Superpixel based spectral classification of hyperspectral images in different spaces

In this paper, superpixel based spectral classification of hyperspectral images are compared using different spaces. In conventional pixel-wise HSI classification systems only use spectral information. Unlike conventional pixel-wised HSI classification, superpixel based HSI classification consider both spectral and spatial information of HSI. Support vector machine (SVM) is used as the classification method. Simple Linear Iterative Clustering (SLIC) superpixel algorithm is used to segment hyperspectral dataset into superpixels. Classification performance of hyperspectral data is compared in RGB space, LAB space, PCA space, Spectral space, and SVM. The classification performances of the methods used are tested for two different sets of data and the classification performance results are compared. It is shown that superpixel based spectral classification in PCA space and LAB space gives better classification accuracy.