基于深度神经网络的蜡染图像鉴别模型研究  

Research on batik image identification model based on deep neural network

作　　者：唐文豪 罗维平 杜焱铭[1] 余中祈 张亚鹏 TANG Wenhao;LUO Weiping;DU Yanming;YU Zhongqi;ZHANG Yapeng(School of Mechanical Engineering and Automation,Wuhan Textile University,Wuhan 430200,China;Hubei Key Laboratory of Digital Textile Equipment,Wuhan Textile University,Wuhan 430200,China)

机构地区：[1]武汉纺织大学机械工程与自动化学院,武汉430200 [2]武汉纺织大学、湖北数字纺织设备重点实验室,武汉430200

出　　处：《纺织工程学报》2025年第2期65-74,共10页JOURNAL OF ADVANCED TEXTILE ENGINEERING

基　　金：国家自然科学基金(62103309);湖北省数字化纺织装备重点实验室开放课题(DTL2022001)。

摘　　要：为了传承蜡染技艺和学术研究,提高蜡染图像归类的准确率,结合印度尼西亚蜡染图像建立了一种基于Inception V3深度神经网络相结合的蜡染图像鉴别模型。为了解决数据集样本数目少的问题,采用风格迁移的方法对数据集进行扩充。对于Inception V3网络,经扩充的数据集比未经扩充的数据集准确率提升了9.56%;在扩充数据集上训练模型,Inception V3的准确率为95.1%,VGG16为76.1%,ResNet50为91.07%,Inception V3模型各类别的精确率和召回率均高于另外两类模型。In order to inherit the batik technique and academic research,and to improve the accuracy of batik image categorization,a batik image identification model based on the combination of Inception V3 deep neural network was established by combining Indonesian batik images.In order to solve the problem of the small number of samples in the dataset,the dataset is expanded by using the method of style migration.For the Inception V3 network,the accuracy of the expanded dataset is improved by 9.56%compared with the unexpanded dataset;training the model on the expanded dataset,the accuracy of Inception V3 is 95.1%,VGG16 is 76.1%,and ResNet50 is 91.07%,and the accuracy and recall of each category of the Inception V3 model are higher than the other two alternative models.

关 键 词：深度学习 蜡染图像 风格迁移 图像鉴别 精确率 召回率 

分 类 号：TS195.644[轻工技术与工程—纺织化学与染整工程]