Research On Pre-perception Recognition Of Coal-rock Interface Based On Semantic Segmentation

Pre accurate identification of coal rock interfaces is a key technology for achieving intelligent mining of coal mining machines in fully mechanized coal mining faces and safe and efficient production underground.In recent years,with the continuous increase of mining depth,the distribution of coal and rock in coal mines has become increasingly complex.Once the coal mining machine cuts hard rocks during the mining process,it will cause the cutting teeth of the coal mining machine to wear or even damage,affecting the cutting efficiency of the coal mining machine.However,with the increasingly harsh mining environment,traditional coal rock interface recognition methods can no longer meet the requirements of precise identification and pre perception of coal rock interfaces.To solve the above problems,this paper studies the distribution and trend of coal seams and rock strata in the coal rock interface and the infrared characterization of coal rock interface under active infrared excitation,and studies a pre perception recognition algorithm of infrared image of coal rock interface based on semantic segmentation.The specific work of this paper is as follows:(1)Analyze the mining environment of coal mine fully mechanized mining face and the influencing factors of coal rock identification based on active infrared thermal excitation,and obtain the influencing factors and boundary conditions of active infrared excitation recognition of coal rock interface.(2)Build an experimental platform based on the analyzed influencing factors and boundary conditions,pour coal rock specimens with random orientations,develop corresponding experimental plans,and conduct experiments to obtain a large amount of infrared image sample data of coal rock interfaces.And by using data augmentation methods to increase the sample dataset,and finally labeling and partitioning the dataset,a dataset for infrared image recognition of coal rock interfaces is constructed.(3)Construct a network model for infrared image recognition of coal rock interfaces,and improve the selected Pyramid Scene Parsing Network(PSPNet)network model.Replacing the original Visual Geometry Group Network(VGG network)and Residual Network(Res Net network)with Mobile Net V2 as the backbone feature extraction network,further improving recognition accuracy and speed by adding the Convolutional Block Attention Module(CBAM module).(4)Apply traditional image segmentation algorithms to infrared image segmentation and recognition of coal rock interfaces,set and train the parameters of the improved network model,and qualitatively analyze the improved network model and comparison model through semantic segmentation evaluation indicators.Finally,conduct random experiments to verify the recognition effect of the improved network model.According to the experimental results,based on the improved PSPNet network model,the intersection to union ratio(Io U)of coal and rock is 98.07% and 98.38%,respectively,and the pixel accuracy(PA)of coal and rock is 98.68% and 99.50%,respectively,which is significantly improved compared to the previous PSPNet network model;The improved network model has a prediction speed of 38.46ms/sheet and occupies 9.12 MB of memory.Compared with the previous PSPNet network model,the prediction speed has increased by30.78% and the memory occupied has decreased by 94.88%.Compared with other network models,the improved network model has significantly better recognition performance.