Research On Transmission Line Panoramic Monitoring Technology Based On Convolutional Neural Network

Electric energy is an important basic energy for national economic development and social construction.As an important carrier for transmitting electric energy,overhead transmission lines are crucial for providing stable electric energy.Due to the wide coverage of transmission lines and the complex terrain involved,transmission line inspection work is facing enormous challenges.Transmission line patrol inspection partially relies on video equipment on the transmission line,while traditional video monitoring can only provide video monitoring and video recording functions,unable to actively identify hidden danger targets around the transmission line.In recent years,with the continuous development of deep learning theory,the deep learning-based target detection algorithm has a broad application prospect in the power industry.It is of great practical significance to study the deep learning-based algorithm for hidden danger target detection around the transmission line tower.In order to eliminate the impact of hidden danger targets on the safety of transmission lines,and in accordance with the need to prevent external damage to transmission lines,this article uses the YOLOv5s network as the basis to identify and monitor hidden danger targets around transmission line towers.The main research content of this article is as follows:(1)Based on the video images collected by the transmission line video monitoring equipment,a transmission line dataset is created that includes five potential transmission line targets,such as construction machinery,cranes,tower cranes,wire foreign objects,and fireworks,and the dataset is expanded.Experiments are conducted on both the original dataset and the extended dataset to verify the effectiveness of the extended data corresponding to the network performance improvement.(2)According to the monitoring requirement of transmission line hidden target,the deficiency of YOLOv5s network in detecting transmission line hidden target is improved.Firstly,in order to solve the problem of low recognition accuracy caused by complex background of transmission lines and low quality of video surveillance images,combining the PVT network with YOLOv5s network,the model’s ability to distinguish target from background is strengthened,and the detection accuracy is improved by 7.93%.Secondly,attention mechanism was added to YOLOv5s network to enhance the expressive ability of target channel characteristics and spatial characteristics,resulting in a 6.73%increase in network mAP.Then SIoU is used to improve the network loss function,which increases the network mAP by 2.54%.Finally,the non-maximum suppression in the post-processing phase is improved by using the Gaussian function,which allows the network to retain more prediction boxes when judging overlapping targets,improves the recognition effect of occluded targets,and improves the mAP by 1.25%.Finally,several methods are combined to improve the recognition accuracy by 14.54%.(3)According to the lightweight requirements and development process of the model,a filter grouping method based on spectral clustering is first proposed.After converting the grouped filter banks into vectors,the clustering center points of the convolutional kernel vectors are clustered,and the Mahalanobis distance from each vector to the center point is used as the pruning basis for pruning.Then,the pruned convolutional layer is reconstructed.Finally,a knowledge distillation algorithm is designed to improve the network detection accuracy after pruning.