GEP‐YOLO: A Lightweight Real‐Time Detection Model for Unmanned Aerial Vehicle Road Cracks Integrating Multi‐Module Optimization

Rapid timely detection of road surface cracks is critical for precise pavement maintenance and vital to extending pavement service life. Based on the YOLOv11n architecture, this paper proposes GEP‐YOLO, an improved lightweight real‐time object detection algorithm for pavement cracks. Firstly, by adjusting Unmanned Aerial Vehicle (UAV) parameters to capture images, followed by data augmentation and annotation, a dedicated UAV road crack dataset was constructed. Secondly, targeted improvements were made to YOLOv11n: GSConv was embedded into the C3K2 bottleneck blocks of the Backbone to replace partial standard convolutions, thereby reducing computational complexity and optimizing feature extraction; EUCB was introduced into the top‐down feature propagation path of the Neck to replace the original conventional upsampling operations, thereby restoring the spatial resolution of feature maps while enhancing the local detail representation of the upsampled features; and PReLU replaced the activation functions in the Backbone and Neck to strengthen nonlinear feature representation capability. Finally, a series of ablation studies and comparative experiments were conducted on the self‐built dataset, validating the superiority of the proposed model. The experimental results demonstrate that compared to YOLOv11n, GEP‐YOLO achieved improvements of 2.6% in mA p @0.5, 2.8% in mA p @0.5:0.95, 4.0% in Precision, 3.2% in Recall, and 14.8% in Frames Per Second (FPS). Furthermore, compared to other mainstream models, GEP‐YOLO exhibits superior capability in identifying small‐sized cracks, higher classification accuracy, and stronger robustness against low‐contrast interference.