基于改进YOLOv8的路面缺陷检测模型  

Pavement Defect Detection Model Based on Improved YOLOv8

作　　者：黄配乐 王生怀[1] 陈晓辉 王宸[1] 张伟[1] Huang Peile;Wang Shenghuai;Chen Xiaohui;Wang Chen;Zhang Wei(Hubei University of Automotive Technology,Shiyan 442002,China)

机构地区：[1]湖北汽车工业学院,湖北十堰442002

出　　处：《湖北汽车工业学院学报》2025年第1期15-19,共5页Journal of Hubei University Of Automotive Technology

基　　金：国家自然科学基金(51675167);湖北省重点研发计划项目(2021BAA056);湖北省高等学校优秀中青年科技创新团队计划项目(T2020018);湖北省自然科学基金(2020CFB755);湖北省教育厅科研项目(Q20191801)。

摘　　要：基于改进YOLOv8提出一种路面缺陷检测模型。设计了C2f_SimAM模块,采用GhostNet替换YOLOv8的骨干特征提取网络,引入SimSPPF模块,将MPDIOU损失作为边界框回归损失。结果表明:与YOLOv8n模型相比,改进模型的平均精度提高了3.6%,参数量降低了10%,FPS达到了85,同时满足实时检测要求。An improved YOLOv8-based pavement defect detection model was proposed.The C2f_SimAM module was designed,and the backbone feature extraction network of YOLOv8 was replaced with GhostNet.Moreover,the SimSPPF module was introduced,and the MPDIOU loss was regarded as the bounding box regression loss.The results indicate that compared to the YOLOv8n model,the improved model achieves a 3.6% increase in average precision,a 10% reduction in the number of parameters,and an FPS of 85,meeting the requirements for real-time detection.

关 键 词：路面缺陷检测 YOLOv8 C2f_SimAM GhostNet 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] U418[自动化与计算机技术—计算机科学与技术]