Comparative analysis of endmember extraction methods for lithological mapping using hyperspectral imaging

Abstract

Hyperspectral imaging has become an important tool in remote sensing, delivering vast data across many spectral bands and revolutionising our understanding of Earth’s surface. Lithological mapping, a critical application of hyperspectral imaging, identifies and maps minerals and rocks based on their spectral signatures. This is useful in a variety of sectors, including geology and mineral exploration. However, correct extraction of endmembers, which reflect pure spectral signatures, is critical yet difficult due to noise and spectral variation. This research examines various endmember extraction methods for lithological mapping with hyperspectral images. The methodology of preprocessing, hyperspectral unmixing, subclass categorization, and calculation of relative abundance are followed here. In this paper, several methods, including CyCU-Net, VCA, N-FINDR, PCA, and ICA, are used, and their performances are evaluated. The results show the effectiveness of combined approaches in producing accurate lithological maps aligned with XRD results.