Research On Surface Defect Detection Algorithm Of Yarn-dyed Fabric

Yarn-dyed fabric defect detection is central to automated visual inspection and quality control in textile manufacturing.However,the traditional manual detection method has the disadvantages of high error detection rate,slow detection speed and large labor intensity,and the detection results are susceptible to subjective factors and external environment.Since the types of yarn-dyed fabric defects are various types and different forms,in which the rapid and accurate real-time detection is the primary challenge.Yarndyed fabric defect detection is set as main emphasis in this study,which aims to pursue general detection solution and the accuracy algorithm.The specific work is listed as follows:In order to detect the detection of patterned fabric,the patterned fabric defect detection algorithm based on Local Binary Pattern and Histogram of Oriented Gradient feature is proposed.Firstly,histogram equalization is applied to the patterned fabric to make the gray value of image is evenly distributed and the image details are enhanced.The equalized image is decomposed into multiple repeating units to extract LBP and hog features.Secondly,the labeled RUs samples that contain the class of features are used to train a support vector machine.Finally,the classifier is used to discriminate that with or without defect in RUs features and the location of the repeating unit in the fabric image is returned to realize the defect detection of patterned fabric.In view of the problem of fabric defect detection under small samples,a fabric defect detection algorithm based on convolution neural network is proposed.In order to detect and locate defects at the same time,the repetitive unit after image decomposition is used in the training stage,and the original fabric image is detected in the test stage according to its period.First of all,the MNIST data set is used to pre-train the network model and save the trained model parameters,which are used as initialization parameters of training repetitive units.After that,the fabric image is decomposed into multiple repeat units by autocorrelation function and the label of defect class is given manually.The pre-trained model parameters are loaded and the labeled fabric data is fed into the network to fine-tune the parameters,thereby accelerating and optimizing the learning efficiency of the model.At last,defects are detected during image inspection by sliding on the image to test local patches using the learned model.A defect detection algorithm is proposed to defect the detection of net color fabric and patterned fabric based on nonsubsampled contourlet transform and finite mixture of generalized gaussians.First of all,the size of the detection window is preset,and the characteristics in each window are extracted by using the nonsubsampled contourlet transform.Then the finite mixture of generalized gaussians is used to modeling the features.Finally,the defect-free sub-block is selected as a template,the relative entropy between the template and the sub-block model to be detected is calculated,and the defect detection is implemented by the threshold.There are 35 figures,8 tables and 95 references in this paper.