YOLOv11-SR: An Enhanced YOLOv11-Based Framework for Defect Detection in Complex Substation Inspection Images

Abstract

Accurate detection of explicit faults in substation equipment is critical for intelligent substation inspection. However, practical inspection scenes often involve small objects, blurred boundaries, background clutter, occlusion, and large-scale variations, which significantly challenge the reliability of existing object detection methods. To address these issues, this paper proposes YOLOv11-SR, an improved object detection framework for explicit fault detection in complex substation environments. Specifically, a Shallow Detail Preservation Attention Module (SDPAM) is introduced to preserve fine-grained shallow details during downsampling, while a Position-Relation Guided Adaptive Fusion Module (PRGAFM) is designed to enhance adaptive multi-scale feature fusion under cluttered backgrounds. Experimental results show that YOLOv11-SR achieves 89.9% Precision, 87.8% Recall, 89.0% F1-score, 90.6% mAP@0.5, and 64.8% mAP@0.5:0.95, outperforming representative detectors including YOLOv5, YOLOv7, YOLOv11, YOLOv13, and YOLO-SS-Large. Category-level analysis and ablation studies further verify the effectiveness of the proposed method, particularly for detecting small and blurred targets in complex scenes. These results demonstrate the effectiveness and robustness of YOLOv11-SR for intelligent substation defect inspection.