Study On Residual Yarn Inspection Of Bobbins Based On Computer Vision

Residual yarn inspection is an important function of ring spinning-winding unit.Since the bobbin which is retracted from the winder may not be completely retracted,it is necessary to detect the amount of remaining yarn and sort the bobbin according to the result.At present,the automatic winder equipped with the fine-coupling device generally adopts a mechanical detecting device,and the remaining yarn amount of the bobbin is determined by touching the bobbin by various forms of tapping.The disadvantage of this type of solution is that the rapid drop of the beater and the impact on the tube yarn may adversely affect the formation of the tube and the quality of the yarn.With the increasing use of machine vision in the field of industrial inspection,attempts have been made to apply machine vision technology to residual yarn detection in recent years.The purpose of this study is to develop a non-contact residual yarn detection schema based on machine vision.In order to realize the yarn volume detection and the yarn tube classification of the bobbin conveying device in the automatic winder,this study firstly proposes an algorithm for segmenting bobbin and yarn based on texture features and these two parts are separated by filtering response.Then,by combining the shape,color and texture features,a feature vector construction method is introduced and then imported to Support Vector Machine to classify inputs as full yarn bobbin,residual yarn bobbin or empty tube.Finally,based on the mixed background model and tracking algorithm,this study builds a detection and tracking system of bobbins in motion under complex background,and the classification algorithm mentioned above is used to inspect the residual yarn amount of the bobbin.In order to divide tube pixel area and yarn pixel area in each image,an algorithm based on odd Gabor filter designed for bobbin inspection is introduced.In this study,the parameter setting methods of the odd-part Gabor filter are optimized according to the principle of maximal response.These parameters include central frequency ?,wavelength ?,Gaussian distribution standard deviation ? and filter direction ?.The tube surface and the yarn edge are strengthened by the parameter-optimized odd-part Gabor filter bank,and the filtered image is subjected to the connected domain inspection to segment the yarn region.In experiments,bobbins of different tube colors,yarn colors and yarn fineness were selected to test proposed method.The results show that the algorithm performs robustness to these samples without supervision.Qualitative test of light intensity proves that the algorithm has certain illumination invariance.On the issue of classification of bobbins,Support Vector Machine is imported as classifier in this study.The image of bobbin is captured and then divided into several areas.In each partition,a batch of geometric features is constructed by the sizes of foreground,background and convex hull area,which can be used to represent the shape and symmetry of corresponding partition.The Gabor filter bank is used to enhance the target texture information,and the texture features of each partition are constructed by the dominant color extraction and color difference calculation.The combination of all features is then fed into a multi-classification support vector machine,which classifies input samples as empty bobbins,dirty-yarn bobbins or residual-yarn bobbins.The cross-validation results of the classification algorithm showed that the classification accuracy of the cotton yarn bobbins with various tube wall colors by the classifier reached more than 96% under various parameter levels;the tests of multi-species yarns showed that the classification true positive rate of yarn bobbins with yarn of different fineness and color reached more than 92%.In the aspect of detecting and tracking moving bobbin objects,this study adopts Gaussian mixture background model for foreground detection under fixed camera position situation,and Camshift object tracking algorithm is used for the trace of moving targets and realize time domain state recording of each foreground region in the Gaussian mixture background model.After the background model detects the moving target,the foreground region is identified by the classifier.If the foreground region is identified as a bobbin,it is classified and the tracking model keeps recording the movement of the bobbin;if the foreground region is identified as noise,then the background model is updated.The online video monitoring test results show that the algorithm can effectively detect and track the bobbins in the spinning-winding unit.