Accurate identification of coal and gangue is an important prerequisite for coal and gangue sorting and clean and efficient utilization of coal. In view of the many shortcomings of traditional methods such as low efficiency the need to install radiation isolation and environmental interference a classification model called Slime Mold Algorithm Extreme Learning Machine SMA-ELM was proposed to recognize coal and gangue based on multi-spectral image characteristics and spectral characteristics. A multi-spectral data acquisition system was built to complete the acquisition of spectral images of coal and gangue. The extracted feature vectors were downscaled by LBP and PCA principal component analysis, and input to SMA-ELM classification model Antlion Algorithm Optimized Extreme Learning Machine ALO-ELM classification model and Whale Algorithm Optimized Extreme Learning Machine WOA-ELM classification model for comparison. It focused on the recognition accuracy of coal and gangue under different wavelength responses to screen the best wavelength and the optimized optimal wavelengths were compared by multiple evaluation indexes. The experimental results showed that SMA-ELM had the best classification effect and the 6th band was the optimal band the average recognition accuracy of SMA-ELM in this band was 95. 08% and the recognition F1-Scores of coal and gangue were 96. 47% and 92. 68% respectively with a time of 10. 6 s. The proposed method can achieve the accurate recognition of coal and gangue which has important research significance for the intelligent separation of coal and gangue.