¹Pal Patel et al. (>10)
Icarus (in Press) Link to Article [https://doi.org/10.1016/j.icarus.2026.117000]
¹Department of Computer Science and Engineering, Institute of Technology, Nirma University, Ahmedabad, Gujarat, 382481, India
Copyright Elsevier

This paper presents a methodology for analyzing hyperspectral data obtained from the imaging infrared spectrometer (IIRS) onboard ISRO (Indian Space Research Organisation)’s Chandrayaan-2 mission. The method uses unsupervised learning and classical spectral unmixing techniques to process hyperspectral data from the Mare Crisium region. The processing starts with the identification and removal of bad spectral bands using auxiliary metadata. This is followed by total variation (TV) based denoising to reduce noise and improve data quality. The denoising performance is measured using the peak signal-to-noise ratio (PSNR), with a minimum value of 36 dB, showing effective noise reduction while preserving important spectral information. The main focus of the methodology is endmember extraction. Vertex component analysis (VCA) is used for this purpose and its performance is compared with five existing methods: N-FINDR, automatic target generation process (ATGP), fast iterative pixel purity index (FIPPI), pixel purity index (PPI), and
-norm based pure pixel identification (TRI-P). Abundance estimation is carried out using fully constrained least squares (FCLS). The results are evaluated using spectral similarity measures such as normalized cross-correlation (NormXCorr), spectral angle mapper (SAM), spectral information divergence (SID), normalized euclidean distance (NED), spectral correlation measure (SCM), and spectral gradient angle (SGA). The VCA method shows the best performance with values of NormXCorr = 0.999, SAM = 0.006, SID = 0.00061, NED = 0.0312, SCM = 0.999, and SGA = 0.066. The results show that the validation of the TV-VCA-FCLS pipeline over Mare Crisium confirms its ability to deliver clear spectral endmembers, geologically meaningful abundance maps, and strong spectral fidelity with high computational efficiency, making it a reliable and practical approach for IIRS hyperspectral data analysis.