基于GEM-YOLOv11n的道路缺陷检测模型OA

To address issues in road defect detection,such as low accuracy in multi-scale object recognition and the tendency for small cracks to be missed or misidentified,a road defect detection model based on the global feature enhanced multi-scale-you only look once version 11 nano(GEM-YOLOv11n)was proposed.First,a global attention mechanism(GAM)was introduced into the YOLOv11n backbone network to enhance focus on key area features while background interference was suppressed.Second,a multi-scale edge information enhance-convolutional three-scale kernel-adaptive dual-path(MSEE-C3k2)module was designed to improve feature extraction capability for defects of different scales.Finally,a bidirectional feature pyramid network(BiFPN)was employed to reconstruct the feature fusion layer,utilizing a bidirectional connection mechanism to achieve deep fusion of multi-level features.The experimental results on the road damage detection 2022(RDD2022)dataset demonstrated that GEM-YOLOv11n model increased by 4.8 and 3.4 percentage points with the mean average precision at intersection over union thresholds of 0.50 and 0.50～0.95 respectively compared with original YOLOv11n model,while a low parameter count and floating-point operation speed were maintained.Furthermore,the effectiveness of each module and the model's robustness across scenarios were validated by ablation experiments and generalization experiments.GEM-YOLOv11n model significantly improved the detection accuracy and adaptability while maintaining low computational cost,providing an efficient and feasible technical solution for road inspection tasks.