Detection Method Of Bird’s Nest On Transmission Lines Based On Attention Mechanism

As the scale of science and technology and industrial production in China continues to expand,the demand for power resources in various fields is getting higher and higher,and ensuring the safe and reliable operation of power transmission lines has become a top priority.However,a bird’s nest on transmission lines can pose a threat to the safe operation of power equipment and even affect the stability of the entire power system.In recent years,UAV inspection technology has made rapid progress,and the use of UAVs to obtain inspection images and locate and identify bird’s nests on transmission lines with the help of target detection technology is of great significance to reduce the cost of power inspection and improve inspection efficiency.However,the complexity of scenes in transmission line inspection tasks and the uncertainty of bird’s nest targets lead to the existing transmission line bird’s nest detection algorithms with low detection accuracy and high leak detection rate.This paper focuses on how to further explore the feature information of bird’s nests in inspection images to improve the target detection algorithm’s ability to locate and identify transmission line bird’s nests while maintaining the real-time nature of the algorithm.The main research content of this paper is as follows:(1)A method of detection of bird’s nest on transmission lines based on attentional feature fusion is designed.The method solves the problem of underutilization of bird’s nest information in different levels of features in the network by designing an attentional feature fusion network and embedding it into the YOLOv3 target detection algorithm.The attentional feature fusion network uses the channel attention mechanism to calculate the semantic weights of deep-level features and uses this semantic weight as a guide to filter the shallow-level features to remove the redundant information in the shallow-level features and help the shallow-level features to obtain the category differentiation ability like the deep-level features.After that,the filtered shallow features are fused with the deep features to obtain features containing rich bird’s nest information and improve the detection performance of the target detection algorithm for transmission line bird’s nests.(2)A detection method of bird’s nests on transmission lines based on feature balance and feature enhancement is designed.The method takes into account the information differences existing between different levels of features in the YOLOv5 target detection algorithm and designs a feature balance network by combining two attention mechanisms,channel,and spitial.The feature balance network is guided by semantic weights and spatial weights,respectively,to achieve the balance of semantic and spatial information between different levels of features.At the same time,to avoid the problem that the information of bird’s nest features is continuously weakened due to the gradual increase of the number of network layers,the feature enhancement module is designed to capture the channel information and location information related to the bird’s nest in combination with the coordinate attention mechanism to enhance the detection algorithm’s ability to distinguish the bird’s nest from the background and improve the applicability of the algorithm for transmission line bird’s nest detection while maintaining the real-time nature of the algorithm.This paper constructs a bird’s nest on a transmission line dataset for algorithm training and testing by collecting UAV inspection images and conducting a series of comparison experiments.The experimental results show that the transmission line bird’s nest detection method designed in this paper has strong generalization capability and applicability for a variety of inspection scenarios,and can significantly improve transmission line inspection efficiency and reduce operation and maintenance costs.