Research on detection technology of textile defects based on deep learning
The traditional approach to textile defect detection relies on manual labour,which is subject to a num-ber of limitations in terms of its reliability,efficiency and accuracy.The application of deep learning techniques in automated textile defect detection is discussed,with a particular focus on the use of MLP,CNN,VGG16,and ResNet50 models.The performance of these models on specific datasets was evaluated through a series of met-rics,including accuracy,training loss,confusion matrix,receiver operating characteristic(ROC)curve,and t-dis-tributed stochastic neighbor embedding(t-SNE)clustering effect.The results demonstrate that the ResNet50 model exhibits optimal detection performance on the test set,with an accuracy of 96.0%,an AUC value of over 0.91,a Silhouette Score value of 0.7731,and a Davies-Bouldin Index value of 0.3195.Furthermore,a hierarchi-cal classification method was explored for hundreds of defect types present in the actual production process.The accuracy of defect detection was enhanced to 85.4%through the implementation of a tree-structured classi-fication strategy.Effective technical support is provided to facilitate the automation of textile quality inspection.
deep learningtextile materialdefect detectionimage classificationconvolutional neural network
国家科技期刊平台
NSTL
万方数据
