Nonlinear Spectral Unmixing of Hyperspectral Imagery Based on Hapke Model and Relevance Vector Regression Algorithm

Abstract

Abstract This paper put forward a nonlinear hyperspectral imagery unmixing method based on Hapke nonlinear spectral mixture model and the relevance vector regression. This paper presents a method of relevance vector regression algorithm for unmixing of hyperspectral images. The spectrum was mixed to train the relevance vector regression algorithm according to Hapke non-linear mixing model. This method made our training process coincident with the physical feature of non-linear spectral mixing process. The experiments of smiulated data and hyperspectral imagery are conducted to validate the methods. The smiulated data and hyperspectral image was used for accuracy assessment. Results show that relevance vector regression algorithm can provide better result as compared with linear spectral unmixing model and support vector regression method. And our method provides better result with relative less relevant vector. The proposed algorithm reduced computational complexity of the spectral unmixing problem.