Taking the coal and rock samples retrieved from the Huainan Xieqiao Mine and the Paner II Mine as the research object, the sample reflectance spectrum curve was collected by a ground spectrometer, and the sample’s oxide content, moisture, ash and volatile content were simultaneously detected to reflect the sample’s reflection. The rate spectral curve and the sample component content are used as independent variables, and the sample type is used as the dependent variable to establish a coal and rock identification model to classify coal and rock. This paper mainly adopts three models, which are principal component analysis combined with support vector machine (PCA-SVM), principal component analysis combined with BP neural network (PCA-BP) model and kernel principal component analysis combined with support vector machine (KPCA-SVM) model. The results show that among the three models based on visible light near-infrared spectroscopy, nuclear principal component analysis combined with support vector machine model has the highest recognition accuracy, the average accuracy of modeling is 95.5%, and the average accuracy of verification is about 90.56%; three based on sample components. In the model, the kernel principal component analysis combined with the support vector machine model has the highest recognition accuracy, the average accuracy of modeling is 98.5%, and the average accuracy of verification is about 95%.