Research On Fabric Defect Detection Algorithm Based On Deep Learning

With the globalization of the economy,the requirements for enterprises are getting higher and higher.In order to keep textile enterprises competitive,the unmanned knitting workshop is undoubtedly an important way to improve the productivity of enterprises.Applying vision technology to workshop production is an important way to achieve unmanned production.Fabric defect detection is an important step in the production of fabrics.In the past,most of the defects were detected by manual inspection.The detection efficiency is low and the error rate is high.This paper mainly uses the deep learning technology to realize the automatic detection of fabric defects,which can liberate labor and achieve workshop automation production.This article focuses on the following aspects:(1)The overall framework of the automatic inspection platform was studied,and the main hardware of the vision system including the camera,lens,and light source were introduced and selected.(2)The characteristics of fabric pattern and background are often very different.In order to achieve more accurate and effective fabric detection,this paper proposes a fabric pattern segmentation algorithm based on improved U-net model,which uses the deeper VGG16 model to replace the features in the original model.Extract the part and add a hollow convolution to enhance the receptive field.After the experiment,the improved algorithm can effectively segment the fabric printing pattern,and the average segmentation accuracy reaches 98.33%.(3)The classification of fabric defects is realized by the deep convolutional neural network VGG16 model.Based on the VGG16 model,the activation function,loss function,gradient descent method,etc.are adjusted.Through multiple comparison experiments,it is verified that the network model can effectively classify fabric defects,and the average classification accuracy rate reaches 98%.(4)Aiming at the problem of fabric defect location,a modified fabric defect detection algorithm based on the original depth learning target detection model faster-RCNN is proposed.The fabric parameters are improved according to the characteristics of the fabric,and the newly trained model parameters are used to extract fabric images and defects.The feature,while using the feature extraction part of the original network,proposes a method of classifying the texture of the fabric first and then detecting the defect.Finally,based on the realization of the defect classification,the precise positioning of the defect position is completed.The average accuracy of the improved algorithm for fabric defect detection is 98.68%,which fully meets the detection accuracy requirements in actual production.