Research On Machine Vision-based Repair And Quality Assessment Technology For Yarn Missing And Hairy Defects

With the rapid development of the textile industry,yarn quality assessment has become increasingly critical in the textile production process.Traditional yarn quality assessment methods mainly rely on manual inspection,which is inefficient and susceptible to subjective factors.Furthermore,the uneven local brightness caused by light source problems during yarn image imaging,resulting in missing fuzz,further increases the difficulty of yarn quality grading.To improve the accuracy and efficiency of yarn quality assessment,this study explores a yarn quality assessment technology based on machine vision,aiming to predict and repair yarn missing fuzz and accurately evaluate the quality of yarn images.This research focuses on yarn images captured by industrial cameras,proposes a fuzz repair model and technology based on machine vision,and conducts yarn quality assessment research.The main research contents include:(1)To effectively verify and test the proposed methods,this paper constructs a comprehensive dataset required for the experiments,including open-source datasets and private datasets,and expands the dataset using image data augmentation techniques.Subsequently,we employ a method based on the UNet network model to segment the yarn core and fuzz and verify the effectiveness of the method through relevant experiments.(2)A missing fuzz repair model based on the dual attention mechanism network is designed.Firstly,semantic information of both yarn core and fuzz is obtained through yarn core segmentation and fuzz segmentation methods.Then,semantic acquisition technology is used to extract the valid feature semantics of both parties,and morphological methods are employed to eliminate the impact of impurity fuzz around the yarn core.Finally,the missing fuzz is successfully repaired with the help of the DANet network.(3)A yarn quality assessment network based on the fusion of Efficient Net and Attention modules is proposed.Firstly,the difficulties in predicting yarn image quality assessment are analyzed.To solve these difficulties,the method is introduced and explained,and transfer learning and model optimization strategies are used based on this method.Secondly,a series of experiments are conducted to verify the effectiveness of the method in yarn image quality assessment,achieving yarn quality assessment classification and accurate grading.Finally,the superiority of the method is validated through comparative experiments,and the necessity of the Attention module is demonstrated through ablation experiments.This method effectively improves the accuracy and practicality of yarn quality assessment prediction.(4)In response to the research topic requirements,an application system and user interface are designed and developed.The system is easy to operate and can be widely applied to yarn production and quality inspection fields.This study provides an efficient,accurate,and objective yarn quality assessment method for the textile industry,which is expected to further enhance the quality and efficiency of textile production.