Abstract:

An improved algorithm, which is based on support vector machine, is proposed for synthetic aperture radar (SAR) target and shadow image segmentation. A classification idea is used to perform SAR image segmentation. Training samples sent to support vector machine (SVM) are updated continuously by iterative processing. These iterations are repeated until the convergence, which is determined by checking the relative change of the entropy between two consecutive segmented images. The algorithm is applied to SAR imagery coming from defense advanced research project agency (DARPA) and Sandia Laboratory. Experimental results show that the classifier performance acquired from this algorithm is much better. Besides, it also dramatically reduces the influence of the choice of initial segmentation thresholds on the classifier performance and greatly increases the segmentation quality of SAR target and shadow image.