An integrated method for image enhancement and defect detection in UAV inspection of transmission lines

UAV inspection is a core technology for the intelligent operation and maintenance of transmission lines. However, the images captured by UAVs are susceptible to light fluctuations, atmospheric turbulence, and UAV jitter, leading to quality degradation issues such as grayscale distortion and edge blurring. Meanwhile, transmission line defects (e.g., insulator self-explosion, pin loss) are characterized by complex backgrounds and a high proportion of small targets, which restrict the recognition accuracy of traditional methods. To address these challenges, an integrated method of image enhancement and defect detection is proposed: in the enhancement stage, Retinex-MSRCR is fused with super resolution reconstruction to separate illumination and reflection components, thereby optimizing image quality; in the detection stage, YOLOv5s is adopted as the backbone, with the introduction of Swin Transformer to enhance global semantic modeling, embedding of CBAM attention mechanism to focus on defect regions, and integration of BiFPN to achieve multi-scale feature fusion. Experiments conducted on a self-built dataset of 8,200 images covering 6 types of defects, as well as public datasets PLD and CPLID, demonstrate that the proposed method achieves a mAP@0.5of91.3%, representing an improvement of 18.6% and 7.5% compared with the traditional Hough transform and the original YOLOv5s, respectively. Additionally, it achieves a precision of 92.1%, a recall of 88.7%, and a frame rate of 58FPS. This method can effectively break through the bottlenecks of image degradation and defect detection, providing technical support for the intelligence of UAV inspection in transmission line maintenance.