Fabric Defect Detection Method Based On Deep Convolution Neural Network

The textile industry has an important proportion in the development of modern national economy.Improving the testing level of textile quality is of great significance for promoting the transformation and upgrading of the textile industry and creating new competitive advantages.Fabric defect detection is an important part of fabric quality control.Automated inspection of fabric defects has become one of the keys to improve the quality of textile production.But the fabric defect detection is still dependent on the traditional manual detection,time-consuming and difficult to ensure that the detection precision is difficult to guarantee,seriously affecting the quality of fabric products and the efficiency of mass production of fabric.Therefore,improving the automation and intelligence level of textile equipment has become an inevitable trend in the development of textile industry.In recent years,the method of computer vision detection based on depth learning has been widely used in the field of image recognition,which shows obvious advantages over traditional algorithms.Therefore,this paper mainly studies the image quality detection of deep learning in textile industry.In order to improve the precision of the defect detection of complex texture fabric,the application value of deep convolution neural network in texture analysis is analyzed,and two different deep neural network are proposed to identify and classify the complex texture fabric defects.The main contributions of this paper include the following aspects:(1)Aiming at the problems of different defects in fabric detection and the existence of texture background interference,a fabric defect location framework based on Faster R-CNN is proposed to locate and identify fabric defects quickly and accurately.Unlike traditional fabric detection methods,the model uses RPN network to quickly calculate the candidate area of fabric defects,and then uses softmax classifier to classify the defects,thus realizing the on-line detection of fabric defects at end to end.(2)Aiming at the gradient dispersion in depth neural network with the increase of depth,a new method of fabric defect detection based on improved deep residual network Cross Net is proposed in this paper.The experimental results show that the model notonly solves the problem of gradient dispersion in depth neural network with increasing depth,but also achieves more than 97% detection precision on the fabric image database.(3)Aiming at the shortage of fabric defect database and small sample size,a network training optimization strategy based on transfer learning is proposed.First of all,the Cross Net network model is optimized by using the learning strategy of pet and line,and then the fabric defect samples are used to improve the training efficiency of the network model.