SBD-YOLOv5: An Enhanced YOLOv5-Based Method for Transmission Line Fitting Defect Detection

One key method to ensure the proper functioning of the power transmission system is the detection of defects in transmission line hardware. The utilization of deep learning technology to analyze hardware defects in transmission line photos captured by drones has emerged as a significant technique in power system inspections. However, current techniques still exhibit limitations in terms of recognition accuracy. To address this issue, this paper presents a YOLOv5-based algorithm for detecting defects in transmission line hardware, named SBD-YOLOv5, derived from the initials of its three improvement methods. Firstly, to address the deficiency in recognizing small targets, a feature extraction module based on Swin Transformer is incorporated into the backbone network, enhancing the model's capability for feature extraction. Secondly, BiFPN is introduced to further enhance feature fusion. Lastly, DRConv is employed to emphasize key features in different semantic regions. The experimental results demonstrate that the proposed model outperforms the original model with a 4.1% improvement in mAP@0.5:0.95. GradCam heatmaps visually confirm that the improved model places higher significance on the detected targets, underscoring their significance within the enhanced model.