摘要
针对道路场景出现的目标遮挡、重叠以及小目标缺失等检测问题,文中提出一种基于YOLOv8n改进算法的车辆及行人目标检测模型。首先,设计了一种新的金字塔池化层(SPPF-CREP)结构替换原网络的金字塔池化层(SPPF)结构,提高模型在训练和推理阶段的效率和性能;其次,增添了小目标检测头(P_2)来提高网络对小目标的检测能力;然后,将在线卷积重参数化(OREPA)融入到C2f模块中,从而提高在高密度环境下对车辆及行人检测的精确率和效率;最后,采用WIoUv2作为替代损失函数,以实现更高的定位精度。在KITTI车辆检测数据集上的实验结果表明,与原始算法相比,改进算法的检测精确率提升了3.6%,平均精度均值提升了4.2%,证明了其在车辆及行人检测方面具有高效性和优越性。
In view of the object occlusion,overlap,and small object missing in road scene detection,a vehicle and pedestrian object detection model based on the improved YOLOv8n algorithm has been proposed.Firstly,a new pyramid pooling layer structure called SPPF-CREP is designed to replace the original network's pyramid pooling layer structure SPPF,enhancing the model's efficiency and performance during training and inference.Secondly,a small object detection head(P2)is added to improve the network's detection capability for small objects.Then,the online convolutional re-parameterization(OREPA)is integrated into the C2f module to enhance the accuracy rate and efficiency of vehicle and pedestrian detection in high-density environments.Finally,the WIoU-v2 is adopted as an alternative loss function to achieve higher localization accuracy.Experimental results on the KITTI vehicle detection dataset demonstrates that,in comparison with the original algorithm,the accuracy rate of the detection of the improved algorithm is increased by 3.6%,and the mean average precision(mAP)is enhanced by 4.2%.This experiment has proven its efficiency and superiority in the detection of vehicles and pedestrians.
作者
周建新
郝英杰
侯自川
ZHOU Jianxin;HAO Yingjie;HOU Zichuan(College of Electrical Engineering,North China University of Science and Technology,Tangshan 063210,China)
出处
《现代电子技术》
北大核心
2025年第17期35-40,共6页
Modern Electronics Technique
基金
河北省自然科学基金项目(F2018209201)。
作者简介
周建新(1977-),男,河北唐山人,博士研究生,副教授,硕士生导师,研究方向为智能控制理论及应用;通讯作者:郝英杰(2000-),男,河北张家口人,硕士研究生,研究方向为智能控制与模式识别;侯自川(2000-),男,河北保定人,硕士研究生,研究方向为电力设备状态检测与故障诊断。
