Research On Mesh Fabric Defect Detection System Based On Machine Vision

Textile is one of the daily use and consumption of commodities,a large number of textiles are consumed every year.The price and user experience of defective textiles will be greatly affected.At present,textile testing is mainly done by hand,but there are some problems in manual testing,such as low reliability,slow detection speed and subjective interference.The defect detection technology based on machine vision can better solve these problems.This paper studies the mesh fabric defect detection system based on machine vision.The main contents of the research include:(1)Construction of mesh fabric defect detection system.In order to ensure the detection system to collect clear images,not affected by the harsh external environment,it is necessary to design a reasonable hardware installation framework.Secondly,according to the precision requirements of mesh fabric defect detection,the hardware equipment of the defect detection system is designed and selected.(2)Research on Image Preprocessing Algorithm of Mesh Fabric.In order to solve the problem of low contrast and many noise points in mesh fabric images,the author proposes three effective image enhancement and denoising algorithms.Gray transformation algorithm.On the basis of the histogram equalization algorithm to enhance the image,the Sigmoid curve is used to transform the image to increase the contrast of light and dark in the mesh fabric image,making the difference between the fabric pixels and the mesh pixels more obvious.Gaussian pyramid algorithm.The Gauss pyramid algorithm is used to downsampling and upsampling the mesh fabric image,which can not only enhance the fabric texture contour,but also eliminate the noise points in the image.Image fusion algorithm.Firstly,histogram equalization algorithm and bilinear interpolation algorithm are used to enhance the fabric center region and suppress the mesh and fabric edge region.Then,two-dimensional information entropy algorithm is used to enhance the fabric edge region and suppress the mesh and fabric center region.Finally,the two images are fused to enhance the fabric pixels.The three image preprocessing algorithms can provide high quality mesh fabric images for image segmentation and defect detection algorithms.(3)Research on mesh fabric image segmentation algorithm.It is an important step to segment a complete mesh from the mesh fabric image.According to different segmentation principles,three mesh fabric image segmentation algorithms are proposed.Image segmentation algorithm based on improved FCM.Adding spatial information to FCM algorithm increases the segmentation accuracy of mesh fabric image.Firstly,a cross filter is established to obtain the spatial information of each fabric pixel.Then,the membership matrix of FCM algorithm is combined with the double threshold classification algorithm to add the spatial information to the segmentation principle.Regional gray minimum segmentation algorithm.Firstly,the theory of regional minimum segmentation algorithm is established to obtain the local interval of each pixel,and the pixels determined as regional minimum are divided into mesh pixels.Then,the theory of multi image fusion algorithm is created to fuse multiple images segmented based on regional gray minimum algorithm.The algorithm not only avoids the threshold selection,but also reduces the phenomenon of mesh adhesion.Gray projection correction algorithm.Firstly,the traditional Otsu algorithm is used to segment the mesh fabric image into the image with only partial mesh adhesion.Then the gray projection algorithm is used to find the adhesion mesh and remove the adhesions.The problem of mesh adhesion in mesh fabric image segmentation is solved.Three image segmentation algorithms can provide a complete mesh segmentation map for the defect detection step,and improve the accuracy of the defect detection of the mesh fabric.(4)Research on mesh fabric feature extraction and defect detection algorithm.Mesh fabric image belongs to periodic texture image,and defects will seriously damage the distribution and shape of mesh in mesh image.Therefore,this paper transforms mesh fabric defect detection problem into mesh classification problem.Firstly,the geometric features of mesh are extracted as eigenvalues.Then the decision tree classifier and BP neural network classifier are created by training sample data.Finally,different types of mesh are divided into defect free mesh and defect mesh to complete the defect detection of mesh fabric.(5)Experimental study.According to the research content of this paper,the image segmentation algorithm and mesh classification algorithm are verified by experiments.On the one hand,three improved image segmentation algorithms are tested and the accuracy of mesh segmentation is calculated by the standard of full mesh segmentation,which verifies the effectiveness of the proposed segmentation algorithm.On the other hand,the decision tree classifier and BP neural network classifier designed and trained are tested to verify the accuracy of mesh classification and defect detection.The experimental results show that the mesh fabric defect detection system designed in this paper can complete the defect detection task better.