Research On Fabric Defect Detection Method Using Optimized Sparse Dictionary

The "Made in China 2025" plan and the planning of textile intelligent manufacturing equipment clearly put forward the priority development of advanced intelligent equipment.In recent years,although weaving equipment has made significant progress in automation and informatization,the degree of automation and production efficiency of some key links still need to be improved,which restricts the automation and unmanned process of the whole weaving process.This paper focuses on the current fabric quality detection process,which still relies on manual labor,and realize the automatic detection of fabric defects by developing advanced machine vision detection algorithms,in order to gradually reduce the manual detection and trimming of fabrics.To this end,this paper conducts the following research on the basis of the existing sparse dictionary-based fabric defect detection algorithm and addresses the problem of insufficient real-time.(1)In order to make the performance of the studied algorithm closer to the industrial field inspection needs,this paper uses a machine vision fabric detection system that conforms to the industrial scenario of defect detection,and introduces the image acquisition module and its acquired data.Firstly,we introduce the structure and important technical parameters of the image acquisition system in the machine vision fabric detection system developed by the group,analyze and summarize the advantages of the system and the acquisition effect that can achieve.Then we choose three different woven fabrics as the test objects,and use this system to acquire images under different light conditions,and account for the number of images acquired and the types of defects contained in each fabric.Finally,all the collected defect images are sorted and classified to build a defect image dataset for subsequent testing.(2)The sparse dictionary algorithm is studied to address the problem of insufficient real-time performance of existing sparse dictionary-based defect detection algorithms.Firstly,we introduce the directions and ideas to improve the detection speed of the sparse dictionary algorithm,and analyze the feasibility of the optimization method based on the sparse coding process.Then we introduce the dictionary grouping optimization strategy in detail,and choose gray fabric as the main object of study to optimize the dictionary parameters and analyze the approximation effect of the images with different parameters.Finally,the feasibility and effectiveness of the dictionary grouping strategy are initially verified by testing a small number of defect images of different types of fabrics.(3)In order to cope with the high sensitivity of the algorithm and the high number of false detections of image noise,a false detection suppression algorithm is designed.Firstly,the principle of non-maximum suppression is introduced,and the feasibility of its application to defect detection is analyzed.Then the reconstruction error is selected as the image patch scoring criterion,and the suppression coefficient is set to control the degree of false detection suppression,and experiments with different suppression coefficients and different overlapping methods of the algorithm are conducted.Finally,the relatively good control coefficient is selected based on the experimental results,and the effectiveness of the algorithm is verified.(4)In order to objectively evaluate the performance of the proposed fabric defect detection algorithm,static and dynamic experiments of the algorithm are conducted respectively.Firstly,the overall process of the algorithm is introduced and the evaluation criteria of static and dynamic tests are determined.Then static tests are conducted,the dataset for the tests is described and two related algorithms are selected for comparison,and the static experimental results show that the highest positive detection rate of the algorithm can reach96.09%.Afterwards,the algorithm is deployed into the self-developed machine vision fabric detection system,the detection process is described and the algorithm is used to process the fabric edge,and two different kinds of fabrics were selected for real-time detection.The dynamic experimental results show that the detection accuracy can reach up to 97.02%,and each 2432×896 pixels image detection only takes about 36 ms after the design of parallel computing process,which proves the effectiveness and real-time of the algorithm.Through the sparse dictionary grouping strategy,the dictionary atoms used in the sparse dictionary coding process are preferentially grouped,and through the optimization of dictionary parameters,the algorithm achieves the effect of greatly improving the detection speed while ensuring the detection accuracy,which is in line with the real-time application of automated fabric defect detection in industrial scenarios,and provide a reference for the development of algorithms and systems for real-time detection of fabric defects...