Spectral Unmixing for End Member Extraction and Abundance Estimation

A hyperspectral sensor can provide various hyperspectral images in the same area multiple times. Endmember reflectance typically varies as a result of environmental change, and this has a significant impact on the unmixing process. The spectral unmixing technique can extract the material spectral information and proportions from hyperspectral data. However, it is challenging to precisely determine abundance maps in spectral unmixing due to the unavoidable spectrum variation driven by illumination and topographical change, atmospheric impacts, and other factors. The current study highlights the significance of classifying land cover features by unmixing spectral data from remotely observed hyperspectral images. A spectrum library, image-based endmembers, or ground truth data are required to analyze and classify this massive volume of hyperspectral information to assist separation of the mixed pixels and map their spatial distribution. Real datasets and ground truths are used in this research work, and the suggested method is compared with existing benchmark techniques to give a better result. The findings indicate that the proposed approach performs better than the most recent procedures discussed in the literature. In this study, work abundance maps are obtained through linear and nonlinear constrain, non-negative, fully constrained least square methods to formulate the resulting unmixing solution. This problem is then resolved using a descent methodology, which includes revising the end members and abundances results for the related dataset.