Insulator defect detection based on YOLOv11s

Insulators play a crucial role in complex power grids, primarily responsible for insulation and support during the electricity transmission process. However, they often face challenges from external factors such as lightning strikes and environmental pollution. In intelligent inspection processes, it is essential to detect the damage to insulators accurately and promptly to avoid potential safety hazards and equipment failures caused by damaged insulators. As a result, insulator defect detection has become a research hotspot in the field of object detection. The YOLO algorithm is an object detection method based on convolutional neural networks (CNN), differs from traditional detection methods in that it can obtain the positions and categories of all targets through a single forward inference, enabling efficient real-time object detection. To enhance the detection accuracy of insulators and their defects in complex environments while minimizing the computational burden of the model, this paper introduces a defect detection algorithm named Yolo-Net for transmission line insulators based on an improved version of YOLOv11s. The proposed approach adopts YOLOv11s as the base detection model and incorporates a Bidirectional Feature Pyramid Network (BiFPN) to improve accuracy by leveraging multi-scale feature pyramids and bidirectional information propagation. Additionally, a multi-scale attention mechanism (MSDA) is incorporated, and the loss function is changed to WIOU. Compared with yolov11, the final average accuracy achieved in testing from 89.2% to 91.4%. The experimental results are then validated by comparing them with those obtained using other YOLO versions, including YOLOv5s, YOLOv7s, YOLOv8s, YOLOv9s, and YOLOv10s.