EFC-YOLO for aerial pavement crack detection

At present, there are problems such as severe background interference, weak crack characteristics and difficulty in balancing detection accuracy and real-time performance in the road surface crack detection in the aerial photography scene. Especially, when the algorithm is to be deployed on unmanned aerial vehicles for real-time processing, this requires extensive optimization of accuracy and speed under limited computing resources. To solve these problems, this paper proposes the EFC-YOLO model for aerial pavement crack detection. This model is based on the YOLOv10 framework and integrates the Feature Enhancement Module (FEM), the Feature Fusion Module (FFM), and the Spatial Context Awareness Module (SCAM). These three modules respectively improve the local perception ability, multi-scale feature fusion ability, and global association ability across channels and Spaces of the network, while avoiding increasing complexity as much as possible. Thereby enhancing the weak feature characterization of fine cracks and suppressing the easily confused background. The validity of EFC-YOLO was verified using the UAV-PDD2023 dataset. Experiments on the unmanned aerial vehicle (UAV) road crack dataset UAV-PDD2023 show that the accuracy of the improved EFC-YOLO reaches 0.929, which is 0.128 higher than the original model. mAP50 reaches 0.909, which is 0.076 higher, surpassing several benchmark models and the most advanced methods. Meanwhile, different tasks were adopted to verify the generalization performance of the model. The experimental results proved that EFC-YOLO has stronger detection performance and can be effectively applied to the detection of pavement cracks from the perspective of unmanned aerial vehicle aerial photography.