GSE-YOLO: An Edge-Enhanced Lightweight YOLO Framework for Sustainable Infrastructure Crack Detection

Aging road and concrete infrastructure requires efficient crack inspection methods that can support timely maintenance while reducing manual labor and computational cost. Existing lightweight detectors often improve speed by compressing features, but this may weaken the representation of crack edges and fine textures. This paper proposes GSE-YOLO, an edge-enhanced lightweight crack detection framework based on YOLO11n for resource-efficient infrastructure monitoring. The framework introduces GhostEdgeConv, which combines Ghost-style feature generation, Sobel edge responses, and an edge-gated enhancement mechanism to strengthen slender and low-contrast crack features with limited overhead. SPPELAN and EMA are further incorporated to improve multi-scale contextual aggregation and key-region feature recalibration. Experiments on the Crack-Seg dataset show that GSE-YOLO achieves a mAP of 81.7%, improving the YOLO11n baseline by 2.5 percentage points. Precision and Recall increase by 4.7 and 1.5 percentage points, respectively, while parameters and GFLOPs are reduced by 17.8% and 14.3%, and the inference speed reaches 500 FPS. These results indicate that GSE-YOLO can provide a practical balance between detection accuracy, computational efficiency, and deployability for sustainable infrastructure maintenance.