Research On Automatic Detection Of Bobbin Yarn Paper Tube Based On Machine Vision

In the context of small batch and multi-variety spinning processing,the same type of yarn often uses the same type of paper tube for winding,and the type of paper tube has a one-to-one correspondence with the variety of bobbin yarn.In order to make full use of the space of the site,the spinning enterprise transports the bobbin yarn by arranging the air guide rail.In this case,it cannot rely on manual resolution,and can only rely on the information of the mechanical gripper to determine the variety of bobbin yarn.If the operation is wrong,give the mechanical hand the wrong variety,failed to detect it and used for weaving or packaging into the market,will affect the product quality or be idle,discarded waste of raw materials.In the process of bobbin yarn transportation,the combination of paper tube information and mechanical gripper information can improve the detection effect.Aiming at the problem of paper tube detection on the bobbin transportation track,this paper proposes a detection method based on machine vision.The specific research contents are as follows:（1）By analyzing the structure of the bobbin yarn transportation track and the image acquisition conditions of the bobbin yarn,a set of image acquisition scheme for the bobbin yarn image of the air transportation track is explored.Preprocessed the image of bobbin yarn.Firstly,positioned the paper tube area,compared different image segmentation methods,the paper tube area segmentation method combining Otsu threshold method and image contour is optimized.The segmented paper tube color ring is imaged by polar coordinate transformation and interpolation algorithm.Then,the fast guided filtering method is usedto extract the illumination component of the brightness channel of the paper tube HSV image,and the contrast of the image os enhanced by combining the two-dimensional gamma transform.By comparing the filtering smoothing effects of different convolution kernel sizes,the 5×5 bilateral filtering method is used to denoise the image and filter out the interference of iurrelevant information.Finally,the illumination of the paper tube image is adjusted by applying a custom illumination medel.（2）According to the color and pattern characteristics of paper tube printing,color and texture features are selected to vectorize the image information.The SVM classification accuracy of paper tubes and the separability of t-SNE dimension reduction features are proposed as the basis for feature selection.When comparing the color features,only different color types of paper tubes are used,and the color moment of CIEL*a*b*space is preferred to represent the color information of paper tubes.The classification accuracy of paper tubes composed of different colors can reach 100%,and the characteristics of each category are obviously different.When analyzing the texture features,only paper tubes with different patterns are used,and the rotation invariant equivalent pattern LBP_8,2 with a sampling point of8 and a sampling radius of 2 is obtained by comparison,as the optimal texture descriptor of the paper tube image.The average classification accuracy of this feature for paper tube types with different patterns is 98%.Color and texture features are used to characterize the image information of paper tube.（3）According to the different requirements of paper tube detection task,the solutions for known paper tube detection and new paper tube detection are proposed.When facing the detection of known paper tubes,it is proposed to use as few samples as possible to construct a high-precision random forest classification model for detection.The average accuracy,recall rate and F1 value of the model for M blue,stripe blue and mei blue paper tubes are 97.41%,97.33%and 97.32%,respectively.It has better performance when detecting paper tube types with different colors or color patterns.For the detection task with new classes,on the basis of constructing a random forest,an isolated forest is introduced to detect abnormal samples,and a new class is detected by the outlier degree of abnormal samples.In the case of paper tubes with the same color,the same pattern and different color patterns as training samples,the false detection rate of the known class is 0,and the detection rate of the new class is 100%,95.83%and 86.67%respectively,which verifies the effectiveness of the proposed method for paper tube detection.A GUI interface is designed to visualize the detection process.Based on the given type of paper tube,this paper proposes a paper tube detection method based on classification model,anomaly detection and outliers.It can obtain accurate detection results for known paper tubes,and has strong detection ability for new paper tubes.It meets the detection requirements of spinning enterprises and provides a new idea for the application of machine vision in the field of textile detection.