Research On The Hidden Danger Detection Method Of Mine Belt Conveyor Jamming Based On Deep Learning

China is a large energy-consuming country and mainly relies on coal as its primary energy source,while other energy sources are relatively scarce.Due to the complex mining conditions,coal transportation usually relies on conveyor belts.However,during the transportation of coal,there are hidden dangers in the carbon flow of the conveyor belt,such as large chunks of carbon(gangue),screws,wire mesh,etc.,which can easily cause the conveyor belt drive point or head to jam,affecting normal production.This thesis focuses on the hidden dangers of conveyor belt transportation and conducts research on related methods of deep learning.The main research contents are as follows:(1)Annotation and pre-processing of coal mine conveyor monitoring video image dataIn the process of training the recognition of hidden dangers in coal mine conveyor monitoring video images,a large amount of image data needs to be collected,which must be ensured to be real,accurate,complete,and rich to meet the algorithm’s requirements.In addition,the dataset must be free of redundancy and errors so that the algorithm can process and predict images more quickly.Therefore,the quality of the dataset is one of the key factors affecting the design of hidden danger detection algorithms and recognition accuracy.The dataset used in this study is a special scenario dataset,which has the characteristics of high annotation cost,insufficient data volume in public datasets,and insufficient quality of related annotation in public datasets.In addition,the coal mine dataset needs to meet the safety factor in a real environment.To solve this problem,the dataset used for model training in this thesis is manually annotated according to the standardized model annotation standards.In the process of monitoring hidden dangers in coal mine conveyor monitoring video images,the feature scale of the target object is different,and the scale will affect the calculation results when calculating the distance.To solve the difference between feature scales,this thesis uses image standardization methods to preprocess the coal mine conveyor monitoring video image dataset to ensure the convergence of the model and improve the model’s generalization ability,ultimately ensuring the detection results of hidden dangers in coal mine conveyors.(2)Coal mine conveyor hidden danger detection method based on SwinTransformer and Faster-RCNNThe private dataset annotated by coal mine conveyor monitoring video images has the characteristics of a small sample size,which can affect the model’s generalization ability and even lead to overfitting.To solve this problem,this thesis proposes to use transfer learning technology as the model training method.In image detection,image feature extraction is closely related to the feature backbone model,and existing methods have the problem of incomplete image feature extraction.The detection effect of the image is closely related to the detection algorithm,and existing methods have the problem of varying detection capabilities.To solve this problem,this thesis proposes a coal mine conveyor hidden danger detection model based on Swin Transformer and Faster R-CNN.Firstly,Swin-Transformer(Tiny)was selected as the feature extraction network to construct the object detection algorithm.To solve the multi-scale problem in object detection,a feature map pyramid network was introduced.To select the best candidate boxes generated by RPN,the traditional non-maximum suppression method was used.Next,the suitable loss function was determined.Finally,the structure of the coal mine conveyor hidden danger detection algorithm based on Swin Transformer and Faster R-CNN was constructed.The experiment shows that in the real coal mine conveyor monitoring video image annotated dataset test of a certain coal mine,the coal mine conveyor hidden danger detection model based on Swin-Transformer and FasterRCNN proposed in this thesis achieves a recall rate of 96.8% and 97.7% for large coal and foreign objects detection,respectively,and the accuracy of these two methods reaches 88.3% and 90.7%,resulting in effective detection.In addition,the detection speed is fast,and real-time performance is ensured during the detection process.(3)Prototype and applicationTo verify the effectiveness of the proposed method in industrial scenarios,the AI model trained by the algorithm was deployed and applied in the conveyor AI monitoring system.Firstly,the overall deployment architecture of the conveyor AI detection system was introduced;then,the deployment locations for foreign object detection and large coal monitoring of the camera were explained based on actual scenarios.Finally,the application effect of the system was explained based on the actual detection effect.This thesis has 19 figures,10 tables,and 90 references.