A Fabric Defect Detection System Based On Autoencoder Network

The detection of fabric defects is an important part in the production of the textile industry,which can control the quality of fabrics and ensure product quality.Using computer vision system to monitor fabric in real time can effectively avoid the disadvantage of high labor cost and low detection accuracy of manual fabric inspection.Traditional visual fabric defect detection algorithms require elaborate design of feature extractors to extract image’s color,edge and other characteristics,and often can’t effectively detect defects which have complex shape or with small area.In recent years,convolutional neural networks that can automatically extract features have had significant application effects in computer vision.In this paper,the convolution network is applied to the detection of fabric defects.According to the actual needs of enterprises,I design a detection algorithm and construct an effective fabric defect detection system.First,based on the actual industrial environment,using image acquisition equipment to obtain clear and complete fabric images,a fabric defect database is build.Consider about the characteristics of fabric images,an effective deep convolutional network for binary classification is designed.The Focal loss function was used to solve the problem of data imbalance.Finally the convolution model has achieved excellent classification results.Aiming at the problems that deep learning requires a large number of labeled samples for model training but the data is difficult to obtain,the strict detection rate on defect free samples,an unsupervised fabric defect detection method based on fully convolution autoencoder network is proposed.The model’s training process only need defect free fabric images which are easily obtained.What’s more,This method can adjust the threshold to not misjudge the defect free samples.In order to improve the speed of detection,the separable convolution module is adopted and the depth of the feature smap is reduced to realize the real-time detection of the fabric.In addition,the algorithm can not only determine the weather the fabric have defects,but also mark the location of defects at the pixel level,so the model is more robust.Finally,a fabric defect detection system which combines the software platform and the hardware equipment is constructed.The actual operation of the system in the factory can fully meet the real-time,stability and accuracy requirements of defect dection.