文献类型：期刊文献

中文题名：改进YOLOv8n的尘雾环境下目标检测算法

英文题名：Improved YOLOv8n Object Detection Algorithm in Dust and Fog Environment

作者：王子钰[1];张建成[2];刘元盛[2]

第一作者：王子钰

机构：[1]北京联合大学城市轨道交通与物流学院,北京100101;[2]北京联合大学机器人学院,北京100101

第一机构：北京联合大学城市轨道交通与物流学院

年份：2025

期号：6

起止页码：1-7

中文期刊名：汽车技术

外文期刊名：Automobile Technology

收录：;北大核心:【北大核心2023】；

基金：国家自然科学基金项目(62371013,61931012);国家重点研发计划项目(2021YFC3001300);国家重点研发技术项目(2022YFB4601100)。

语种：中文

中文关键词：自动驾驶;目标检测;注意力机制;多尺度特征融合;尘雾环境

外文关键词：Autonomous driving;Object detection;Attention mechanism;Multi-scale feature fusion;Dusty and foggy environment

摘要：针对扬尘、雾霾等恶劣环境下,车辆目标检测中漏检、误检及远小目标检测精度低等问题,提出了EPMYOLOv8的目标检测算法。将高效通道注意力(ECA)模块集成到YOLOv8n算法的C2f模块,使骨干网络更加关注浅层较小的目标特征信息;通过增加目标检测层,并设计多尺度特征融合架构,提高模型目标特征提取能力与检测精度;使用基于最小点距离交并比(MPDIoU)损失作为损失函数,提高检测框回归精度。试验结果表明:EPM-YOLOv8模型检测框查准率达到83.6%,检测精度达到76.8%,对尘雾恶劣环境的检测性能具有一定优越性。
To address the issues of missed detections,false detections and low accuracy in detecting small and distant objects under adverse conditions such as dust and haze,this paper proposes the EPM-YOLOv8 object detection algorithm.The Efficient Channel Attention(ECA)module is integrated into the C2f module of the YOLOv8n algorithm,enabling the backbone network to focus more effectively on shallow and smaller object features.By adding an additional detection layer and designing a multi-dimension feature fusion architecture,the model’s ability to extract target features and its detection accuracy are significantly improved.Furthermore,a loss function based on the Minimum Point Distance Intersection over Union(MPDIoU)is employed to enhance the precision of bounding box regression.Experimental results demonstrate that the EPM-YOLOv8 model achieves a precision ratio of 83.6%and a detection accuracy of 76.8%,exhibiting superior detection performance under challenging conditions such as haze and dust.