Research On Coal And Gangue Recognition Method And Experiment Based On Improved YOLO Algorithm

In recent years,China has vigorously promoted the green development of the coal industry.Traditional coal preparation technology and coal preparation theory are facing severe challenges,and more intelligent,environmentally friendly and efficient coal gangue separation technology is urgently needed.The convolutional neural network technology realizes the automatic extraction of high-dimensional features of coal gangue targets,so that the gangue selection robot can adapt to the complex environment and the diversity of coal gangue.The coal gangue identification and positioning method based on deep learning can output coal gangue target detection results in real time.The category information and location information of coal gangue are included,which improves the accuracy and stability of the coal gangue detection system.The main research work is as follows:(1)Designing and building a coal gangue target identification and positioning platform: the overall design of coal gangue target detection platform design and platform hardware construction,design camera selection,lens selection,lighting scheme design,etc.,through the target detection experimental platform built to obtain Coal gangue target detection data set,and use the labelImg tool to manually mark the coal gangue target in the image to make a data set for the training of the subsequent coal gangue target detection optimization algorithm.(2)Lightweight YOLO’s coal gangue target recognition and localization algorithm:proposed a coal gangue detection method based on convolutional neural network technology,combined the deep learning method with the coal gangue detection problem,and realized the real-time end-to-end detection of coal gangue targets.At the same time,the lightweight feature extraction network ensures the full extraction of coal gangue features and reduces the complexity of the network,which can realize real-time recognition and location of coal gangue targets.(3)Improve two-scale feature prediction: In this paper,two-scale feature prediction + multi-scale feature fusion method is used to extract coal gangue features and target prediction,so that coal gangue feature maps with different resolutions in the coal gangue feature extraction stage have rich location information and deeper semantic information;by improving this method,the influence of the size change of coal gangue targets can be reduced,and the accuracy of coal gangue target recognition and localization can be further improved.(4)Realizing the visualization of coal gangue feature extraction process: realize the visualization of the feature extraction process based on the class activation map CAM,and use the CNN visualization model to explore the feature areas of coal gangue identification,and further explore the reliable basis for the model to identify coal gangue.Figure [37] Table [8] Reference [61]...