Spectral imaging technology based on on-chip splitting provides services for aerospace, industrial and consumer electronics applications. Since each application requires a different set and number of spectral bands, the lack of scalable and high-cost customized splitting schemes hinders the wide application of multispectral imaging. Here, we demonstrate a compact, highly freely customizable imaging spectrometer with initial validation for coal and gangue classification and recognition applications. And the results reflect the potential application of this spectral imaging system in coal and gangue classification and identification. A supervised classification method using support vector machines (SVM) was used to recognize coal and gangue, and the evaluation of classification accuracy shows that more than 82% of the pixels can be correctly classified, and this study provides strong support for the visual sensors with complete spectral band combinations to achieve higher accuracy. © 2023 SPIE. All rights reserved.