基于卷积神经网络的服饰纹样风格自动标注

Automatic annotation of clothing pattern style based on convolutional neural network

针对当前服饰纹样的标注多关注纹样本体类别信息,忽略纹样本体属性的问题,对服饰纹样风格,即本体属性进行自动标注。以清代服饰植物纹样为例,在自建的纹样风格数据集的基础上,构建一种基于卷积神经网络的服饰纹样风格自动标注方式。以AlexNet模型结构为基础,提出了Model-1、Model-2等2种纹样自动标注模型,基于Keras框架在自建数据集上进行了对比实验,采用P、R、F_(1)评价指标对模型测试结果进行了量化评估。结果表明,Model-1在P、R、F_(1)3个指标上的值均优于Model-2,最终选择Model-1作为纹样风格自动标注模型。

In view of the fact that the current clothing pattern has focused to the pattern sample body category information,without attention paid to the problem of the pattern sample body attribute,an automatic labeling method based on convolutional neural network was proposed.Based on the AlexNet model structure,two automatic pattern annotation models were proposed:Model-1 and Model-2.Based on the Keras framework,comparative experiments were carried out on self-built data sets,and P,R,and F_(1) were used for evaluation indications,with which the model test results were quantitatively evaluated.The results show that Model-1 is better than Model-2 on the three indicators of P,R and F_(1),with Model-1 finally selected as the pattern style automatic labeling model.