Research On Foreign Body Detection Algorithm For Power Transmission Lines Based On Deep Learnin

As an important carrier of electric energy transmission,transmission lines are of great importance.If the transmission line fails,it will cause irreversible damage to the power system.Therefore,the maintenance personnel of the power system need to frequently inspect foreign objects such as bird’s nests,garbage,kites and balloons on the transmission line to prevent these foreign objects from causing short circuits on the transmission line and causing sudden power outages.However,manual inspection not only consumes a lot of manpower and material resources,but also some missed inspections may occur.In order to reduce the consumption of manpower and material resources during power inspections and improve the accuracy of inspections,this paper conducts research on the detection of foreign objects on transmission lines based on UAV inspections combined with mainstream algorithms of deep learning.The main research contents are as follows:(1)This paper proposes a transmission line foreign object detection algorithm based on improved Center Net.Aiming at the problem of poor detection of small targets on transmission lines,an improved Center Net algorithm is proposed.The specific improvements are as follows:in order to extract more abundant semantic features,the residual network of the Center Net algorithm is improved to a wide-body residual network.And combined with the attention mechanism,the extraction of feature information is strengthened.It has been verified by experiments that the improved Center Net algorithm can effectively detect foreign objects on transmission lines,and the m AP has increased by 5.03% compared with that before the improvement.(2)This paper proposes a transmission line foreign object detection algorithm based on improved Center Net.Aiming at the problem of poor detection of small targets on transmission lines,an improved Center Net algorithm is proposed.The specific improvements are as follows:in order to extract richer semantic features,the residual network of the Center Net algorithm is improved to a wide-body residual network.And combined with the attention mechanism,the extraction of feature information is strengthened.It has been verified by experiments that the improved Center Net algorithm can effectively detect foreign objects on transmission lines,and the m AP has increased by 5.03% compared with that before the improvement.(3)Due to the frequent downsampling of the C-YOLOv5 algorithm,the semantic information of small objects is lost,which affects the overall performance of the algorithm.This paper proposes a transmission line foreign object detection algorithm based on CYOLOv5-new.Before the frequent downsampling of the C-YOLOv5 algorithm,the CYOLOv5-new algorithm jumps and connects the shallow feature map to the multi-layer Neck network,combines the residual connection,and introduces the designed Multicat module to combine large,medium,and The fusion of feature maps of three different sizes can retain more semantic information of small targets,so that the feature information of small targets of foreign objects on transmission lines can be preserved.Experiments show that compared with the CYOLOv5 algorithm,the C-YOLOv5-new algorithm is better in foreign object detection on transmission lines,and the m AP is increased by 1.6% compared with that before the improvement.Finally,compared with other mainstream algorithms,the performance of CYOLOv5-new algorithm is also better.