基于深度学习的细粒度车辆识别算法及其优化分析  

Analysis of Fine Granular Vehicle Recognition Algorithm and Its Optimization Based on Deep Learning

作　　者：黄胤铭 HUANG Yinming(School of Computer Science,South China Normal University,Guangdong 510631,China)

机构地区：[1]华南师范大学计算机学院,广东510631

出　　处：《集成电路应用》2023年第3期270-273,共4页Application of IC

摘　　要：阐述神经网络为ResNet-152以及VGG16,采用迁移学习的方式,将Stanford-car196数据集按照80:20分为训练集和测试集,进行训练,最终得到两种神经网络的训练模型。ResNet152和VGG16_dropout的准确度分别为89.80%以及77.54%,使用dropout&Batch-Normalize优化后的VGG16,性能由77.54%增加至82.26%。This paper describes that the neural networks are ResNet-152 and VGG16.The Stanford car 196 dataset is divided into a training set and a test set according to 80:20 by using the transfer learning method.After training,the training models of the two neural networks are finally obtained.The accuracy of ResNet152 and VGG16_dropout was 89.80%and 77.54%,respectively.The performance of VGG16 optimized using dropout&Batch Normalize increased from 77.54%to 82.26%.

关 键 词：细粒度分类 深度学习 车辆分类 DROPOUT Batch-Normalize 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]