基于标签嵌入方法的纺织品瑕疵识别网络  

Textile defect recognition network based on label embedding

作　　者：刘颖[1] 姜威 李冠典 陈磊[1] 赵爽[1] LIU Ying;JIANG Wei;LI Guandian;CHEN Lei;ZHAO Shuang(College of Electronic Information Engineering,Changchun University of Science and Technology,Changchun 130000,China)

机构地区：[1]长春理工大学电子信息工程学院,吉林长春130000

出　　处：《光学精密工程》2023年第10期1563-1579,共17页Optics and Precision Engineering

基　　金：吉林省自然科学基金资助项目(No.20210101182JC)

摘　　要：卷积神经网络(Convolutional Neural Network,CNN)可用于工业生产环境下的纺织品疵点的鉴别与分类。针对实际场景下的纺织品瑕疵存在瑕疵类型视觉区分度小和实际数据样本采集时的瑕疵类别不平衡问题,本文提出了基于标签嵌入方法的纺织品瑕疵识别网络(Textile Defect Recognition Network Based on Label Embedding,TDRNet)。首先,算法调整了基础骨干网络的结构,从而提高模型的分类精度;接着算法还设计了标签嵌入模块(Label Embedded Module,LEM),并使用该模块来生成模型的类别权重偏移;然后,本文提出了分布感知损失函数(Distribution Perception Loss,DP Loss)调整算法的类别分布,从而减小同类瑕疵特征的类内距并增大异类瑕疵特征的类间距;最后,本文引入了Seesaw Loss损失函数,通过抑制少数类别的负样本梯度并提高对误分类时的样本损失来动态平衡模型训练过程中在不同样本下的更新梯度,以缓解少数类别的误分类率。在自制的“广东智能制造”布匹瑕疵分类数据集中,本文提出的框架在粗粒度分类和细粒度分类两个任务上的top1错误率可达16.35%和17.12%,而top5错误率在细粒度分类任务上低至5.20%。与其他分类模型相比,TDRNet在对比实验中取得了最优的结果。此外,TDRNet与近5年经典的细粒度分类模型进行了比较,并取得了SOTA结果,这充分表明了TDRNet的先进性。A convolutional neural network(CNN)can be used in the industrial production environment to identify and classify textile defects.To overcome the problems in the visual discrimination of small defect types and imbalance of textile defect categories in actual scenes,a textile defect recognition network(TDRNet)based on label embedding method is proposed.First,the backbone structure is adjusted to improve the classification accuracy of the model.Then,a label embedded module(LEM)is constructed to generate the category weight offset of the model.Subsequently,a distribution perception loss function(DP loss)is proposed to adjust the class distribution of the algorithm;this reduces the distance of homogenous defect features and increases the distance of heterogeneous features.Finally,the seesaw loss function is introduced to dynamically balance the gradient update for different samples during the model training process by suppressing the negative sample gradient of a few categories and increasing the sample loss during misclassification,thereby alleviating the misclassification rate of a few categories.In the self-made"Guangdong intelligent manufacturing"cloth defect classification dataset,the top1 error rate of our framework for rough-grained and fine-grained classifications reached 16.35%and 17.12%,respectively,whereas the top5 error rate of fine-grained classification was as low as 5.20%.Compared with other classification models,TDRNet achieved the best results.In addition,TDRNet was compared with the classical fine-grained classification model in recent five years and achieved state-of-the-art(SOTA)performance,fully demonstrating the enhancements provided.

关 键 词：卷积神经网络 纺织品瑕疵识别 标签嵌入 分布感知损失 Seesaw损失 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]