Experimental study on coal-rock perception based on reflectance spectroscopy

In order to study the perception and recognition of coal and rock using reflectance spectroscopy, 75 carbonaceous shale and bituminous coal samples with similar appearance were collected from the boundary of coal seam and roof in the same fully mechanized coal face. An experimental device for collecting the reflectance spectra of these coal and rock samples was built in the laboratory, which consists of one near-infrared spectrometer, four tungsten halogen lights, one fiber collimator, one Y-type fiber and so on. According to the heights of common seam roofs, near-infrared (1 000-2 500 nm) reflectance spectra of the surfaces of these samples were obtained with the distance of 3 m between sample and fiber collimator. Ash yield of spectral acquisition region in the surface of each sample was then measured. Four methods including first derivative (FD), second derivative (SD), continuum removal (CR) and standardized normal variate (SNV) were employed to preprocess the spectral reflectance curves of these samples after Savitzky-Golay (SG) convolution denoising with 13 points. The correlations between ash yields and preprocessed spectra of 50 out of the 75 samples were analyzed. It was found that the maximum correlation coefficient is 0.777 obtained by CR, and its wavelength point falls at 1 698 nm very approaching to the spectral band of 1 700 nm which is related to the main organic components of coal and rock. Based on the continuum removal spectra at 11 wavelength points in the left and right interval-[1 693 nm, 1 703 nm] of 1 698 nm with the maximum correlation coefficient, ash yields and coal-rock categories of the 50 samples, the models of support vector regression (SVR) of coal-rock ash yields and support vector classification (SVC) of coal-rock categories were established. With the two models of in-situ coal-rock perception and recognition used, the prediction accuracies of the remaining 25 test coal and rock samples were 92% and 96%, respectively, and the total time taken for single sample recognition was both less than 0.1 s. Meanwhile, the root mean square error (RMSE) of predicted surface ash yields of the 25 test coal and rock samples reached 5% and the coefficient of determination was 0.88 by the model of SVR of coal-rock ash yields. © 2019, Editorial Office of Journal of China Coal Society. All right reserved.