摘要
为了数字化传承与创新传统的蓝印花布纹样,需要将蓝印花布纹样进行分类。为此,提出一种改进的VGGNet卷积神经网络模型的纹样分类方法。首先,采集原始的蓝印花布图案,通过图像增强技术扩充样本,形成训练数据集。其次,改进经典的VGGNet 16卷积神经网络结构,增加卷积组及调整网络参数,增加丢弃层。同时,分析、验证训练优化策略对蓝印花布纹样分类的影响。最后,利用训练集及验证集中的图像样本,通过自动学习获取网络模型参数,得到纹样分类的最佳网络模型并获得较为理想的分类结果。实验结果显示,改进的卷积神经网络模型针对5类蓝印花布纹样进行分类训练,其平均分类准确率达89.73%,为蓝印花布纹样的继承和创新研究提供了新思路。
To inherit and innovate the traditional blue calico vein pattern,it is necessary to classify the vein pattern of blue calico.Therefore,a method of vein pattern classification based on the improved convolution neural network model of VGGNet is proposed.Firstly,the original blue calico vein pattern images are captured,and the image samples are expanded by utilizing image enhancement technology to construct the training data set.Secondly,the classic VGGNet 16 convolutional neural network structure is improved,the convolution group is added and the network parameters are adjusted,and the Dropout layer is added.At the same time,this paper analyzes and verifies the effect of training optimization strategy on the classification of the vein pattern for blue calico.Finally,using the image samples in the training set and validation set,the network model parameters are acquired through automatic learning,the optimal network model of vein pattern classification is acquired and the ideal classification results are obtained.The experimental results show that the improved convolutional neural network model can classify 5 types of vein patterns of blue calico,and the average classification accuracy is 89.73%,which provides a new idea for the inheritance and innovation of blue calico vein patterns.
作者
贾小军
邓洪涛
刘子豪
叶利华
JIA Xiao-jun;DENG Hong-tao;LIU Zi-hao;YE Li-hua(College of Mathematics,Physics and Information Engineering,Jiaxing University,Jiaxing 314001,China;College of Design,Jiaxing University,Jiaxing 314001,China)
出处
《光电子.激光》
EI
CAS
CSCD
北大核心
2019年第8期867-875,共9页
Journal of Optoelectronics·Laser
基金
嘉兴市科技局公益性研究计划项目(2018AY11008)资助项目
关键词
蓝印花布纹样
VGGNet
卷积神经网络
训练
分类
blue calico vein pattern
VGGNet
convolutional neural network
training
classification