Research On The Detection And Classification Of The Defects In The Weaving Plaid Yarn-dyed Fabric

The textile industry occupies an important position in China’s GDP.With the development of technology,more and more types of textile products have entered the lives of the people.The large-scale industrial automated production can not help but produce flaws of textiles.The labeling process in the actual production link is often mainly based on artificial testing,resulting in problems such as low identification efficiency and waste of labor costs.With the rapid development of computer vision technology,automated cloth detection system has been used to detect the defects of fabrics.However,the existing detection algorithm is mainly for white gray cloth.The complicated texture structure and pattern characteristics of the yarn-dyed fabric increase the difficulty of defect detection.The conventional white gray cloth detection method cannot be applicable.The research is still in the theoretical stage,and there are few mature fabric defect detection algorithms.This article is mainly for engineering applications,studying the defect detection of the plaid yarn-dyed fabric.The main research content is as follows:(1)Combined with the actual needs of industrial testing,the overall scheme of the plaid yarn-dyed fabric is designed,and the hardware and software systems are combined to complete the experimental platform-based motion control,image collection,image processing detection,and defective classification work.(2)For the problem of determining the position of the vertical period position in the continuous movement of the plaid yarn-dyed fabric,a position period test algorithm with the combination of the Hoff straight line detection and the histogram matching is applied.Use the Sobel operator to extract the border unique to the plaid yarn-dyed fabric and corner coordinates of their intersection.At the same time,the mid-value filtering is used to treat the noise.Calculate the chart template algorithm to obtain the vertical period position of the current image.(3)For the image processing of complex working scenarios and controlling cost conditions,a defect detection method based on LBP and gray-level co-occurrence matrix is improved.Through Fourier Frequency Domain transformation,the fabric image is transferred to the frequency domain space for high-pass filtering to remove the interference of high-frequency noise.Taking the four minimum period units as the statistical area,after the texture characteristics of the local binary pattern are performed,analyze the statistical method of the GLCM calculation results,establish a reasonable normal distribution model,and select the defect to determine the threshold.This algorithm determines the defect area in the minimum period unit,and the detection success rate is about 70%,which can effectively identify the larger types of defects.(4)In response to the problem of large statistical calculation quantities and fluctuations in the threshold,it proposes a defect detection method based on the cyclical division average template subtraction.The detection accuracy and identification accuracy of statistical characteristics detection methods can be further extracted in detail the accurate position and contour characteristics of the defect.The texture period characteristics within the minimum period unit are filtered on mean value,and then automatically generate the average template of the area.The morphological characteristics of the defect are extracted through the image subtraction operation.Integrate defect information as the output of the final detection result.The detection success rate of the method is more than 90%,and the detection accuracy is 0.3mm.It has good generality and characteristic extraction capabilities.Industrial defect detection applied to the plaid yarn-dyed fabric.(5)Based on defects of the small sample capacity obtained by the existing experiments,the classification model of the SVM support vector machine is established,and the simple classification of the common defect type is completed.By selecting a linear nuclear function and constructing 1-v-1 SVMs,the model is trained by the defective sample data.The accuracy of the classification of defects after experimental analysis is about 80%,which can basically solve the problem of defects classification of common types.