Research On Remote Sensing Image Segmentation Algorithm Based Support Vector Clustering

Remote sensing image segmentation is not only the key implementation step for understanding remote sensing image but also a comprehensive study of the subject which across the image processing, pattern recognition, computer vision, neural network and other areas of multidisciplinary research. It has broad application prospects and has been universal attention in the recent years. Because of the characteristics of remote sensing image is large size, multi-band, rich and varied content, texture feature rich, multi-scale and so on, The segmentation of remote sensing image is more difficult than the general image segmentation.Support vector clustering is cluster analysis method which has been highly concerned currently. Through carefully studies the traditional cluster analysis methods and support vector machine, Support vector clustering method will be applied to the field of remote sensing image segmentation and implement image segmentation. The main idea of this method is that the sample points are mapped to a high-dimensional feature space through a nonlinear mapping. And in this space the smallest radius of the hyper spherical that surrounded by all the sample points is found. When the spherical surface is mapped back to data space that can be divided into several parts and each part contains an independent data points cluster for Partitioning the data sample points. Firstly, the advantages of the method are able to get a global optimal solution through solving quadratic programming problem. Secondly, the method is able to deal with the clustering of arbitrary shape and the noise is able to be effectively analyzed. The results of the experiments were satisfied, and indicated the validity of the method.High time complexity is the difficult problem of Support Vector Clustering. In order to make this method be further applicable to remote sensing image segmentation. At the basis of the original method an improved support vector clustering method for remote sensing image segmentation is proposed through the introduction of clustering identification based on the proximity graph and automatic parameters selection function. The improved method is not only able to achieve image segmentation, but also largely reduces the time complexity. The method improved the timeliness of large-scale data and achieves the parameters of automatic selection. Experimental results show that the improved method was well performed.