广西科技大学学报2025，Vol.36Issue(3)：85-91,122,8.DOI:10.16375/j.cnki.cn45-1395/t.2025.03.011
基于改进YOLOv8的复杂路况下的目标识别
Object recognition in complex road conditions based on the improved YOLOv8
摘要
Abstract
Object recognition and detection is the key technology in automatic driving technology,but the existing object recognition algorithm exhibits low detection accuracy in complex road conditions.Therefore,this paper proposed an improvement based on YOLOv8 algorithm.Multi-head self-attention(MHSA)mechanism was introduced into the feature detection layer.Multi-head has the characteristic of selectively paying attention to the area where there are vehicles and pedestrians,and finally captures higher-level semantic features.The region was adapted to complex visual tasks and was better able to capture spatial changes and shape information of the object.Finally,the ablation experiment and comparison experiment results were obtained on the urban road data set.The experimental results show that the improved YOLOv8 algorithm performs better than the original algorithm in complex scenes,with the average precision MAP reaching 93.14%,an increase of 5.29%,demonstrating better object detection performance.关键词
目标识别检测/YOLOv8/复杂路况/多头自注意力(MHSA)机制/可变形卷积Key words
object recognition and detection/YOLOv8/complex road condition/multi-head self-attention(MHSA)mechanism/deformable convolution分类
信息技术与安全科学引用本文复制引用
张成涛,李习刊,徐伟航,王瑞敏..基于改进YOLOv8的复杂路况下的目标识别[J].广西科技大学学报,2025,36(3):85-91,122,8.基金项目
广西科技重大专项(桂科AA22068100) （桂科AA22068100）
广西科技重点研发计划项目(桂科AB24010197)资助 （桂科AB24010197）
