Abstract
The availability of hyperspectral images has significantly facilitated clay minerals identification and mapping. Based on the Hyperion L1R hyperspectral dataset, this research aims to improve previous scientific work related to identifying and mapping clay minerals in the Djebel Meni region (northwestern of Algeria). Firstly, we enhance the dataset quality through pre-processing techniques like the removal of bad bands, radiometric calibration, and atmospheric correction. Secondly, the Spectral Information Divergence (SID) algorithm was employed by introducing endmembers derived with the Sequential Maximum Angle Convex Cone (SMACC) algorithm initially and then using Jet Propulsion Laboratory (JPL) spectral signatures of Illite, Kaolinite, and Montmorillonite. The classification results show a correspondence between areas occupied by endmembers and JPL spectral signatures, which helps in matching endmembers with specific minerals. Finally, we conducted a comparative analysis of the two classifications outcomes against a reference map. This last is generated using SID algorithm, which takes ground truth spectral signatures as input. Our proposed approach using the SID classifier has given an overall accuracy score of 97.13% and 84.17% using endmembers image and JPL library respectively. The Kappa coefficient is respectively, with endmembers image and JPL, 0.93 and 0.57. These results show that the SID classification with the endmembers image is better than the SID classification with the JPL library because extracting the endmembers from the image is more accurate than doing that from the pure minerals analyzed in the laboratory. These promising results suggest that our approach could be extended to the identification of clay minerals in diverse hyperspectral datasets.