Segmentation of Watery Low Land Area using Hyperspectral Imaging Technique: A Comparative Study with PPI, N-FINDR, ATGP, and FIPPI

Segmentation of hyperspectral images can be achieved by utilizing the abundance map. Before generating an abundance map, endmember detection is necessary. The algorithms for endmember detection are Pixel Purity Index (PPI), N-FINDR, Automatic Target Generation Process (ATGP), and Fast Iterative Pixel Purity Index (FIPPI). Based on the spectral signature of the extracted endmembers, the abundance map can be generated using Fully Constrained Least Squares (FCLS), Non-negative Constrained Least Squares (NNLS), and Unconstrained Least Squares (UCLS) algorithms. This paper proposes to explore the possible combinations of endmember detection and abundance map generation algorithms. A comparative study was conducted on the combination of algorithms, and it was discovered that FIPPI and UCLS together take the least amount of time to execute, whereas PPI and FCLS take the most time to execute. Though, the trade-off analysis shows that ATGP and NNLS could be an optimum option for this case study of segmentation of watery low land area.