Research Of Fabric Defect Detection Based On Deep Learning

Textile industry plays a very important role in the national economy.Through automatic detection of fabric defects,fabric with quality problems can be found,and timely repair and prevention can be carried out to reduce the deterioration of fabric quality caused by various reasons.Deep learning algorithm has the advantages of fast detection speed,high accuracy and strong robustness,which has become the development trend of fabric defect detection algorithm at present.However,when using deep-learning-based convolutional neural network to detect fabric defects,the network pays little attention to the spatial information and expression ability of the feature map in the convolutional layer,which resulting in poor classification performance.Therefore,the methods to solve these problems are researched in this paper.The specific research results of this paper are as follows:(1)A fabric defect detection method based on the improved class activation mapping is proposed.At first,to solve the problem that the traditional CAM method has changed the convolutional neural network's structure,which will reduce the classification performance,the deep and shallow convolutional layers in convolutional neural network are combined,and the Squeeze-Excitation(SE)block is embedded behind the convolutional layer,so that the network can generate a fine-grained class activation map while improving the classification performance.Then,in order to improve the accuracy of defect localization,the class activation map of the two resolutions were fused to generate an improved class activation map.At last,five different kinds of fabric images are used to test the feasibility of the proposed method,which includes 1500 images with no defects,holes,stains,yarn defects and soft sides.Experimental results show that the recognition accuracy of the proposed method is 96.40%,which is 1.37% higher than the baseline method.Furthermore,it can locate fabric defects accurately when there is only image-level labeling available in the data set.(2)In order to improve the classification performance of fabric defect images,this paper proposes a fabric defect detection algorithm based on the combination of DCL(Destruction and Construction Learning)and attention mechanism.At first,in order to solve the problem of poor image classification performance of fabric defects,the DCL network is applied to fabric defect detection network,meanwhile,the DCL network and the idea of fabric defect detection based on DCL network is introduced.Then,in order to solve the problem that the classification network of DCL ignores the fine-grained expression ability of feature maps of feature maps and causing the classification performance of the network fall into the bottleneck,a fabric defect detection method combining the attention mechanism with the DCL network is proposed,and the structure of the attention module,the embedding position and the parameters of the DCL network are designed.Finally,the proposed method is compared with 7kinds of colorful fabric defect image data sets.Experimental results show that the classification accuracy of the proposed method for the data set reached 95.62%,which is7.98% higher than that of the baseline network,and proves the feasibility of the proposed method.At the same time,the CAM method is used to visualize the fabric defect image.This method has a good visualization effect on the fabric defect and can locate the localization of the defect area in the fabric image.