Research On Defect Detection Algorithm For Jacquard Knitted Fabric Based On Multi-feature Extraction And Optimized SVM

Defect detection in the production process of textiles is an vital factor of quality assurance.Currently,defect detection techniques for plain weaves and other common fabrics are relatively mature,but relatively little research on jacquard knitted fabrics.Due to the complex color patterns and knit textures of jacquard knitted fabrics,traditional defect detection algorithms cannot effectively distinguish fabric texture and defects,which affected the automation process of textile detection and seriously constrained the production efficiency of enterprises.To address this issue,this paper conducted a texture feature analysis and defect detection classification study on jacquard knitted fabrics,solving the interference of complex fabric textures on the defect detection process and accomplishing automatic detection and classification of fabric defects.In view of the problem of the detection interference caused by complex textures in jacquard knitted fabrics,this paper proposed a preprocessing method based on an improved PM model.Firstly,according to the gradient characteristics of fabric defects edges,this method adopt different diffusion behaviors for defect regions and texture regions.By introducing a new diffusion coefficient function with stronger smoothing ability during the diffusion process and using the frequency maximization method and correlation analysis method to calculate the optimal gradient threshold for different regions,the ability of PM model to autonomously select diffusion behavior is further improved.Through comparative experiments with other preprocessing algorithms,it is verified that this preprocessing algorithm can effectively smooth and suppress complex textures in fabrics while preserving defect detail information.Aiming at the problem of difficulty of defect detection in jacquard knitted fabrics,a defect localization and segmentation algorithm on account of the multi-feature fusion was proposed.This paper combined improved LBP,correlation,and information entropy to describe the grayscale feature differences between fabric textures and defects.The neighborhood mean removal method was proposed to normalize the feature maps.And weighted fusion reconstruction was performed based on the peak proportion in each feature map to further highlight the feature value differences between fabric textures and defects.Defect locations were identified using feature value mutations caused by defects in the reconstructed image,then using the region-growing segmentation algorithm to extract the complete shape of defects.Experimental results verified the proposed multi feature fusion algorithm can accurately locate and segment fabric defects..In order to classify the defect types of jacquard knitted fabrics,this paper proposed a SVM classification method based on an improved arithmetic optimization algorithm.By constructing a feature extractor to cream off the geometric and statistical properties of fabric defects as inputs for the classifier,the sample data volume was effectively reduced.To obtain the optimal parameters of the SVM classifier and achieve the recognition and classification of fabric defect types,this paper employ an improved AOA algorithm based on Levy Flight to optimize the parameter search process for the classifier.Finally,a jacquard knitted fabric image acquisition system was used to construct a fabric defect image database for comparative experiments.The experimental results showed the proposed defect detection method achieved a detection rate of 98.5%for defects,and the optimized SVM classifier achieved an accuracy rate of 98.3% for defect type discrimination.