基于YOLOv5的障碍物图像检测算法研究  

Obstacle Image Detection Algorithm Based on Yolov5

作　　者：张霞[1] 王月诚 ZHANG Xia;WANG Yue-cheng(College of Intelligent Manufacturing,Longdong University,Qingyang 745000,Gansu)

机构地区：[1]陇东学院智能制造学院,甘肃庆阳745000

出　　处：《陇东学院学报》2025年第2期48-54,共7页Journal of Longdong University

基　　金：甘肃省教育科技创新项目“基于脑电信号的癫痫自动检测及预测研究”(2023A-114);陇东学院校企合作横向科研项目“老旧小区安防系统改造设计安装与调试”(HXZK2426);陇东学院校企合作横向科研项目“工地空气质量检测系统的设计”(HXZK2458)。

摘　　要：无人驾驶传统目标检测算法中,进行障碍物检测的时候会出现侯选区域生成速度慢和检测准确率低的问题,提出了一种改进的YOLOv5障碍物检测算法。首先在YOLOv5算法中加入了通道注意力机制SE模块,并用FReLU激活函数代替原激活函数Leaky ReLU,采用nuscenes数据集对检测算法进行验证。实验表明,在YOLOv5算法中只加入通道注意力机制SE模块时,其检测速度提高了0.05ms,检测精准度提高了4.18%。加入通道注意力机制SE模块和FReLU激活函数,检测速度降低了0.2ms,精准度却提高了9.05%。In the traditional object detection algorithms used in autonomous driving,there are issues of slow candidate region generation and low detection accuracy.To address these issues,an improved YOLOv5 obstacle detection algorithm is proposed.Firstly,the channel attention mechanism SE module is incorporated into the YOLOv5 algorithm,and the activation function Leaky ReLU is replaced with the FReLU activation function.The detection algorithm is validated using the nuscenes dataset.The experiments show that when only the channel attention mechanism SE module is added to the YOLOv5 algorithm,the detection speed improves by 0.05ms and the detection accuracy increases by 4.18%.When both the channel attention mechanism SE module and the FReLU activation function are added,the detection speed decreases by 0.2ms,but the accuracy improves by 9.05%.

关 键 词：深度学习 无人驾驶 障碍物检测 YOLOv5 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]