基于迁移学习的VGG-16网络芯片图像分类  

Image classification of migration learning chip based on VGG-16 network

作　　者：马俊 张荣福[1] 郭天茹 张喆嫣 李卿 王蓉 李子莹 MA Jun;ZHANG Rongfu;GUO Tianru;ZHANG Zheyan;LI Qing;WANG Rong;LI Ziying(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《光学仪器》2020年第3期21-27,共7页Optical Instruments

摘　　要：针对芯片图像分类过程中图像数量过少、需要大量人工标注以及效率低的问题,提出一种基于迁移学习的VGG-16网络芯片图像分类方法。该方法通过VGG-16网络直接从原始像素中自动学习图像特征,有效减少人工标注的成本,同时对比了VGG-16网络模型和基于迁移学习的VGG-16网络模型的准确率及其混淆矩阵。实验结果表明,所提出的基于迁移学习的VGG-16网络模型对芯片图像分类效果要优于原VGG-16网络模型。To solve the problems of lack of images,too much manual marking and low efficiency in the process of chip image classification,a VGG-16 network chip image classification method based on migration learning is proposed.This method is based on the VGG-16 automatic learning in network and can extract directly from the original pixel image characteristics,effectively reducing the cost of manual annotation.In comparison with VGG-16 network model and VGG-16 network model based on transfer learning accuracy and confusion matrix,the experiment results show that the proposed VGG-16 network model based on the migration study on chip image classification effect is better than the original VGG-16 network model.

关 键 词：图像分类 卷积神经网络 迁移学习 VGG-16 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]