An unmixing algorithm based on the relevance vector machine for hyperspectral imagery

Aiming at the defects of the low unmixing accuracy and the abundance map fuzzy existing in the traditional hyperspectral image unmixing methods, a new hyperspectral imagery unmixing algorithm based on the relevance vector machine(UARVM) is proposed in this paper. The core idea of the proposed UARVM is to improve the one-against-rest relevance vector machine, which changes the multi-classification problem into the multiple binary-classification problem, and then to solve each sample's corresponding attribution class probability value, i. e. the abundance, to complete the unmixing process of the hyperspectral imagery. Theoretical analysis and simulation results show that, compared with the traditional unmixing methods, the UARVM method has better unmixing performance and abundance map effect. ©, 2015, Editorial Board of Journal of HEU. All right reserved.