A Lightweight and Efficient Insulator Defect Detection Model for Unmanned Aerial Vehicle Inspection

ABSTRACT Insulator defect detection is crucial for ensuring the safety of power grids, but it faces challenges such as complex environmental interference, small‐scale target recognition, and edge device deployment. This paper proposes an improved lightweight model based on YOLOv11. The model takes Faster‐Net as the backbone and uses partial convolution to reduce computational costs. The neck network is constructed through hierarchical feature fusion blocks to enhance multiscale feature fusion. Introduce a Parallel Patch‐aware Attention (PPA) module into the detection head to enable the model to adaptively focus on the key defect areas in the image. Meanwhile, a Lightweight Adaptive Extraction (LAE) module is integrated to optimize the feature calculation process. Experiments show that this model achieves a mAP50 accuracy rate of 93.2% and a precision of 93.7% on public datasets, with the parameter count and computational load being only 2.3 M and 4.6 GFLOPs respectively. The ablation experiment verified the effectiveness of the module, and cross‐dataset tests confirmed the excellent generalization performance. The model was successfully deployed on the Jetson Xavier NX platform and the M350 Unmanned Aerial Vehicle (UAV), achieving real‐time end‐to‐end detection. This research provides a complete solution for intelligent power inspection in complex environments that is highly accurate, lightweight and easy to deploy.