Research On Fabric Defect Detection Method Based On Low-rank Decomposition

Fabric defect detection plays a key role in fabric quality control.In the traditional production process,fabric defect detection is done manually,which cannot meet the requirements for accuracy,consistency and efficiency of fabric defect detection.With the rapid development of machine vision technology,there have been mature detection algorithms for simple unpatterned fabric defect detection,but the defect detection algorithm for patterned fabric is not yet mature.This study uses image fusion to enhance the saliency of defects,and uses low-rank decomposition with structural constraints to obtain a sparse term containing defect pixels.Finally,the sparse term is automatically thresholded to achieve defect detection for patterned fabrics.The main research contents of this paper are as follows:(1)This paper introduces the theory of low-rank decomposition and analyzes the relationship between low-rank decomposition and fabric defect detection.Theoretically,low-rank decomposition can separate low-rank fabric background from sparse defects to realize defect detection.However,through the verification of several traditional defect detection methods based on low-rank decomposition,it is found that there are many problems in the application of low-rank decomposition in fabric defect detection.This paper summarizes and analyzes the causes of the problems and clarifies the direction of improvement.(2)This paper proposes a defect enhancement algorithm based on image fusion.The algorithm first constructs the energy image by extracting the energy features of the fabric image.Then pixel-level image fusion is performed on the energy image and the fabric image to realize the complementary enhancement of the detail information of the defect in the original image and the energy information of the defect in the energy image.Finally,a fusion image with high contrast between the fabric defect and the background is obtained,which solves the problem of poor detection effect of the traditional fabric defect detection method based on low-rank decomposition when the fabric defect and the background are similar.(3)This paper proposes a low-rank decomposition model with structural constraints.Defect pixels are spatially continuous and clustered,while noise pixels are scattered.The model takes this structural difference into account and adds constraints related to the spatial structure on the basis of the traditional low-rank decomposition model.The sparse term obtained by low-rank decomposition tends to retain defective pixels with high spatial aggregation,while rejecting scattered non-defect pixels,thereby reducing false detections.The method in this paper uses a public database of fabric image to carry out defect detection experiments,and compares it with several other excellent defect detection methods,including the comparison of defect detection results and evaluation metrics.The experimental results show that this method is suitable for different types of fabrics and different types of defects,and the defect detection evaluation metrics also verifies that the comprehensive detection ability of this method is better than other fabric defect detection methods.