CNN coal and rock recognition method based on hyperspectral data

Abstract

Aiming at the problem of coal gangue identification in the current fully mechanized mining face and coal washing, this article proposed a convolution neural network (CNN) coal and rock identification method based on hyperspectral data. First, coal and rock spectrum data were collected by a near-infrared spectrometer, and then four methods were used to filter 120 sets of collected data: first-order differential (FD), second-order differential (SD), standard normal variable transformation (SNV), and multi-style smoothing. The coal and rock reflectance spectrum data were pre-processed to enhance the intensity of spectral reflectance and absorption characteristics, as well as effectively remove the spectral curve noise generated by instrument performance and environmental factors. A CNN model was constructed, and its advantages and disadvantages were judged based on the accuracy of the three parameter combinations (i.e., the learning rate, the number of feature extraction layers, and the dropout rate) to generate the best CNN classifier for the hyperspectral data for rock recognition. The experiments show that the recognition accuracy of the one-dimensional CNN model proposed in this paper reaches 94.6%. Verification of the advantages and effectiveness of the method were proposed in this article.