Intelligent Separation Technology Of Coal And Gangue Based On LiDAR Imaging

The 14th Five-Year Plan pointed out that in the coming years,coal will still be the most demanded energy in China,and it will play an irreplaceable role in the development of national industry,economy and living standards.However,coal mining process is often accompanied by a large number of gangue output,the density of the gangue is high and the heat is low.If a large amount of gangue is mixed in coal,transportation costs will rise,and combustion efficiency of coal will be reduced and toxic and harmful gases will be released.Therefore,it is an important procedure to separate coal and waste rock in mine or coal wash plant.Due to the difference of gray and texture between coal and gangue surface,it is possible to use automatic separation method of coal and Gangue Based on image recognition.With the development of imaging equipment and high-performance computer,in the laboratory environment,the image-based coal gangue recognition method has been developed rapidly in equipment and algorithm.However,in a real mine or coal washery,due to the complex use environment,traditional image recognition methods often have technical defects in hardware and software,resulting in the safety or accuracy of recognition cannot meet the production requirements.In view of the above problems,this paper puts forward for the first time the recognition method of coal gangue using LiDAR active imaging,obtains the reflection intensity image of coal gangue from point cloud information,studies the recognition algorithm for the reflection intensity image of coal gangue,compares two traditional texture feature extraction algorithms and an improved neural network recognition algorithm,and finds the improved neural network calculation.The method can better identify the reflection intensity image of coal gangue,and the result is satisfactory.In this paper,firstly,the type selection of LiDAR is completed for the on-site environment of coal washing plant,and the information acquisition system of gangue point cloud is established.Subsequently,10000 pieces of coal and gangue cloud information were collected in the actual production environment of the coal washer,and a series of pretreatments were carried out for the point cloud information,including distance information and reflection intensity information.A method of background denoising based on distance point cloud information is proposed.The background denoising is accomplished from the physical level through the distance difference between conveyor belt and ore.Based on this,the reflected intensity image of gangue without background information can be obtained.Subsequently,normalization,filtering and closing operations are performed on the reflected intensity image to improve the image quality.Then,the reflective intensity image containing multiple gangue pieces is divided into several reflective intensity images containing single gangue pieces by using the region of interest algorithm,and adjusted to a uniform size to obtain a data set containing 10000 reflective intensity images of gangue pieces.Two traditional feature extraction and classification algorithms were designed:the first one is to extract features of Gangue by using feature extraction algorithm of Gray Level Co-occurrence Matrix(GLCM)and classify features with Support Vector Machine(SVM)feature classification algorithm;the second one is to extract features of Gangue by using gradient direction histogram(HOG)combined with local binary mode(LBP)feature special zone algorithm and combining classification of SVM classification algorithm.The algorithm classifies the features.The structure of VGG-prune network based on VGG(Visual Geometry Group Network)convolution neural network is optimized,and the VGG-prune network in this paper is obtained by pruning,and compared with the classical VGG-16 and VGG-19 network architecture by experiment,which proves that the VGG-prune network can greatly improve the recognition speed without affecting the recognition accuracy.At the end of this paper,the model training and testing of the above three algorithms are carried out using image data base of reflection intensity of gangue,and their test results are evaluated by a comprehensive evaluation method.The test results show that the recognition accuracy of VGG-prune model for coal and gangue is 92.13%,the F1-score is 0.926,and the recognition speed is 37.4 pieces per second.The data of VGG-prune model are better than the other two traditional texture feature recognition algorithms.The experimental results prove that LiDAR combined with VGG-prune network can quickly and accurately identify coal and gangue in the environment of coal washery,which proves that LiDAR has scientific research significance and application value in ore identification,and makes up for the defects of existing image gangue identification methods,and also provides a new idea for the application of LiDAR.