Coal is an important natural resource in our country and plays an important role in the development of industry and the national economy. In the process of underground mining, the traditional manual identification of coal-rock interface to cut coal and rock by the shearer is relatively inefficient, the recognition accuracy is poor, and there are many uncertain factors. In the future underground mining, “unmanned” has gradually become the technological development trend of underground mining. The realization of unmanned mining first needs to accurately and efficiently determine the coal-rock interface, and the coal-rock recognition algorithm will become the “brain” of unmanned equipment. Hyperspectral is an emerging technology that has developed rapidly in recent years and has a wide range of substance identification and classification applications. In this paper, hyperspectral is used as the technical means of coal and rock identification, collecting coal and rock hyperspectral data, and designing algorithms to realize coal and rock identification by extracting the characteristic bands of hyperspectral. Coal and rock identification are based on the difference between coal and rock composition. Coal and rock have different forms of aluminum in elemental components. The aluminum in coal samples is alumina, while the aluminum in rock samples is aluminum hydroxide. The vibration of the crystal lattice of AL-OH causes it to produce a strong absorption peak in the 2 130~2 250 nm band. Alumina does not have a strong absorption peak in this band, so 2 130~2 250 nm is used as the characteristic band design algorithm. Taking the mining area of Huainan area as the research area, sampling was conducted in multiple mining areas to obtain 23 sets of coal samples such as coking coal, gas coal, and lean coal; and 25 sets of rock samples such as floor mudstone, sandstone, and shale were obtained. After grinding the sample, use the FieldSpec4 spectrometer produced by the American ASD company to collect the reflectance spectra of coal and rock samples between 350 and 2 500 nm. After pretreatment, use continuum removal method, first-order differential method, second-order differential method and SCA- The SID model method extracts features from the 2 130~2 250 nm band of coal and rock, trains the extracted feature vectors with random forest and SVM algorithms, and applies the model to the test set for classification. In the end, the performance on the test set is good, the overall recognition rate is high, the recognition of the first-order differential and continuum removal methods is 83.3%, and the Kappa coefficients are 0.66 and 0.68, respectively. The recognition rates of the second-order differential method and SCA-SID model method are both above 90%, and the Kappa coefficient is 0.83. From the model’s time complexity and space complexity, the second-order differentiation method is more efficient and reliable than the SCA-SID model method. These identification methods provide an application reference for the automatic coal and rock identification technology underground in actual engineering.