FDAT:基于AlexNet迁移学习的纺织物疵点分类方法

FDAT:Fabric Defect Classification Model Based on AlexNet Using Transfer Learning

针对现有的纺织产品疵点分类方法数据集小,网络训练耗时较长以及准确率较低等问题,论文提出了一种使用迁移学习基于ALexNet模型的纺织产品疵点分类算法(Fabric Defect Classification Model based on AlexNet using Transfer Learning,FDAT),首先,针对纺织产品疵点数据集数据量少的问题,通过基于大型数据集训练得到模型训练参数权重,利用迁移学习方法构建基于AlexNet的纺织产品疵点分类方法;其次,对输入纺织产品疵点数据进行特征提取,使用softmax分类器针对特征提取结果进行分类;最后,在TILDA纺织产品疵点数据集上进行了计算机模拟实验,实验结果表明,提出的FDAT模型对比传统小波变换算法,人工神经网络,DenseNet,ResNet以及Xception,可以有效地解决小样本分类问题,提高算法的准确率的同时,缩短网络分类耗时。

For the problems of the existing textile product defect classification methods such as small data sets,long network training time and low accuracy,this paper proposes the fabric defect classification model based on AlexNet using transfer learning(short of FDAT).First of all,for the problem of the small amount of data in the textile product defect data set,the model training parameter weights are obtained through training based on large data sets,and the transfer learning method is used to construct a textile product defect classification method based on AlexNet.Then,perform feature extraction on the input fabric defect data and use the softmax classifier to classify the feature extraction results.Finally,a computer simulation experiment is carried out on the TILDA fabric defect data set.The experimental results show that the proposed FDAT can effectively solve the small sample classification problem and improve the algorithm's performance compared with the traditional wavelet transform algorithm,artificial neural network,DenseNet,ResNet and Xception.It can improve the accuracy of the algorithm and shorten the time-consuming network classification.