Research On Light Coal Mine Pyrotechnic Detection Algorithm Based On Global Context Information

The environment of coal mine is complex.Once the fire occurs,the flame will explode when it meets the gas in the mine,causing serious consequences.Therefore,the detection of fireworks under the coal mine is particularly important.In view of the long detection time and high false detection rate of traditional coal mine fireworks detection,this paper proposes a lightweight coal mine smoke and fire detection algorithm that integrates global context information.The YOLOv5 s model is selected as the benchmark model of this algorithm,and the deep learning network is used to extract image features to determine whether exist fireworks.The main work of this paper is as follows:(1)Efficient lightweight smoke and fire detection model.Firstly,the improved KMeans++ clustering was used to obtain the anchor frame most suitable for this datasets.Secondly,in order to make up for the problem that the number of parameters and computation caused by the subsequent improvement affected the subsequent deployment of the model on the mobile terminal,this paper used GhostNet network to reconstruct the neck of the YOLOv5 s model,making the model more lightweight.At the same time,the Shuffle Attention(SA)mechanism was added to each detection head position to make the model focus more on small target information and suppress the interference of noise.Finally,the boundary frame regression loss function CIoU is replaced by SIoU to improve the regression accuracy and accelerate the convergence of the model.(2)Fireworks detection model integrating global context information.Firstly,based on the improvement of RFB module,this paper proposes multi-scale receptive field module MRF,which uses the atrous convolution to extract different receptive fields information and integrate them.Secondly,we proposes a feature extraction module BoT3 combining CNN and Transformer.This module integrates the advantages of CNN focusing on local information and Transformer focusing on global information,which can effectively solve the problem of small target detection.The experimental results show that the improved YOLOv5 s can meet the requirements of coal mine pyrotechnic detection with less parameters,less calculation and higher accuracy.