广西科学院学报2025，Vol.41Issue(1)：33-44,12.DOI:10.13657/j.cnki.gxkxyxb.20250423.001
基于YOLOv8-MCMA模型的道路缺陷检测应用研究
Application Research of Road Defect Detection Based on YOLOv8-MCMA Model
摘要
Abstract
Road defects have multi-scale characteristics,resulting in low detection accuracy.In order to im-prove this problem,this article proposes a lightweight multi-scale convolutional mobile attention model for road defect detection(YOLOv8 Multi-scale Convolutional Mobile Attention,YOLOv8-MCMA).Firstly,the MobileViT network can make the model maintain a high recognition accuracy while reducing the number of parameters.Secondly,the Content-Aware Reassembly of Features(CARAFE)module is used as the up-sam-pling module to improve the detection ability of small cracks.Thirdly,a Multi-scale Inverted Residual Atten-tion(MIRA)module is designed to enhance the sensitivity of the model to multi-scale features.Finally,the ordinary convolution of the neck is replaced by an Alterable Kernel Convolution(AKConv)to better capture irregular crack information,thereby reducing the detection error.The experimental results show that com-pared with the YOLOv8n model,the average accuracy@0.5(mAP@0.5)of the proposed model on Road Damage Detection Dataset,RDD2022_China and Crack-forest Dataset is increased by 3.7%,1.4%and 2.6%respectively,and the parameter amount is reduced by 23.3%.Compared with other models,this model shows significant advantages and has strong adaptability to multi-scale road defects.关键词
计算机视觉/目标检测/道路缺陷检测/MobileViT网络/MIRA模块/YOLOv8-MCMA模型Key words
computer vision/object detection/road defect detection/MobileViT network/MIRA module/YOLOv8-MCMA model分类
计算机与自动化引用本文复制引用
徐克圣,孙蓉..基于YOLOv8-MCMA模型的道路缺陷检测应用研究[J].广西科学院学报,2025,41(1):33-44,12.基金项目
辽宁省教育厅科研经费项目(LJKMZ20220838)资助. （LJKMZ20220838）
