Research on sound recognition of coal mine gas and coal dust explosion based on VGG and CNN

YU Xingchen; PENG Cheng; LIU Yongtao; WANG Yang; CUI Yuchen

doi:10.19956/j.cnki.ncist.2025.06.003

Abstract：Aiming at the problems of insufficient recognition accuracy and generalization degree of coal mine gas and coal dust explosion sounds,a recognition method for coal mine gas and coal dust explosion sounds based on VGG and CNN is proposed.Deploy mine microphones in key monitoring areas of coal mines to cap-ture the operation of coal mine equipment and environmental sounds in real time.Use the VGGish network to process the collected audio signals,extract the sound spectra as sound representations,and then input them into CNN to construct sound recognition models for gas and coal dust explosions.For the sound signal to be tested,the spectrogram is also extracted and input into the trained model for classification.The effectiveness of this method was verified through experiments.Firstly,spectral graph extraction and analysis were carried out on the sounds of gas explosion,coal dust explosion,coal mining machine operation,roadheader operation and ventilator operation,verifying the effectiveness and reliability of the spectral graph extraction method.The spectrogram comparison experiment shows that this spectrogram provides more accurate and richer acoustic fea-tures,increases the discrimination between explosive and non-explosive sounds,and facilitates model training.The experimental results show that the average recognition rate of the algorithm proposed in this paper is 97.87%,the accuracy rate is 94.26%,and the recall rate is 100%,which is significantly better than the existing literature methods,verifying the effectiveness and robustness of the algorithm.The time-consuming test results show that the average training time is 46.4 s,the average recognition time is 1.175 s,and the total average duration is 47.55 s.This algorithm maintains a short processing time under different training set ratios,indicating that the proposed algorithm has excellent recognition efficiency.