Research On Coal Gangue Recognition Based On Convolutional Neural Network Models

Coal,one of the most influential fossil fuels,is usually accompanied by coal gangue in the mining process.The accumulation of coal gangue leads to the waste of land resources and environmental damage,therefore the separation of coal and gangue is an important research direction for the coal industry.Existing coal gangue sorting is done by various techniques such as manual sorting and mechanical sorting,which has serious problems.Manual sorting requires the experience of skilled workers,but the working environment has a significant impact on the health of the workers,while mechanical procedures seriously affect the working environment and production quality.The introduction of computer vision and neural networks has not only ensured the safety of workers and played a huge role in improving the efficiency of the separation system,but they still suffer from significant problems such as visual similarity of gangue,and cannot directly use the original image information to effectively identify gangue.Therefore,this paper combines the perception mechanism to study the gangue identification problem.Based on the fact that each material has different thermal emissions according to its composition,various thermal image based coal and gangue classification models are proposed,and in addition,based on the relationship between the volume and weight of the object,several gangue recognition models based on the perception and prediction of the volume of the gangue are proposed.The main research content of this dissertation is as follows.(1)By studying the thermal radiation effect of coal gangue under certain circumstances,a thermal image based SVM model for coal gangue sorting YCb CrSVM is constructed.firstly,using the principle of infrared radiation,the physical principle model related to coal gangue identification is constructed by comparing the emissivity power and wavelength of black body coal and coal gangue at different temperatures to provide a reference basis for the thermal image based coal gangue classification model.On this basis,the thermal image data acquisition platform is built,and based on the YCb Cr colour space representation,the thermal image Cb colour information is extracted as features,and the Gaussian SVM’s are used to classify the coal and coal gangue,which is compared with the method based on ordinary digital images to prove the superiority of the thermal image based method.(2)Aiming at the problem that the recognition of coal gangue based on SVM model needs a lot of analysis and extraction of thermal image features,which leads to the reduction of model recognition efficiency,the research carries out rapid extraction of thermal image features of coal gangue.According to the advantages of convolutional neural network(CNN)in image processing,we propose a feature extraction method of coal gangue thermal image based on CNN,and construct a coal gangue recognition model CGR-CNN.Through the training and testing of coal gangue thermal image features,the accurate and efficient recognition of coal gangue is realized.(3)Aiming at the problem that the thermal environment of coal gangue recognition model based on thermal image features affects the use of the model,the research carries out general coal gangue recognition method.According to the density difference of coal gangue,we propose a recognition method based on the relationship between sample volume and weight,and construct a coal gangue classification model VW-SVM based on sample volume,and design Ex MW＿SVM,CGVP＿CNN and MCGVP＿CNN three models to predict the sample volume for the test of VW-SVM model to realize the coal gangue recognition based on the relationship between volume and weight.