CLASSIFICATION OF HYPERSPECTRAL IMAGES USING SPECTRAL SPATIAL METHOD

In Remote sensing, hyperspectral images gives detailed information about the surface of the earth and the object present in it. The larger number of bands provides more information about the target. This paper describes a sparse representation method and discontinuity preserving relaxation method (DPR). The Multi-scale adaptive sparse representation (MASR) gives scaling the pixels in different class regions exploits better classification performance. The relaxation uses the local association among the neighboring pixels to correct the distortion in the spatial and spectral domain. The both technique gives massive achievement in the classification of hyperspectral images. This relaxation method is done in two steps. In the first step, noises present in the image are reduced and class separability is improved as a preprocessing method. In the second step, probabilistic relaxation is used as post processing method; discontinuity preserving relaxation technique is developed to achieve the information in the spatial domain that considers the discontinuity existing in the hyperspectral images. Experiment is conducted with AVIRIS sensor over Indian Pines hyperspectral dataset for MASR and relaxation methods and ROSIS sensor over Pavia University hyperspectral dataset for DPR method and performance metrics such as average accuracy, overall accuracy, and kappa coefficients are calculated.