SPECTRAL UNMIXING USING MODIFIED PICTURE FUZZY CLUSTERING METHOD

Abstract

Unsupervised clustering methods are used in spectral unmixing of multispectral images since to estimate the endmembers. But their performance is hampered due to inadequate data and imprecise initialization of endmembers. In this study we propose modified picture fuzzy c-means (PFCM) technique to initialize the endmembers. Experimental assessments carried out on multi-spectral datasets reveal that our approach consistently surpasses current leading techniques. Performance indicators such as the BVI, UI, and IRCM validate its enhanced capabilities in endmember separability, accuracy of abundance estimation, and spatial consistency. These findings underscore the efficacy and adaptability of the proposed method across various multi-spectral imaging contexts. Incorporating the heuristic consistently enhances performance in various datasets. Significantly, a revised version of PFCM that includes our heuristic achieving a BVI of 0.004, UI of 0.999 and IRCM of 0.466 on the multispectral image dataset. Tests performed on both the Iris dataset and a multispectral image reveal enhancements in the BVI, UI, and IRCM metrics, as well as faster convergence rates compared to traditional implementations.