Wavelet-SVM classifier based on texture features for land cover classification
Ning Zhang,Bingfang Wu,Jianjun Zhu,Yuemin Zhou,Liang Zhu
DOI: https://doi.org/10.1117/12.816101
2008-01-01
Abstract:Texture features are recognized to be a special hint in images, which represent the spatial relations of the gray pixels. Nowadays, the applications of the texture analysis in image classification spread abroad. Combined with wavelet multi-resolution analysis or support vector machine statistical learning theory, texture analysis could improve the quality of classification increasingly. In this paper, we focus on the land cover for the Three Gorges reservoir using remote sensing data SPOT-5, a new classification method, wavelet-SVM classifier based on texture features, is employed for this study. Compare to the traditional maximum likelihood classifier and SVM classifier only use spectrum feature, this method produces more accurate classification results. According to the real environment of the Three Gorges reservoir land cover, a best texture group is selected from several texture features. Decompose the image at different levels, which is one of the main advantage of wavelet, and then compute the texture features in every sub-image, and the next step is eliminating the redundant, every texture features are centralized on the first principal components using principal component analysis. Finally, with the first principal components inputted, we can get the classification result using SVM in every decomposition scale, but what the problem we couldn't overlook is how to select the best SVM parameters. So an iterative rule based on the classification accuracy is induced, the more accuracy, the proper parameters. © 2008 SPIE.