A lightweight multi-type substation defect detection method based on YOLOv11n

Under the background of the rapid development of smart grid, UAV and other devices have become the main force of transmission line inspection. However, the images captured by UAV have the characteristics of complex background and dense small targets, which lead to the problems of few detection types, low efficiency and high false alarm rate in the existing models in substation defect detection, restricting the intelligent process of inspection. This paper proposes a lightweight multi-type defect detection model based on YOLOv11n for real-time online detection on lightweight devices. The main work includes: The neck network structure of YOLOv11n was optimized, the C3k2 modules of the 13th, 16th, 19th and 22nd layers of the neck network were replaced with the C3k2_Faster module to reduce the number of convolution layers. The channel-wise distillation (CWD) technology was introduced, and the improved YOLOv11s was used as the teacher model and YOLOv11n as the student model. Through the optimization of channel dimension information transmission, the student model feature expression ability was improved. Experimental results show that the improved model achieves a 2.7% increase in mAP50, a 5.6% reduction in parameters, and a 4.6% reduction in GFLOPs compared to YOLOv11n.