基于YOLOv5改进的雾天行人与车辆检测算法  

Improved Foggy Pedestrian and Vehicle Detection Algorithm Based on YOLOv5

作　　者：苏彤 王颖[1,2] 邓启扬[1] 李兆彬 Su Tong;Wang Ying;Deng Qiyang;Li Zhaobin(School of Information Engineering,Nanchang Institute of Technology,Nanchang 330000,China;Jiangxi Province Key Laboratory of WaterInformation Cooperative Sensing and Intelligent Processing(Nanchang Institute of Technology),Nanchang 330000,China)

机构地区：[1]南昌工程学院信息工程学院,江西南昌330000 [2]江西省水信息协同感知与智能处理重点实验室(南昌工程学院),江西南昌330000

出　　处：《系统仿真学报》2024年第10期2413-2422,共10页Journal of System Simulation

基　　金：江西省科技厅重点研发项目(20161BBG70055)。

摘　　要：由于在恶劣天气下汽车对环境的感知能力差,且对动态目标的检测能力有极大的影响,使得基于深度学习的目标检测网络在雾天行人车辆检测中出现精度低、鲁棒性差等问题。本文提出一种基于DehazeNet去雾算法与改进YOLOv5算法相结合的雾天检测方法—YOLOv5-SGE检测网络。通过取消初始锚框,实现锚框自适应计算,生成适合当前数据集的锚框;在特征提取模块加入三维权重注意力机制,使网络可以快速捕捉到感兴趣区域,抑制无用信息;使用轻量级卷积GSConv代替融合模块的标准卷积,弥补语义信息损失,减轻模型的复杂度;使用EIoU损失函数替换YOLOv5网络原损失函数CIoU,加快网络收敛速度。实验结果表明:所提算法具有较高的检测精度,mAP达到84%,验证了算法的有效性。Due to the poor environment perception of car in bad weather,the detection ability on dynamic targets is significantly reduced,and thus the problems such as low accuracy and poor robustness of the deep learning-based target detection network will occur when detecting pedestrians and vehicles in foggy days.A YOLOv5-SGE foggy detection network is proposed on the basis of the combination of image dehazing DehazeNet and the improved YOLOv5.The adaptive calculation of anchor frame is realized by canceling the initial anchor frame of YOLOv5,and the anchor frame suitable for the current dataset is generated.A three-dimensional weighted attention mechanism is added to the feature extraction module,so that the network can quickly capture the region of interest and suppress the useless information.Instead of the standard convolution of the fusion module the lightweight convolution GSConv is used to compensate for the loss of semantic information and reduce the complexity of tmodel.EIoU loss function is used to replace the original loss function CIoU of network to accelerate the network convergence speed.The experimental results show that,compared with the other four algorithms,the proposed model has a higher detection accuracy and mAP reaches 84%,which verifies the effectiveness of the proposed algorithm.

关 键 词：自动驾驶 图像去雾 YOLOv5 轻量化卷积 损失函数 目标检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]