A The Study of Hyperspectral Image Classification Based on Support Vector Machine

The remote sensing technique has been widely use on the field of land cover classification. Traditionally, image classification is based on per-pixel which with the spectral of object. This kind of method do not considerate the spatial relationship between neighbor of pixels. Therefore many of study were focus of utilizes the texture information to distinguish between spatial characteristic. It promote classification overall accuracy effectively. But, there have the same deficiency while the texture with different object gives similar each other. Finally, Knowledge-based image interpretation method was adopted to overcome these kinds of drawback. Finding the best classified approach is usually a topic to be discussed on the image process. To testify the validity of SVM, this study chooses the dataset of hyperspectral imagery sensed by AVIRIS, with band selection by minimization of dependent information, added different scales of texture information as the origin information of image for classification. To improve the problem of characterizing different scales of textures, wavelet co-occurrence parameters, mean and standard deviation of different level discrete wavelet transform images are used as texture features. In this paper, the texture features and the gray scale value of image will be adopted as characteristic vector of training samples for SVM algorithm. Finally, several classification schemes based on different methodology are comparatively studied. The effectiveness of including texture measures in the classification is also analyzed.