This paper addresses the challenges of complex environmental interference and small target recognition in power system insulator defect detection by proposing an enhanced YOLOv11-based algorithm. We implement three novel mod ules: the Multi-Resolution Edge Feature Integration module (MREFI), the PoolingFormer module (PFM), and the Adaptive Convolutional Gated Linear Unit attention mechanism (ACGLU). The MREFI module strengthens defect boundary recognition by extracting and fusing edge features across multiple spatial scales. The PFM module replaces conventional self-attention mechanisms with efficient pooling operations, substantially re ducing computational complexity while preserving global infor mation interaction capabilities. The ACGLU module enhances feature representation by integrating convolutional operations with gating mechanisms. Experimental results demonstrate that our improved algorithm significantly outperforms the original YOLOv11modelindetection accuracy, with mAP@0.5 increasing by 3.3 percentage points and mAP@0.5-0.95 by 5.1 percentage points. Simultaneously, the model achieves reduced computa tional complexity and parameter count, making it more suitable for practical deployment in power system inspection applications.

defect detection, yolov11, PFM, ACGLU