Research On Safety Device Identification Of Loading And Unloading Truck Based On Deep Learning

In recent years,with the rapid development of the chemical industry,the demand for chemical product logistics and transportation has been growing,and people have been paying more attention to the safety of the chemical product loading and unloading truck operation process.In the loading and unloading operation process contains many safety devices to assist the safety of loading and unloading operations,such as grounding wire,brake block,safety rope,etc.The grounding wire can conduct away the static electricity generated by the operating vehicle to avoid static electricity generating sparks to ignite flammable substances in the environment,and the stop block can prevent the vehicle from slipping to avoid collisions.With the help of deep learning technology to strengthen the supervision and alarm analysis of safety devices in loading and unloading operation,it can reduce the chance of accidents and provide important reference information for safety management department,and make more scientific and reasonable safety management strategy according to the reference information,so as to guarantee the safety of personnel and environment to the greatest extent.Due to the special characteristics of the loading or unloading truck operation environment,only explosion-proof cameras with poor pixels can be installed at the site,so the identification of safety devices has been suffering from low recognition rate,false alarm and omission,which makes it difficult to apply the image recognition technology to the supervision of loading or unloading truck operation in practice.With the continuous development of deep learning technology,image recognition algorithms have stronger feature extraction capability and can effectively recognize small target objects in low pixel images.In this paper,based on YOLOv5 algorithm,we introduce self-attention mechanism and lightweight module to improve the algorithm,collect images of loading and unloading truck operation sites in factory areas and make a safety device dataset to verify the performance of the improved algorithm and design a recognition system based on the improved algorithm,the specific work is as follows:(1)A safety device recognition method based on Co TNet improved YOLOv5 is proposed.The ground wire in the safety device takes up a small percentage of the monitoring screen with few features and many forms,thus making recognition difficult.Therefore,this paper incorporates the Co TNet self-attention mechanism in the network structure of the YOLOv5 algorithm to improve the algorithm feature extraction capability,and then replaces the loss function at the output of the algorithm with SIOU Loss to improve the recognition accuracy.After the introduction of Co TNet and SIOU Loss,the model accuracy is improved by 5.1% and the accuracy is improved by 2.9%.(2)A security device identification method based on Mobile One improved YOLOv5 is proposed.Whether the model can run with limited hardware resources and time constraints is an important prerequisite for the practical application of the model to the project.In order to reduce the model size and improve the inference speed of the algorithm,this paper incorporates the lightweight module Mobile One into the network structure of YOLOv5 to reduce the number of training parameters and decrease the model size.After the introduction of Mobile One,the model improves the inference speed by 27.4Hz at a sacrifice of only 2.2%,and reduces the size to 15% of the original one.(3)A security device identification method based on MoCot-YOLOv5(Mobile and Co TNet,MoCot)is proposed.The above two improvement methods have their own advantages but also disadvantages,in order to maximize the accuracy of algorithm identification,inference speed and reduce the size of the model,and find the best balance of the three,the Mobile One module and Compared with the original YOLOv5 algorithm,the MoCot-YOLOv5 algorithm has the best detection effect.YOLOv5 improved by 3.2%and the inference speed FPS improved by 15.4Hz.(4)A safety device identification system based on the MoCot-YOLOv5 algorithm was designed and implemented.Using the explosion-proof camera video data of the plant and the information of the loading and unloading operation process,the MoCot-YOLOv5 algorithm based on the experimentally selected MoCot-YOLOv5 algorithm is used to intelligently supervise the safety devices in the loading and unloading operation and to analyze and count the alarm information.