SVM多窗口纹理土地利用信息提取技术_张伐伐教材.pdf

1007-4619 (2012) 01-067-12 Journal of Remote Sensing 遥感学报 Technologies of extracting land utilization information based on SVM method with multi-window texture 1 1 2 2 1 ZHANG Fafa , LI Weizhong , LU Liuye , ZHANG Qingfeng , KANG Le 1. College of Forestry, Northwest AF University, Yangling, 712100, China; 2. College of Resources and Environment, Northwest AF University, Yangling, 712100, China Abstract: In order to overcome the problem of fragmentation of ground objects and low accuracy in the single window texture classification, we present a new method of classification using SVM based on multi-window texture, using the Changjiaoba town of Foping county in Shaanxi Province as the test area. First we established the SVM classification model combined with texture analysis based on texture extraction from SPOT 5 remote sensing image. Then we used the model to classify and analyze the types of land use in the area by comparing it with single window texture classification and single data source (spectrum) SVM classification. The research result showed an overall accuracy for multi-window texture classification of 85.33%, which was 13.11% higher than the single window texture classification and 24.10% than single data source (spectrum) SVM. Therefore, we conclude that the method is effective and could solve the problem of fragmentation of ground objects and low accuracy in the single window texture classification. Key words: support vector machine, texture feature, land use, single window texture, multi-window texture C