Multi-feature fusion identification algorithm for VNIR-SWIR spectra of molybdenite and associated minerals

Rapid and accurate identification of molybdenite is critical for optimizing beneficiation processes but remains challenging due to complex gangue interference and environmental noise. This study proposes an intelligent identification framework based on VNIR-SWIR spectroscopy that integrates novel chemometric algorithms with multi-feature fusion. To address spectral coupling mechanisms, we introduce the Multi-scale Band-coupled Interquartile Range (MC-IQR) algorithm, a preprocessing technique that combines adaptive sliding-window filtering to robustly extract overlapping spectral features. Unlike conventional single-feature methods, our approach employs a dynamically weighted fusion model incorporating Euclidean Distance, Cosine Similarity, and Spectral Information Divergence (SID). This strategy synergistically optimizes multidimensional spectral information-spanning amplitude, geometric shape, and probability distribution-to ensure adaptability in complex mining environments. Statistical validation on an independent test set reveals an Overall Accuracy of 97.8% across all target mineral categories, with an ultra-fast processing speed of <10 ms per sample. An ablation study further confirmed that the full ED-CS-SID fusion configuration outperformed all single-feature and dual-feature variants. Furthermore, blind field tests on jaw-crushed raw ore samples provide additional practical evidence for the model's robustness to surface dust and mixed-mineral interference under field conditions. These results support the potential of the proposed method as a non-destructive and real-time tool for intelligent mineral sorting in complex mining environments.