As an important link in the transmission and use of electrical energy, the safety and stability of transmission lines play a crucial role in the normal operation of the power system (EPS). However, the infrastructure of transmission lines has long faced challenges from the natural environment, and its physical state is inevitably damaged, with frequent problems such as internal crack propagation or external damage. In response to the low precision of defect recognition in traditional transmission line inspection methods, this paper designs a defect recognition model based on machine learning (ML) and unmanned aerial vehicle (UAV) thermal infrared images. In the model, this article proposes an improved approach YOLOv5-GN-CB (YOLOv5-GhostNet-Coordinate Attention & Bidirectional Feature Pyramid). The infrared image recognition method introduces GhostNet into the YOLOv5 backbone network, adds attention mechanism CA and weighted bidirectional feature pyramid (BiFPN), and has efficient bidirectional cross scale connection and weighted feature fusion, which can achieve more feature fusion. Simulation experiments show that this model can improve the precision of defect identification in transmission lines, while maintaining high detection precision, achieving a certain degree of optimization, reducing the complexity and computational load of the model.
Qi et al. (Sun,) studied this question.