Research On Foreign Object Detection Method Of Coal Mine Conveyor Belt Based On Machine Vision

The coal conveyor belt in the process of transporting coal,coal mixed with foreign objects such as gangue,iron,which will not only cause scratches and tears to the coal conveyor belt,but also affect the quality of coal.Therefore,it is particularly important to accurately detect foreign objects at the initial stage when they enter the coal conveyor belt.Aiming at the problems that existing methods easily lead to missed detection of small-sized foreign objects and excessive dependence on hardware resource performance,this paper explores the effective methods in image preprocessing and foreign object target detection based on machine vision.It has laid an important technical foundation for accelerating the intelligent development of coal mine.The main research contents and relevant conclusions of this study are as follows:(1)Aiming at the interference of dust,light and performance of hardware and equipment in the image of coal mine conveyor belt collected in complex environment,the Wiener filtering algorithm is used to eliminate motion blur in the images;the Gauss filtering algorithm is used for image denoising;and the gamma correction algorithm is used for image enhancement.The experimental results show that the above algorithms can effectively eliminate motion blur in the coal mine conveyor belt foreign object images,eliminate the image noise,protect the edge details of the image and highlight the characteristics of foreign objects in the image,which lays a foundation for subsequent research on foreign body detection of coal mine conveyor belt.(2)Aiming at the problem that the key foreign object features are easy to be lost due to the small target of foreign objects in coal mine conveyor belt,the improved YOLOXm network is adopted to detect small-sized foreign objects in coal mine conveyor belt.The improvement strategies include:1)adopting the Bidirectional Feature Pyramid Network(BIFPN)as the feature fusion network;2)adding the Coordinate Attention(CA)module to the feature extraction network part;3)using Efficient Intersection Over Union(EIOU)as the bounding regression box loss function.Eventually,the effectiveness of the above improvement scheme was verified through a series of experiments.The average accuracy of the improved YOLOXm network model is 92.7%,with a recall rate of 91.1%,a speed of 52 f·s-1,and a model volume size of 25.7MB.Compared with SSD,YOLOv4,and YOLOXm models,the average accuracy of the algorithm has been improved by 10.2%,5.9%,and 5.0%,respectively,indicating that the algorithm has a reasonable structural design and excellent performance.Therefore,it is feasible and effective to use this algorithm for the detection of small-sized foreign objects on coal mine conveyor belt,and can provide a technical reference for intelligent safety detection research in underground coal mines.(3)Aiming at the problems of high structure complexity,large parameters,poor flexibility and over-dependence on hardware resources performance in existing detection methods for foreign objects in coal mine conveyor belt,a lightweight detection method for foreign objects in coal mine conveyor belt based on improved YOLOv5s is proposed.The improvement strategies include:1)using ShuffleNetv2 as the feature extraction network;2)using the Hardswish activation function;3)expanding the size of the depth-separable convolutional kernel to 5×5;4)adding the Efficient Channel Attention Network(ECANet)module between the feature extraction network and the neck network.Finally,a large number of experiments were conducted to verify the effectiveness and advancement of the above improvement strategies.The average accuracy of the improved YOLOv5s model was 90.3%,the recall rate was 88.7%,the frame rate was 61 f·s-1,and the model size was reduced to 3.7MB.Compared with the YOLOv5s original model,the average accuracy increased by 3.8%,the recall rate increased by 4.3%,and the model volume size decreased by 73%.This indicates that the algorithm proposed in this paper can significantly reduce the size of the model while ensuring detection accuracy and can provide technical support for embedded mobile device platform applications.(4)The above-mentioned method is combined with image acquisition equipment to form a coal mine conveyor belt foreign body detection system.The overall structure of the system is designed according to the actual detection needs,including the image acquisition part and the computer software part.The experimental test results show that the system can detect foreign bodies on coal mine conveyor belts in different complex environments and has a strong practical value,which is important for improving the safety of conveyor belt operation.