Lightweight insulator defect detection based on improved YOLOv5

To address the challenge of balancing model complexity and detection accuracy in the task of detecting defects in transmission line insulators, this paper proposes a lightweight detection method based on an improved YOLOv5. The backbone network is reconstructed by integrating depthwise separable convolution and Ghost Net to construct the GDConv module, significantly reducing the model’s parameter size and computational load. The DC-SPP module is designed to replace the traditional SPP structure, leveraging dilated convolution and adaptive average pooling to fuse multi-scale features and mitigate the loss of small target features. Additionally, a structured pruning strategy is employed to further optimize the model size. Experimental results demonstrate that the improved model achieves a mean Average Precision (mAP) of 93.5% on the self-built insulator defect dataset, representing a 9.2% improvement over the original YOLOv5s. The parameter size is reduced to 3.78M, and the detection speed is increased to 35 FPS. Compared to mainstream models such as Faster RCNN and SSD, the proposed method achieves a balance between accuracy and speed, adapts to inspection requirements under complex weather conditions, and provides an efficient solution for deployment on drone-embedded devices, demonstrating significant engineering application value.