Enhanced YOLOv8 with dynamic convolution and separable attention for robust power line defect detection

Insulator and defect detection is a critical technology for the automated inspection of transmission and distribution lines within smart grids. However, the development of a lightweight, real-time detection platform suitable for deployment on drones faces significant challenges. Traditional detection methods struggle with accurately identifying small insulators and various defect types while maintaining computational efficiency for real-time monitoring applications. This paper proposes an enhanced YOLOv8n architecture incorporating Dynamic Convolution (DynamicConv) and Large Separable Kernel Attention (LSKA) mechanisms for improved insulator and defect detection. The DynamicConv module dynamically selects optimal convolution kernels based on input characteristics through expert kernel weighting, enabling parameter-efficient adaptive feature extraction. The LSKA mechanism decomposes large 2D convolution kernels into separable 1D operations, significantly reducing computational complexity while maintaining extensive receptive field coverage. Extensive experiments on the Chinese Power Line Insulator Dataset (CPLID) demonstrate that our approach achieves superior performance with mAP@0.5 of 93.2%, precision of 91.8%, and recall of 92.4%, outperforming the baseline YOLOv8n by 4.7% in mAP@0.5. The proposed method maintains real-time inference capabilities with 67.8 FPS while providing enhanced accuracy for critical infrastructure monitoring applications suitable for future drone deployment.