基于改进YOLOv5轻量化的车辆目标检测算法
VEHICLE TARGET DETECTION ALGORITHM BASED ON IMPROVED YOLOV5 LIGHTWEIGHT
田栋 1魏霞 1袁杰1
作者信息
- 1. 新疆大学电气工程学院 新疆 乌鲁木齐 830047
- 折叠
摘要
无人驾驶汽车在近年来取得了巨大的进展和突破,环境感知技术作为无人驾驶汽车安全行驶的重要前提,则需要在行驶的过程中提前检测其周围环境,并快速且准确地检测出周围目标.基于此问题,提出基于改进YOLOv5 的目标检测算法.将EfficientNetV2 作为YOLOv5 算法的主干特征提取网络;为了提高算法的收敛性,引入MetaAconC激活函数,并在Head中融合BiFPN,增加图像特征融合的多样性,不仅使算法模型减小了39%,更在精度上也有一定的提高.通过实验验证,相比YOLOv5 原方法,该算法在保证目标检测实时性的同时具有更高的检测精度,且设备兼容性更好.
Abstract
Driverless cars have made tremendous progress and breakthroughs in recent years.As an important prerequisite for driverless cars to drive safely,environmental perception technology needs to detect their surroundings in advance during driving,and quickly and accurately detect the surroundings target.Based on this problem,this paper proposes a target detection algorithm based on improved YOLOv5.EfficientNetV2 was used as the backbone feature extraction network of the YOLOv5 algorithm.In order to improve the convergence of the algorithm,the MetaAconC activation function was introduced,and BiFPN was integrated in the Head,which increased the diversity of image feature fusion,reduced the algorithm model by 39%,and there was also a certain improvement in accuracy.Through experimental verification,compared with the original method of YOLOv5,this algorithm has higher detection accuracy while ensuring real-time target detection,and has better equipment compatibility.
关键词
YOLOv5/MetaAconC/轻量化/特征融合/BiFPNKey words
YOLOv5/MetaAconC/Lightweight/Feature fusion/BiFPN引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据