基于混合数据增强的布匹推荐系统  

Cloth recommendation system based on mixed data enhancement

作　　者：赵洁 陈禹健 王光夫 ZHAO Jie;CHEN Yujian;WANG Guangfu(School of Computer and Information Engineering,Tianjin Chengjian University,Tianjin 300384,China;Tianjin Surveillance Technology Company Limited,Tianjin 300392,China)

机构地区：[1]天津城建大学计算机与信息工程学院,天津300384 [2]天津瑟威兰斯科技有限公司,天津300392

出　　处：《通信与信息技术》2023年第5期7-12,18,共7页Communication & Information Technology

基　　金：国家级大学生创新创业资助项目(项目编号:201810792026);教育部产学合作协同育人资助项目(项目编号:202002168032);天津城建大学教育教学改革与研究重点资助项目(项目编号:JG-ZD-22038,项目编号:JG-ZD-22035)。

摘　　要：为解决传统图像检索算法对图像特征复杂度要求高以及深度学习中数据集单一的问题,提出了一种基于混合数据增强的布匹推荐系统,提高布匹花纹的识别准确率。首先使用传统图像增强方式和生成式对抗网络实现对原始数据的图像增强,然后构建卷积神经网络分类模型实现对布匹花纹的分类。实验结果表明,匹配模型可以对6044种花纹正确分类并且测试花纹的匹配置信度在90%以上,相比原始数据集训练得到的匹配模型在置信度上有较大的提高。In order to solve the problem of feature complexity required by traditional image retrieval algorithm and single data set in deep learning,this paper proposes a cloth recommendation system based on mixed data enhancement for improving the recognition accuracy of cloth pattern.Firstly,traditional image enhancement method and generative antagonism network are used to realize image enhancement of the original data.Then,the classification model of convolutional neural network is constructed to realize the classification of cloth patterns.The experimental results show that the matching model can classify 6,044 patterns and test the matching confidence of patterns above 90%。Compared with the original data set,the confidence of the matching model is greatly improved.

关 键 词：布匹推荐 图像增强 卷积神经网络 生成式对抗网络 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]