InsDD-YOLO: Detection of Transmission Line Insulator Damage Based on the Improved YOLOv13 Model

Amidst the rapid global expansion of smart grids, ensuring the safety and reliability of power transmission systems has become paramount. Insulators are critical components of high-voltage transmission lines, providing both electrical insulation and mechanical support. However, their exposure to electrical, mechanical, and environmental stressors renders them vulnerable point within the system. Defective insulators are a major cause of failures in power transmission systems. Consequently, the early and accurate detection of these defects is pivotal for maintaining the integrity and reliability of the power grid. To address this challenge, this study proposes InsDD-YOLO, a novel object detection architecture enhanced from the YOLOv13 framework. The model incorporates a suite of strategic enhancements, including an improved DSConv (IDSConv) module for robust feature extraction, a streamlined Neck architecture augmented with a feature stream from a shallower layer (B2) to improve small-target detection, and a direct Head connection mechanism to maximize the preservation of fine-grained details. Experimental results demonstrate that InsDD-YOLO achieves superior performance, reaching an mAP0.5 of 90.1% and an mAP<inline-formula> <tex-math notation="LaTeX">${}_{0.5:0.95}$ </tex-math></inline-formula> of 46.4%, outperforming the baseline YOLOv13 model by a significant 5.0% in mAP0.5. With an inference time of just 5.4 ms, the proposed model not only establishes a new benchmark for accuracy but also demonstrates an effective trade-off between performance and speed, underscoring its significant potential for deployment in real-time, automated power grid monitoring systems.