| The unmanned or less-humanized of fully-mechanized coal mining working face is becoming extremely important for safety and high efficiency production.As one of key equipment in modern fully-mechanized coal mining working face,the intelligent level of shearer directly affects the safety production and coaling mining efficiency of the entire coal mining working face.The identification of coal-cock is the core technology to realize the intelligence of the shearer,and becomes an urgent technical problem to be solved in the field of coal mining.Based on coal-rock image of fullymechanized coal mining face,this paper considers the quick and accurate recognition of coal-rock distribution as the research target.The methods and technologies focus on coal-rock image data set construction,coal-rock image recognition and semanticsegmentation.The main work in this paper can be summarized as follow:(1)The overall framework of the coal-rock recognition system for fullymechanized mining face is constructed based on the cooperative working process of fully-mechanized mining equipment,combing with the functional requirements of the coal-rock recognition system.The main components and recognition flow of the system are analyzed.(2)The coal-rock cutting experimental platform of the shearer is built,and the coal and rock distribution image after shearer drum cut coal wall is obtained.The image data is expanded by scaling,rotating,reducing,and adding noise to the image.Finally,the coal-rock image classification data set and coal-rock image semantic segmentation data set are established through marking and dividing the coal rock image data set.(3)A CRnet network model for coal-rock image classification in fully-mechanized mining face is designed.The deeply separable convolution and Res2 net modules can reduce the parameters of the model and enhance the performance of the network model,and the network model is optimized based on Dropout,Regularization and batch normalization.The simulation results verify the feasibility and superiority of the CRnet network model proposed in this paper in coal-rock image recognition.(4)A CRSnet network model for semantic segmentation of coal-rock images in fully-mechanized mining face is designed.By decoding-coding structure and skip connection the contradictory relationship between the position information and the semantic information is reduced.The model parameters are reduced by using transposed depth separable convolution,and the segmentation results are optimized based on conditional random field.The simulation results verify the feasibility and superiority of the CRSnet network model proposed in this paper in coal-rock image semantic segmentation.(5)In order to verify the research results of this thesis,the related experiments are carried out based on the coal-rock image of fully-mechanized coal mining face,which is at Sanmenxia Longwangzhuang Coal Mine in Henan Province.The experimental results show that: the CRnet network designed for coal-rock image classification of fully-mechanized mining face can accurately identify coal and rock image,and the CRSnet network model designed for coal-rock image semantic segmentation of fullymechanized mining face can clearly segment coal and rock mixed images.Combining the results of coal-rock image classification and coal-rock image semantic segmentation,the specific distribution of coal and rock in the coal-rock image of fully-mechanized coal mining face can be obtained.In this dissertation,there are 43 figures,19 tables and 103 references. |
