基于迁移学习的红外图像分类  

Infrared image classification based on transferring learning

作　　者：邓玲 张军[1] 周宽 DENG Ling;ZHANGJun;ZHOU Kuan(School of Electronic Engineering,Tianjin University of Technology and Education,Tianjin 300222,China)

机构地区：[1]天津职业技术师范大学电子工程学院,天津300222

出　　处：《天津职业技术师范大学学报》2020年第3期23-29,共7页Journal of Tianjin University of Technology and Education

基　　金：天津市科委项目(14JCTPJC00537).

摘　　要：针对红外目标分类的问题,采用将Inception-v3模型进行迁移并用迭代法对网络结构进行改进的方法,对适用于红外目标分类的神经网络进行了研究。迁移后的模型经过训练能对红外图像的特征进行学习和表达,训练后的卷积神经网络模型能对不同数据格式的红外图像进行分类。实验结果表明,该迁移模型比已有的红外图像分类模型鲁棒性更好,红外图像分类准确率达98%以上,并且能够对不同数据格式的红外目标图像进行分类。To deal with the problem of infrared image target classification,the Inception-v3 model was transferred and iterative method was used to improve the network structure,and the neural network suitable for infrared target classification was examined.The transferred model can learn and express the features of infrared images after training,and the trained convolutional neural network model can classify infrared images in different data formats.The experimental results show that the transferred model is more robust than the existing infrared image classification model with the classification accuracy of infrared image higher than 98%,and it can classify infrared target images in different data formats.

关 键 词：红外图像分类 迁移学习 卷积神经网络 Inception-v3模型 

分 类 号：TN219[电子电信—物理电子学] TP391.41[自动化与计算机技术—计算机应用技术]