摘要
为提高自动驾驶中的道路目标检测精度,设计了一种基于YOLOv5的道路目标检测模型。该模型在YOLOv5s的网络模型基础上,将原始的初始锚框聚类算法改为K-means++算法来减小随机带来的聚类误差;并在Backbone中SPP模块之前引入SENet注意力机制,以增强道路目标重要特征并抑制一般特征,达到提高检测网络对道路目标的检测能力。在VOC2012改进数据集上训练、测试,基于改进的YOLOv5s的模型比原始YOLOv5s模型平均准确精度提高了2.4%。实验结果表明,改进的YOLOv5s模型能较好地满足道路目标检测的精度要求。
In order to improve the accuracy of road target detection in automatic driving,a road target detection model based on YOLOv5 is proposed.Based on the network model of YOLOv5s,this model changes the original initial anchor frame clustering algorithm to K-means++algorithm to reduce the clustering error caused by randomness.The SENet attention mechanism is introduced before the SPP module in Backbone to enhance the important features of road targets and suppress the general features to improve the detection capability of the detection network for road targets.After training and testing on the VOC2012 improved dataset,the average accuracy of the model based on the improved YOLOv5s is 2.4%,higher than that of the original YOLOv5s model.The experimental results show that the improved YOLOv5s model can better meet the accuracy requirements of road target detection.
作者
吴丽娟
黄尧
关贵明
WU Lijuan;HUANG Yao;GUAN Guiming(College of Physical Science and Technology,Shenyang Normal University,Shenyang 110034,China;Troops 31441 of The Chinese People s Liberation Army,Shenyang 110001,China)
出处
《沈阳师范大学学报（自然科学版）》
CAS
2023年第1期85-91,共7页
Journal of Shenyang Normal University:Natural Science Edition
基金
辽宁省教育厅科研项目(LFW202003)。
