<p>Deep learning models deployed on drone platforms for bridge inspection often face challenges in computational complexity and real-time performance. To address these issues, we propose a lightweight detection model, LiteBC-YOLO. First, we construct a dedicated concrete bridge defect dataset that includes typical damages such as cracks, spalling, exposed steel bars, and honeycomb surfaces, providing high-quality support for model training. Second, with YOLOv8n as the baseline, we design a lightweight network architecture: PKINet serves as the backbone to enhance feature extraction of subtle damages in complex backgrounds; the DMGConv module is integrated into the neck to reduce the computational cost of multi-scale feature fusion; and a multi-scale parameter-sharing detection head is introduced to minimize redundancy. Finally, structured pruning and knowledge distillation are applied to further optimize efficiency without sacrificing accuracy. Experimental results show that LiteBC-YOLO achieved an AP50 of 92.84%, reduced parameters by 50.5%, and decreased FLOPs by 59%. In engineering validation, the model shortened inference time by 38%, improved AP50 by 7.16%, increased inference frame rate by 59%, and lowered power consumption by 35%. Overall, LiteBC-YOLO effectively balances detection accuracy and resource efficiency, offering an efficient solution for real-time intelligent bridge inspection using Unmanned Aerial Vehicles (UAVs).</p>