Research And Improvement On Fabric Anomaly Classification

With the increasing requirements of textile industry for fabric production efficiency and quality,manual detection has become increasingly difficult to meet its development needs.At present,the existing detection methods have problems such as high cost,low recall rate and poor accuracy,so it is urgent to develop a set of efficient automatic detection technology.As one of the research hotspots in the field of artificial intelligence,fabric defect automatic detection based on image recognition can greatly improve the production efficiency and product quality of textile industry.In this paper,aiming at the common fabric defect detection problems,around the defect detection method based on shallow features,a multi feature fusion algorithm is proposed;at the same time,based on the deep learning method,two typical networks,Alex Net and Le Net,are improved.First of all,aiming at the problems of imperfect feature representation,data redundancy and limited feature expression ability of the current mainstream fabric image feature extraction algorithms,a word vector fusion classification algorithm based on Laws and Gabor is proposed to extract the point,line,edge,energy and other features of image texture and space,and comparative experiments are carried out on single feature extraction method and feature fusion method.The results show that the average accuracy and recall rate of the fabric defect classification method based on multi feature fusion are 74.7%,which is 3% higher than Gabor based method and 6.9% higher than Laws based method.It can significantly improve the classification accuracy and solve the problem of incomplete expression of fabric defect inherent attributes.Secondly,aiming at the problem of low recall rate and detection rate of traditional fabric defect detection algorithm,the improved algorithm of Alex Net and Le Net based on deep learning is proposed.In addition,the Alex Net network is added with a hole convolution layer to increase the receptive field.The experimental results show that the average accuracy and recall rate of the improved network are 85% and 17% higher;the convolution layer is added to the Le Net network to further extract features,and the convolution bias is introduced to improve the prediction and evaluation accuracy.The results show that the accuracy of the improved network in fabric defect classification is improved and the average recall rate reached 82.5%,increased by 3%.The improved Alex Net network improves the F1 value by 14% compared with the shallow learning method with fusion features,and also has strong robustness.Finally,according to the previous two chapters on the existing problems of traditional fabric detection and the corresponding algorithm proposed,in different data size data sets,texture fusion classification algorithm and deep learning method are compared.The experimental results effectively prove that in different application scenarios,different fabric defect classification algorithms are selected to achieve the optimal effect.The experimental results show that the classification effect of shallow feature learning method is better when the amount of data is less than 800,and the effect of deep learning method is better when the amount of data is larger than 800.The research results have great reference significance for the method selection research in different data scale application scenarios in the future.