Research Of SAR Image Segmentation Based On Improved Spectral Clustering Algorithm

Synthetic Aperture Radar(SAR)has been widely studied because of its unique imaging mechanism that can operate all day and all weather,but the inherent speckle noise of SAR image makes the segmentation of SAR images have always been a key problem in SAR image processing.The raditional segmentation algorithm which is applicable to ordinary optical image is no longer suitable for SAR image segmentation.In this paper,spectral clustering algorithm based on the theory of spectrum divided was introduced into SAR image segmentation and was verified experimentally.Considering the high computational complexity and the low accuracy of the traditional spectral clustering algorithm,an improved spectral clustering algorithm is proposed to overcome these shortcomings.Then it makes full use of the image texture and spatial adjacency information to attain cosine similarity matrix.In the spectral mapping process,using Nystr?m approximation strategy to estimate approximate similarity matrix and its main eigenvectors.Finally we apply a new algorithm that combines improved K-means and improved particle swarm optimization algorithm to the low-dimensional subspace clustering procedure avoiding the defect that K-means algorithm is sensitive to initial value and easy to fall into local optimum.Experimental results show that the new method has obviously better performance and low computational cost than the traditional spectral clustering algorithm.Speckle noise also showed a good robustness and segmentation effect was significantly improved.