Research On Lightweight Coal And Gangue Target Detection Method Based On Improved YOLOv4 Algorithm

In the process of coal production,the separation of coal blocks and gangue is an inevitable and important link.With the national vigorous implementation of coal industry green development-related policies,the coal industry has an increasingly urgent demand for more green and intelligent high-efficiency coal and gangue separation technology.Nowadays,the use of computer vision technology to achieve automatic detection of coal and gangue is gradually being paid attention to,how to achieve more efficient and accurate automatic identification of coal and gangue is the key to improve the efficiency of coal output of the plant and promote the green development of coal industry.In order to be able to identify the gangue quickly and accurately,without placing too high a demand on the performance of the hardware equipment,the paper proposes to use target detection algorithm to achieve automatic detection of coal and gangue targets.The following research work has been carried out in the paper:(1)This paper deeply analyzes the research status of target detection algorithms and coal gangue separation technology at home and abroad,and finally selects a deep learning algorithm with superior recognition accuracy and speed as the research object by comparing the specific performance of deep learning algorithms and traditional algorithms in coal gangue detection Through the use of camera to the total length of the coal block and gangue of Xinjiang origin more than 30 mm to complete the collection of data set,in the collection process through the gangue to flip again to take pictures to improve the utilization rate of the gangue,and to obtain more feature information.The total number of coal block and gangue used in the collection of data set is about 2200,and a total of 1600 images of coal and gangue dataset were obtained.and use the label Img tool to manually mark and classify the coal and gangue targets in the dataset,and finally the dataset is divided into two parts: training set and test set.(2)Analyze and compare YOLO series algorithms,and the YOLOv4 algorithm with higher detection accuracy was selected as the basis for establishing the coal and gangue target detection model.The following improvements were made to the YOLOv4algorithm: the Mobilenetv3 structure with fewer network parameters was selected as the model backbone network for feature extraction to improve the model’s real-time detection performance;Using CBAM attention mechanism and Mish activation function to improve Mobilenetv3 network to optimize model detection accuracy;The Bi FPN network was selected for feature multiscale connection and fusion to optimize the model’s final prediction results.(3)A lightweight gangue target detection method based on improved YOLOv4 is proposed.The improved YOLOv4-based coal gangue target detection model is compared with the original YOLOv4,YOLOv5 s and YOLOv5 m models for experiments,and the experimental results show that the improved model in the paper has a great improvement in detection speed and has a good detection effect,and the model also becomes more lightweight with the best comprehensive performance.