改进Smaller VGGNet的细粒度汽车图像分类算法  

Fine-grained vehicle image classification algorithm based on improved Smaller VGGNet

作　　者：骆绍烨 林子洋 龙秋华 林俊武 黄佳燕 LUO Shaoye;LIN Ziyang;LONG Qiuhua;LIN Junwu;HUANG Jiayan(New Engineering Industry College,Putian University,Putian 351100,China)

机构地区：[1]莆田学院新工科产业学院,福建莆田351100

出　　处：《延边大学学报（自然科学版）》2024年第3期51-55,共5页Journal of Yanbian University（Natural Science Edition）

基　　金：福建省中青年教师教育科研项目(JAT210410,JAT220829);福建省自然基金面上项目(2022J011170);莆田学院引进人才启动项目(2022058);莆田市科技特派员项目(F2020KTP018);莆田学院横向项目(2021AHX154(L),2023AHX036(L));莆田市科技计划项目(2024GZ2001PTXY18)。

摘　　要：为了提高细粒度车辆图像分类的准确度,提出了一种基于改进Smaller VGGNet的细粒度汽车图像分类算法模型.首先,重新调整了神经网络结构,以此提高训练的稳定性和分类的准确率;其次,替换特征的填充方式,以更好地捕捉局部信息,减缓过拟合现象;最后,利用ELU替换原有的ReLU激活函数,以加快模型收敛速度.在公开数据集上,将改进后模型与现有的6种图像分类算法进行对此.结果表明,改进后模型的细粒度汽车图像分类效果显著,且在训练过程表现出更好的稳定性.研究结果可为同类问题解决提供借鉴.To improve the ability of vehicle category information recognition and analysis in the field of intelligent transportation,a fine-grained vehicle image classification algorithm model based on improved SmallerVGGNet is proposed.The model first readjusted the neural network structure to improve training stability and classification accuracy.Secondly,replace the filling method of features to better capture local information and alleviate overfitting.Finally,ELU is used to replace the original ReLU activation function to accelerate the convergence speed of the model.On a publicly available dataset,the improved model with six existing image classification algorithms were compared,experimental results show that the improved algorithm outperforms significantly the compared algorithms and exhibits better stability during the training process.

关 键 词：智能交通 汽车图像分类 细粒度 Smaller VGGNet 神经网络 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]