 This study presents a GCYOLOV5S crack detection network of UAVs that addresses issues such as low efficiency, low detection accuracy due to shadows, occlusions, and low contrast, and influences due to road noise and classic methods. The proposed algorithm uses a focal GIOU loss function with focal loss to address the imbalance of difficult and easy samples in crack images and replaces the original localization loss function CIOU with GIOU for a regular target detection. A transposed convolution layer is added to improve feature representation while the C3Ghost module decreases network parameters while maintaining adequate feature representation. The lightweight CSPCM module successfully reduces model parameters and improves detection accuracy. The study establishes a new UAV road crack detection dataset and conducts extensive trials resulting in increased precision of 8.2%, 2.8%, and 3.1% compared to YOLOV5S with a reduction in model parameters by 16.2%. The proposed algorithm outperforms previous YOLO comparison models in precision, recall MAP underscore 0.5, MAP underscore 0.5 colon 0.95, and PROMS. This article was authored by Xinjian Xiang, Haibin Hu, Yi Ding, and others. We're article.tv, links in the description below.