Research On Coal Calorific Value Detecting Method Based On Hyperspectral Imaging Technology

Reasonable utilization of coal resources is a necessary condition for long-term development.In order to improve the utilization rate of coal and improve economic benefits,coal with different calorific values needs to be combusted in proportion to consume less coal to achieve the target power generation or temperature.This paper adopts hyperspectral imaging technology combined with machine learning method,and combines spatial information and spectral information to establish a coal calorific value detection model to achieve accurate,efficient and non-destructive detection of coal calorific value.Firstly,the effect of different moisture content,particle size grade and density grade of coal on the mean reflectance spectrum was studied.The experimental analysis shows that for the same sample,within a certain range,the average spectral curve height of coal will decrease with the increase of moisture content,decrease with the increase of particle size,and increase with the increase of density.In the follow-up experiments,coal samples with the same particle size and density were used,and the tightness of the packaging was ensured to lay the foundation for the accurate detection of the calorific value of coal.Secondly,the calorific value detection of coal is realized based on the spectral information of characteristic bands.The average spectrum of the region of interest is extracted,processed by four combination methods,and the optimal preprocessing method is selected by comparison.Aiming at the limitations of the traditional BP neural network,the additional momentum method and the adaptive learning rate method are used for improvement,and the initial parameters of the model are selected by the particle swarm algorithm.Competitive adaptive reweighting sampling algorithm and successive projections algorithm are used to select coal's composition and calorific value respectively,and the optimal feature selection method is determined by comparison.Combined with the improved BP neural network,calorific value detection model of ingredient-based and feature-band-based are established respectively.The comparative analysis of the two models proves the feasibility and effectiveness of the coal calorific value detecting method based on the spectral information of characteristic bands.Finally,the calorific value detection of coal is realized by combining the spatial information and spectral information of characteristic bands.Based on the low-rank characteristics of hyperspectral image,a hyperspectral image denoising method based on weighted truncated kernel norm and mixed total variation is designed to remove mixed noise in hyperspectral image.On the basis of calorific value feature selection,the grayscale features of coal calorific value feature band images are extracted,and stacked autoencoders are used to extract depth features.Combined with improved BP neural network,coal calorific value detecting model based on joint spatial-spectral information is established.The model is compared with the ingredient-based calorific value detection model and the feature-band-based calorific value detection model.The analysis results show that the accuracy and stability of the model can be further improved after adding spatial grayscale features.The feasibility and effectiveness of the coal calorific value detecting method based on joint spatial-spectral information are proved.