<p>Hyperspectral imaging, with its ability to capture detailed spectral information, has revolutionized mineral identification and mapping. Conventional methods notably spectral angle mapper (SAM) based on region of interest extraction have been employed for classification but often suffer from accuracy limitations. To address this, we propose an iterative approach that combines the sequential maximum angle convex cone (SMACC) for spectral signatures extraction with the SAM method. It includes five steps: (1) preprocessing, (2) spectral signatures extraction using the SMACC method, (3) spectral matching between the extracted spectral signatures and their corresponding ground truth spectral signatures, (4) spectral angle mapping by introducing the spectral signatures matched to the ground truth spectral library, (5) accuracy assessment. The last four steps are repeated iteratively until stable accuracy and kappa coefficient are achieved. This iterative process has been proposed to automatically identify the best endmembers, eliminating the limitation of relying on just a single choice of endmembers, as was common in traditional similar algorithms. This proposed approach has been implemented in a region of Cuprite, Nevada, which is one of the few publicly accessible areas that has corresponding ground truth data. Our focus was specifically on identifying four key mineral classes: alunite, andradite, sphene, and montmorillonite, due to their widespread distribution throughout the chosen Cuprite area. The quantitative and qualitative evaluations highlight significant improvements in identifying and mapping results. The obtained maps show the high quality of the quantitative evaluation. The qualitative evaluations include the spectral similarity score, the accuracy, kappa coefficient and statistical tests. As the result of each step of the overall proposed approach influences the next step, we noticed that the iterative mechanism allows us to detect the best-extracted endmembers, which improves the identification and mapping results. Thus, the obtained spectral similarity score of alunite and sphene minerals is stable for one extracted endmember, while it enhances progressively with the number of endmembers ‘k’ for montmorillonite and andradite minerals. In terms of overall accuracy, there is a linear increase from 81.79 to 96.87% as the number of endmembers rises, accompanied by a substantial improvement in the kappa coefficient, escalating from 0.65 to 0.92. To further validate the efficacy of our approach, we conducted comparisons with conventional SAM and support vector machine methods. Although our method relies on ground truth spectral signatures and does not address the challenge of mixed pixels due to SMACC’s focus on extracting pure endmembers, the experimental results demonstrate that our proposed method achieves the best performance. This enhancement underscores its potential to advance mineral mapping and exploration efforts significantly.</p>