PROJECT TITLE :

A Data Dependent Multiscale Model for Hyperspectral Unmixing With Spectral Variability

ABSTRACT:

Environmental, lighting, atmospheric, and temporal variables can all contribute to hyperspectral image spectral variability. In the unmixing process, it can cause considerable estimation mistakes to propagate if it occurs. For this problem, expanded linear mixing models have been presented that lead to large scale nonsmooth inverse difficulties. As a result of regularisation procedures utilised to generate meaningful findings, the optimization issue has become even more complex because it now includes interdependencies among numerous solutions. New data dependent multiscale model for hyperspectral unmixing taking into consideration spectra variability presented in this paper. Using a multiscale transform based on superpixels, the new method adds spatial context to abundances in extended linear mixing models. To speed up the procedure, the suggested approach estimates abundance only once at each scale per iteration. Based on simulations with synthetic and actual images, the suggested method is compared to other leading-edge algorithms in terms of accuracy and execution time.