Wavelet texture extraction and image classification of hyperspectral data based on Support Vector Machine

The support vector machine (SVM) was a new machine learning technique developed on the basis of statistical learning theory. It is the most successful realization of statistical learning theory. To testify the validity of SVM, this study chose the data set of hyperspectral images sensed by AVIRIS, with the band selected by Bhattacharya distance. And it added different scales of texture information as the origin information of image for classification. The main difficulty of texture recognition was the lack of effective tools to characterize different scales of textures. To improve the problem, the wavelet co-occurrence parameters, mean, homogeneity, and standard deviation of different level discrete wavelet transform images were used as texture features. In this paper, the texture features combined with PCA band of image were adopted as the characteristic vector of training samples for SVM, and decision tree classification. Finally, traditional classification schemes of maximum likelihood were comparatively studied. The effectiveness of the classification including texture measures was also analyzed. The experimental results showed that SVM method gave the highest correct classification rate within all of these three methodologies while maximum likelihood gave the lowest rate. Adding texture feature information by the proposed approach to images improved classification accuracy for all of SVM, decision tree, and maximum likelihood classification.