基于卷积神经网络的蓝印花布纹样基元分类  

Elements classification of vein patterns using convolutional neural networks for blue calico

作　　者：贾小军[1] 叶利华[1] 邓洪涛[2] 刘子豪 陆锋杰 JIA Xiaojun;YE Lihua;DENG Hongtao;LIU Zihao;LU Fengjie(College of Mathematics,Physics and Information Engineering,Jiaxing University,Jiaxing,Zhejiang 314001,China;College of Design,Jiaxing University,Jiaxing,Zhejiang 314001,China;Zhejiang Hanpu Power Technology Co.,Ltd.,Jiaxing,Zhejiang 314300,China)

机构地区：[1]嘉兴学院数理与信息工程学院,浙江嘉兴314001 [2]嘉兴学院设计学院,浙江嘉兴314001 [3]浙江涵普电力科技有限公司,浙江嘉兴314300

出　　处：《纺织学报》2020年第1期110-117,共8页Journal of Textile Research

基　　金：浙江省科技计划公益技术研究项目(GG20F010032);嘉兴市公益性研究计划项目(2018AY11008)

摘　　要：为更好地数字化传承与创新传统的蓝印花布纹样,并能单独提取构成纹样的图案基元并进行分类,提出一种基于卷积神经网络的纹样基元分类方法。首先,对采集的128张蓝印花布图像进行纹样基元提取,形成图像样本库,共21212张。其次,从库中随机选取80%的图像样本作为训练集,20%作为测试集,利用5×5卷积核对训练样本进行卷积操作,将得到的特征图进行池化。通过3层卷积、3层池化及2层全连接层计算后,利用Softmax分类器得到12种分类结果。最后,通过基元样本的学习获取最佳网络模型参数,并取得较理想的分类结果。结果表明:提出的卷积神经网络模型对12种纹样基元的平均分类准确率达99.61%,检测平均准确率达98.5%,为蓝印花布纹样的研究提供了新思路。To inherit and innovate the traditional blue calico vein patterns making use of the digital technology,the image elements constituting blue calico vein patterns were extracted to form a structured database for vein pattern elements.A classification method based on convolutional neural networks was proposed.The pattern elements were firstly extracted from the captured 128 images of blue calico to form an image sample database with a total of 21212 images.Secondly,80%of image samples in the database were randomly selected as the training set and the rest 20%as the testing set.The training samples were convoluted by a 5×5 convolutional kernel size,and the obtained feature maps were pooled.After computing through 3 convoluting layers,3 pooling layers and 2 full connection layers,12 classification types were obtained by using Softmax classifier.Eventually,the optimal network model parameters were acquired and ideal classification results were obtained through deep learning of the image element samples.The experimental results show that the 12 types of vein pattern elements of blue calico,produced by the convolutional neural networks model,are proved to be with an average accuracy of 99.61%,and detection accuracy of 98.5%.This work provides new ideas for studying blue calico vein patterns.

关 键 词：蓝印花布 纹样基元分类 卷积神经网络 数字纺织 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]