise called speckle makes the SAR image processing so complex. The region based similarity constrain region merging algorithm brings a feasible idea for the image processing of high resolution SAR image. Also the SAR image can be segmented into multi-scale as the variation of the similarity between merged regions. The region based SAR image classification algorithm can reduce the impact of the inherent noise and bring in the information of the texture, structure and size etc. in order to improve the accuracy of the SAR image classification.In this paper, first we make a conclusion about the SAR image segmentation and SAR image classification. Then a SAR region based SAR image segmentation method and a segmentation based SAR image classification method are proposed. The main work of this paper is shown as follow:1. The SAR development is reviewed and summarized, and then the development status of the SAR image segmentation and SAR image classification are introduced in detail which point out that the advantage of the region based segmentation method and the application feasibility on SAR image classification.2. We improve the region merging based segmentation method. First, we use the watershed algorithm to obtain an initial over-segmentation results, the regions are stored in nearest neighbor gragh. Then the neighbor regions will be merged through the global optimum region merging rule in order to ensure that the result is the most exact. Only one region will be merged at each iteration and so the adjacent region list does not need to update wholly. At last, the segmentation results will be output and compared will the common segment methods3. The region based SAR image segmentation method is introduced into PolSAR image segmentation, and it’s applied on SAR and PolSAR image classification. First, the single band SAR image is classified based on the segmentation before with the Support Vector Machine (SVM) classifier. Then the region based SAR image segmentation method is introduced into PolSAR image, and the PolSAR image is classified by SVM based on the segmentation results with the multifeature combination. The classification results are compared with the pixel based methods to identify the effectiveness of the proposed SAR image classification method.