Joint Bayesian Hyperspectral Unmixing For Change Detection

Spectral unmixing allows to extract endmembers and estimate their proportions in hyperspectral data. Each observed pixel is considered to be a linear combination of several endmembers spectra. Based on a novel hierarchical Bayesian model, change detection into hyperspectral images is achieved by unmixing. A Gibbs sampler is proposed to overcome the complexity of integrating the resulting posterior distribution. The performance of the proposed Bayesian change detection method is evaluated on real data. It provides binary detection with a precision rate up to 98.90%.