Research On Fabric Surface Defect Detection And Classification

With the increasingly fierce trade competition in the textile industry,it has become a key research direction to strictly control the quality of textile and reduce testing costs.At present,the detection of fabric defects mainly depends on manual detection in China.This method has low accuracy and high cost,thus automatic detection has become an important way to replace manual detection.In this paper,the convolutional neural network is used to research the detection and classification algorithms of fabric surface defects.The improved AlexNet,the improved VGG16 and the transfer learning method are used to classify fabric defects respectively.The algorithm of Faster RCNN combined with residual network and the algorithm of SSD are used to detect fabric defects.The research contents are as follows:(1)The AlexNet and VGG16 are improved by researching the classification of fabric defects.In the improved AlexNet,the batch normalization layer is used to replace the local response normalization layer,and the effectiveness of the improvement is verified by experiments.In the improved VGG16,a batch normalization layer is added between the pooling layer and the convolution layer,and the classification accuracy is 98.26% when the stochastic gradient descent method is used.(2)The fabric defects classification algorithm based on transfer learning is adopted.The structure and parameters of the pre-trained Inception V3 are transferred by using modelbased transfer learning.Then,the network parameters are fine-tuned by training the fabric defects dataset.The experiments show that this method can reduce the training time greatly and require less hardware equipment.(3)The fabric defects are detected by using the method that combining Faster RCNN and residual network.The deep residual network with better feature extraction effect is adopted based on the Faster RCNN.First,the residual network is used for the feature extraction of fabric.Then,region proposal network and Fast RCNN detection network are used to detect the defects of fabric.The detection effect is compared when Faster RCNN is combined with VGG16 and ResNet101 respectively.The influence of different parameters on the results is discussed.The detection accuracy is 99.6%.(4)In order to achieve the real-time detection of fabric defects,the detection algorithm based on SSD is adopted.The algorithm first uses deep convolutional neural network to extract image features,and then detects the multi-scale targets at each position of feature maps with different scales.The experimental results show that the algorithm has good realtime performance and only needs about 35 ms to detect a single image.There are 32 figures,11 tables and 64 references in this paper.