Bird Detection Algorithm for Transmission Line Based on Improved YOLOv7

Abstract

With the increase in the number of birds, the life of birds will seriously endanger the safety of transmission lines or transmission towers. To ensure peaceful coexistence between transmission towers or lines and birds, this paper proposes to identify birds based on the improved YOLOv7 network structure, which lays the foundation for driving or guiding different birds. Replace part of the Conv layer in YOLOv7 with DOConv layer, improve the network performance through deeply over-parameterized convolution, and do not over-parameterize in reasoning; Add SPDConv layer in small target recognition to improve the recognition rate of small target; SimAM attention mechanism is added to feature extraction to enhance feature extraction. Some birds fly at high speed, resulting in blurred pictures taken by traditional cameras, which will greatly reduce the recognition rate of birds. In this paper, motion blur processing of data sets is beneficial to the recognition of blurred bird images. The model can accurately identify birds in the collected images of power transmission lines, providing a reference for the prevention and control of bird-related faults on power transmission lines.