Experimental Research On Coal And Rock Gangue Classification And Preference Recognition Based On Machine Vision

Coal,rock and gangue identification technology is a key technology for coal gangue sorting in the later stage.The large base,many types and high similarity of gangue lead to low sorting efficiency and poor precision,which cannot meet the needs of my country’s smart mine construction,and become an important problem in underground coal gangue sorting.To solve this problem,this paper focuses on The concept of intelligent sorting of coal,rock and gangue adopts image recognition,deep learning,preference recognition and other methods to build a coal rock gangue sample preference recognition model based on Res Net convolutional neural network,and analyze the recognition accuracy and recognition efficiency of the preference recognition algorithm,build the MYS-200 coal and rock gangue identification test bench,and carry out experimental research on coal and rock gangue sorting.The main research contents and research results of this paper are as follows:1)Putting forward the preference recognition method,expounding the theoretical basis of preference,introducing the preference structure and special preference structure,establishing the preference feature space of Coal Gangue based on the sample feature preference degree,which is mainly characterized by texture feature,gray feature and geometric feature,forms the sample image feature preference degree matrix,putting forward the coal gangue preference matching algorithm,formulates the multi-attribute group decision-making scheme of coal gangue samples,and constructs the coal gangue sample preference recognition model.2)Collect image samples of coal(Fat Coal,Lignite,Coking Coal,Gas Fat Coal,Gas Coal,Anthracite),rock(Granite,Diabase,Sandstone,Limestone,Quartz Sandstone,Quartzite,Anorthosite,Shale)and gangue(lignite gangue,bituminous gangue)with 500 images of each kind,totaling 8000 groups of images as the sample set,build a coal,rock and gangue image data sample library,and collect image data in advanced,Select the optimal pretreatment method.3)Extracting the information of image gray-scale features,texture features and geometric features,analyzing the preference feature parameters of image gray-scale features,texture features and geometric features respectively,dividing the weights of the image according to the recognition of each preference feature parameter,and construct a preference feature space based on gray-scale features,texture features and geometric features.4)A in-depth learning algorithm based on Convolution Neural Network fusion preference recognition model is proposed.8000 groups of sample images are trained and analyzed by using the traditional Res Net convolution neural network algorithm and the improved preference recognition convolution neural network algorithm.The research results show that the coal recognition accuracy was improved from 89% to 98.4%,the rock recognition accuracy was improved from 92% to 98.9%,the gangue recognition accuracy was improved from 86% to 99%,and the average recognition accuracy was improved from 89% to 98.8%,The improvement rate was 9.8%,the recognition time of 1600 groups of images in the test set was reduced from 1050.79 s to 773.28 s,a total of 277.51 s was reduced,and the recognition efficiency was increased to 26.4%,which proves that the preference recognition algorithm improves the recognition accuracy and recognition efficiency.5)The MYS-200 coal and rock gangue identification test bench was built,and 16 samples(20 groups of each sample)were selected for experimental testing,and the single sample test and mixed sample test were carried out respectively.The recognition Res Net convolutional neural network algorithm has higher recognition accuracy and efficiency,which verifies the accuracy and efficiency of the algorithm.Aiming at the technical problems of low recognition accuracy and poor efficiency of coal,rock and gangue sorting,a new algorithm is proposed,which adopts the research strategies of image processing,simulation and experimental verification,and introduces the preference recognition method based on deep learning theory.A preference recognition model for coal and rock gangue samples based on Res Net convolutional neural network is constructed,and a preference recognition algorithm based on Res Net convolutional neural network is proposed.The experimental test results show that the average recognition accuracy of samples using the traditional Res Net convolutional neural network algorithm is 88.44%.,using the improved preference recognition algorithm,the recognition accuracy is increased to 98.43%,the recognition accuracy improvement rate is 9.98%,the recognition time of a single image is reduced from 0.16 s to 0.12 s,and the recognition efficiency is increased by 25%,which solves the problem of underground coal gangue sorting and identification The problems of poor accuracy and low efficiency have a certain role in promoting the intelligent development of underground coal,rock and gangue sorting.The thesis has 115 figures,36 tables and 147 references...