计算机技术与发展2025，Vol.35Issue(9)：30-37,8.DOI:10.20165/j.cnki.ISSN1673-629X.2025.0113
MDSD-YOLO:一种复杂道路场景目标检测方法
MDSD-YOLO:An Object Detection Method in Complex Road Scenes
摘要
Abstract
Object detection in complex road scenarios suffers from severely insufficient performance of existing models due to irregularly shaped detection targets,low accuracy for distant targets,and slow inference speeds.To address this issue,an improved algorithm named MDSD-YOLO based on YOLOv8 is proposed.In the Head part,an MD-C2f module,which integrates MLCA attention and DCNv4,is designed to adapt to irregular targets and enhance the detection accuracy of small targets.After each detection head,the SEAM attention module is introduced to optimize the detection accuracy when targets occlude each other.In the backbone network,a D-C2f module based on DualConv lightweight convolution is designed,which not only improves the accuracy but also reduces the number of model pa-rameters,significantly enhancing the real-time performance of the model.Experimental results show that the improved model performs excellently on the SODA10M and KITTI datasets.The mAP50 reaches 67.0％and 94.4％respectively,an increase of 8.7 percentage points and 5.3 percentage points compared to the YOLOv8n baseline model;the mAP50:95 reaches 44.8％and 70.5％respectively,5.5 percentage points and 4.2 percentage points higher than that of the baseline model.Experiments demonstrate that the proposed model has fast inference speed,high detection accuracy,and shows effectiveness and superiority in object detection in complex road scenarios.关键词
YOLOv8/复杂道路场景/可变形卷积/注意力机制/目标检测/轻量化Key words
YOLOv8/complex road scenes/deformable convolution/attention mechanisms/object detection/lightweight分类
信息技术与安全科学引用本文复制引用
赵龙阳,李天豪,张会兵,刘琦,孟瑞敏..MDSD-YOLO:一种复杂道路场景目标检测方法[J].计算机技术与发展,2025,35(9):30-37,8.基金项目
国家自然科学基金项目(62267003) （62267003）
广东省农业科学院协同创新中心项目(XTXM202203) （XTXM202203）
