基于改进VGG16网络的小尺寸图像识别研究  

Research on Small-size Image Recognition Based on Improved VGG16 Network

作　　者：陈灵方 张鹏 李昆 杨航 邱媛媛 CHEN Lingfang;ZHANG Peng;LI Kun;YANG Hang;QIU Yuanyuan(Xinjiang Institute of Technology,Aksu 843100,China)

机构地区：[1]新疆理工学院,新疆阿克苏843100

出　　处：《现代信息科技》2024年第23期105-109,共5页Modern Information Technology

摘　　要：在嵌入式系统和边缘计算中,为提高VGG16卷积神经网络对小尺寸图像识别的计算效率,通过调整模型全连接层数量、卷积核数量和使用全局平均池化替代全连接层等方式对VGG16网络进行改进,降低网络模型的可训练参数量。将改进的神经网络模型在图像增强的CIFAR-10数据集上进行训练,训练集达到99%以上的识别准确率,测试集可以达到90%以上的识别准确率,改进后的网络模型参数量较VGG16网络参数量减少了89.04%,验证了改进网络模型的有效性。In embedded systems and edge computing,in order to improve the computational efficiency of the VGG16 Convolutional Neural Networks for small-size image recognition,the VGG16 network is improved by adjusting the number of fully connected layers and the number of convolutional kernels in the model,using global average pooling to replace fully connected layers,and other ways,so as to reduce the number of trainable parameters of the network model.The improved neural network model is trained on the CIFAR-10 dataset with image enhancement.The recognition accuracy of the training set reaches more than 99%,and the recognition accuracy of test set can reach more than 90%.The number of parameters of the improved network model is reduced by 89.04%compared with the VGG16 network,which verifies the effectiveness of the improved network model.

关 键 词：卷积神经网络 VGG16 CIFAR-10数据集 网络轻量化 图像增强 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]