An improved algorithm of YOLOv8 for defect detection in high voltage transmission lines

High-voltage transmission lines are exposed to the outdoors for a long time, and the harsh environment is easy to cause damage to high-voltage transmission lines. The damaged area is large or small, and the transmission line background is complex, so there are multi-scale defects and complex image background information interference problems. In order to solve the above problems, a high voltage transmission line detection algorithm based on improved YOLOv8 called MFD-YOLOv8 is proposed. Firstly, Mamba-block is used in the backbone network to enhance the modeling ability of the backbone network for long sequences and realize the capture of global information by the model; Then, a Feature-Focused Diffusion Pyramid Network (FFDPN) is proposed, which makes the features of each scale have detailed context information, which is more conducive to the subsequent defect detection and classification. Finally, a Dynamic Weight Selection (DWS) module is designed to give different weights to different convolutions, which can enhance the feature information of defect areas and weaken the background information. Aiming at the lack of relevant data sets in the field of high-voltage transmission line defect detection, a variety of high-voltage transmission line defect data sets are constructed. The experimental results show that on the self-built dataset, MFD-YOLOv8 improves the precision by 6.0% and the recall by 5.1% compared to the baseline.