YOLOv5s-TC: An Improved Intelligent Model for Insulator Fault Detection Based on YOLOv5s

Insulators play a pivotal role in power grid infrastructure, offering indispensable electrical insulation and mechanical support. Precise and efficient detection of insulator faults is of paramount importance for safeguarding grid reliability and ensuring operational safety. With the rapid advancements in UAV (unmanned aerial vehicle) technology and deep learning, there has been a notable transition from traditional manual inspections to automated UAV-based detection systems. To further enhance detection accuracy, this study conducts a series of systematic improvements to the YOLOv5s model and proposes an advanced intelligent insulator detection model, namely YOLOv5s-TC. Firstly, this new model replaces the C3 (Cross Stage Partial Bottleneck with 3 convolutions) module with Bottleneck Transformers to enhance feature learning ability. Secondly, the CBAM (Convolutional Block Attention Module) is introduced to make the model focus more on the key features of the images, thus improving the target localization ability. Finally, the improved loss function named OSIoU is adopted to further enhance detection accuracy. Comparative experiments demonstrate that YOLOv5s-TC achieves significant performance gains, with mean average precision improvements of 4.4%, 24.5%, and 13.9% over the original YOLOv5s, Faster R-CNN, and SSD models, respectively. The results indicate that YOLOv5s-TC offers superior detection performance and greater reliability for practical power grid inspection applications.