Research On Fabric Defect Detection Method Based On Machine Vision Technology

Since the 21st century,with the development of people's living condition,people's demand for clothing has increased,which has also promoted the production of fabrics.In order to meet this situation,the fabric production process is gradually becoming intelligent,and one of the key parts is to detect fabric defects.At present,most domestic enterprises choose the traditional manual visual inspection method,but this method is easily affected by the subjective influence of the inspectors,and it also extremely tests the physical strength and eyesight of the inspectors,resulting in low detection efficiency and accuracy.So as to solve the above problems,more and more scholars have applied machine vision technology in defect detection field,and made outstanding achievements in this field.Hence this paper chooses to apply machine vision technology to fabric defect detection.The main contents of this paper are as follows:(1)Aiming at the problem of low accuracy and efficiency of fabric surface defect detection,a method of feature selection and synchronously parameter optimization based on binary random drift particle swarm algorithm is proposed.BRDPSO algorithm is used to select features on the original feature set and optimize the parameters of the random forest classifier synchronously,which is guided the search of algorithm by the classification accuracy.Finally,RF classification constructed with the optimal parameters is used to detect fabric surface defects on the selected feature subset.The experimental results show that BRDPSO algorithm that can select features and synchronously optimize parameters can more effectively improve the accuracy and efficiency of fabric surface defect detection.Compared with the proposed optimization algorithms,its detection effect is better.(2)Due to the previous research belongs to solve two-category problem,that is to say,the proposed algorithm can only detect a certain type of defect.BRDPSO algorithm is put forward combined with isolated forests(IF)algorithm,to realize of feature selection on the original feature set,at the same time optimize the parameters of the IF classifier.And the algorithm guides the search by the classification evaluation index AUC score.In the image processing stage,the convolution operation of the optimal Gabor filter is added to make it easier to distinguish the background of the fabric image from the defect area.A lot of experiments show that this proposed method improves the accuracy of defect detection.(3)In recent years,deep learning convolutional neural network(CNN)has shown more powerful capabilities than traditional machine learning in classification and detection problems,so Deep Support Vector Data Description(Deep SVDD)is proposed for the classification detection of fabric surface defects.Experimental results show that compared with the traditional machine learning algorithm,this method has higher classification accuracy,and only uses normal samples in the training process,so this algorithm is more suitable for industrial production process.(4)For solving the problem that the structure of CNN is difficult to be determined,a full deep learning model is proposed,which uses random drift particle swarm optimization(RDPSO)algorithm to automatically search for the optimal one class deep support vector data description(One-class Deep SVDD).Aiming at a specific fabric surface defect data set,this method can achieve end-to-end classification and detection.Meanwhile,only normal non-defect sample data is required during the training process,which makes the algorithm more suitable for actual enterprise needs.This method has better classification and detection performance and has strong versatility.