Enhanced YOLOv11-Based Algorithm for Insulator Defect Detection

With the expansion of power systems and increasing environmental complexity, insulator defect detection has become crucial for ensuring grid safety. Traditional manual inspection suffers from low efficiency and high risks, while existing deep learning algorithms face challenges including insufficient model generalization capability, low precision in small object detection, and high computational complexity. To address these issues, this study proposes an enhanced YOLOv11 algorithm through optimized feature extraction and dynamic computation mechanisms. Firstly, DySample replaces conventional upsampling modules to reduce inference latency and memory consumption via dynamic point sampling. Secondly, Strip Pooling modules are embedded after partial C3k2 layers to enhance contextual feature capture for slender targets through elongated kernel pooling. Finally, the attention layer in the C2PSA module is replaced with BiFormer's Bi-Level Routing Attention mechanism, which reduces redundant computations through sparse routing strategies while strengthening long-range feature modeling. Experimental results demonstrate that the improved model achieves <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$92.6 \% \text{mAP} 50$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$63.1 \% \text{mAP} 50-95$</tex>, representing 3.3% and 3.5% improvements over baseline YOLOv11 respectively. Comparative experiments confirm that our algorithm strikes a superior balance between accuracy and speed compared to mainstream models, validating its efficiency and robustness in complex scenarios. This work provides a viable technical solution for intelligent power inspection.