Research On Fabric Surface Defect Detection Algorithm Based On Machine Vision And Neural Network

In the textile field,the surface quality of fabrics has always been an important factor affecting product beauty and sales,and surface defect detection is the most important step to control product quality.For a long time,the surface defects of fabrics in the textile industry have always been inspected by quality inspectors through human observation.Due to the high load of labor,the quality inspectors are under tremendous mental and physical pressure,and they often cause leakage due to lack of concentration.It is difficult to guarantee the reliability of inspection and false inspection.In recent years,machine vision has become more and more popular in industrial automation detection.For the detection of fabric surface defects,relevant scholars at home and abroad have proposed a large number of detection algorithms,but there are still many shortcomings in accuracy and usability.This article has conducted an in-depth and systematic research on the fabric surface defect detection algorithm.The main work of the article has four aspects: The first is to build a visual inspection hardware platform based on the domestic and foreign machine vision inspection systems,combined with the need for fabric surface defect detection;the second is for fabrics The most frequent defect type "color point" on the surface,a fast detection method is proposed;the third is a traditional machine vision detection algorithm for common defect types;the fourth is for large-scale and multiple types of defects.A detection model based on deep learning is proposed.This article introduces the research content from four aspects:(1)Aiming at surface defects such as splitting and color spots,a rapid detection method for point defects is proposed.Firstly,the noise points on the image are removed,and the defect image is filtered and preprocessed.The processed image is subjected to Hough transform to extract defect information and return to the defect position.In order to test the stability and accuracy of the proposed algorithm,a generative adversarial network is used to expand the data sample,compare with the manual detection result,and calculate the defect detection rate.(2)Aiming at common defects such as broken warp and weft,holes,and oil pollution,a defect detection method based on L0 gradient minimization and K-means clustering is proposed.It is mainly divided into two steps.First,use L0 gradient minimization to smooth the defective image,remove the influence of the background texture,and retain the larger edges of the image,and then use K-means clustering to cluster the smoothed image to segment the defective area.Experimental results show that this method can accurately detect defects and improve the efficiency of detection.(3)Aiming at different types of fabrics,a network model based on deep learning is designed:SE-SSDNet.First,perform Squeeze operation on the feature map after the SSD network convolution to obtain the global feature,and then perform the Excitation operation on the global feature to obtain the weights of different channels and then multiply the original feature to form the final feature.According to different weight values,the model pays more attention to channel features with large amount of information.Effectively improve the accuracy of fabric detection.There are 38 pictures,6 tables,and 82 references.