【摘要】 针对无人机检测路面病害的实时性与准确性需求，提出基于改进YOLOv8的轻量级检测算法LHP-YOLO。将RG-C2f模块替换原C2f模块以减少计算冗余，设计了轻量化检测头Detect-T3G以减少模型参数量，在颈部网络嵌入内容引导注意力融合机制以提升复杂背景下的目标检测精度，应用通道级知识蒸馏以补偿轻量化带来的精度损失。在路面病害数据集RDD 2022上的实验表明：改进模型的mAP50较原模型提升3.4%，参数量与计算量分别为1.76×106和3.9 GFLOPs，较原模型减少了41.3%和51.8%，满足无人机路面病害检测的实时性要求。更多还原

【Abstract】 To meet the real-time and accuracy requirements of unmanned aerial vehicle(UAV) devices in delecting road disease, a lightweight detection algorithm based on an improved YOLOv8(LHP-YOLO) was proposed. The RG-C2f module replaced the original C2f module to reduce computational redundancy, and a lightweight detection head Detect-T3G was designed to reduce the model’s parameter count. In addition, a content-guided attention fusion mechanism was embedded in the neck network to enhance target detection accuracy in complex backgrounds. Channel-wise knowledge distillation was applied to compensate for the accuracy loss caused by lightweight optimization. Experiments on the road disease dataset RDD 2022 demonstrate that the improved model achieves a 3.4% increase in mAP50 compared to the original model, with parameter count and computational load reaching 1.76×106 and 3.9 GFLOPs, respectively, representing reductions of 41.3% and 51.8% compared to the original model. The improved model meets the real-time requirements for road disease detection UAVs.更多还原