基于YOLOv11n的桥梁水下结构病害轻量化快速目标检测模型OA

To achieve efficient detection of bridge underwater defects,a lightweight and fast object detection model CF-YOLOv11n was proposed.A context-aware local enhancement attention mechanism with a dual-branch structure was introduced to fuse global semantic and local detail information,while depth-wise convolution was used to reduce the redundant computation and improve the inference speed.A frequency-modulation feed-forward network performs local-window frequency domain filtering to realize multi-scale feature modeling and suppress background interference,thereby improving the detection accuracy.The results show that the mAP0.5:0.95(mean average precision at intersection over union thresholds from 0.5 to 0.95)of CF-YOLOv11n on a real bridge dataset reaches 45.50%,which is 2.46%higher than that of the baseline.The inference speed is 66.13 frame/s,which is 2.25 times faster than that of the baseline.Com-pared with the baseline model,the proposed model can capture multi-scale information more effectively,accel-erate the inference process,and balance accuracy and speed,demonstrating stronger engineering application value in real-time detection tasks under actual underwater bridge environments.