Design And Implementation Of Beverage Cylinder Label Defect Inspection System

In the food and beverage industry,label packaging is an indispensable process in beverage production.The label carries the key information of the beverage products.The quality of the product determine whether the beverage can be sold in the market.And in the process of label packaging,defects such as label missed and label broken are caused due to various reasons.With the improvement of computer technology and processing performance,machine vision is widely used in various fields,which not only liberated the manual labor force but also grately improved production efficiency.Therefore,it is of great significance for enterprises to realize the automation of the inspection of beverage cylinder label defects.In this thesis,a software and hardware system for cylinder label defect inspection is designed.Based on the hardware platform,a series of image processing techniques are used to obtain the expanded image of cylinder label.In this way,the defects of beverage cylinder label can be detected.The main works of this thesis is as follows:(1)This thesis analyzes the characteristics of beverage packaging label and cylinder beverage,and the experimental platform of the detection system is designed and constructed,and indispensable hardware selection is introduced in detail.Four camera stations are used to collect the label images and four LED strip light sources are used,so that the image has ideal brightness and uniform illumination.(2)The original image preprocessing algorithm is designed.And the median filter and gaussian filter are used to reduce the image noise.A method of segmentation suitable for the cylinder label area is designed,the left and right boundary lines of the cylinder label are detected by edge detection and Hough transformation,and the tilted bottle images are corrected.Finally,a cylindrical adverse projection algorithm is proposed to correct the cylindrical labels with distortion.(3)After preprocessing,registration and fusion are performed between the adjacent label images.Based on the analysis and comparison of the main feature point extraction algorithms applied to the label image,an ORB feature point matching method based on region segmentation is proposed for image registration of adjacent label images.The experiment results show that this method not only reduced the calculation time but also improved the accuracy of feature point matching.Moreover,the fast approximate nearest neighbor searche and random sample consensus algorithm are used to eliminate the error matching points,which provide more accurate feature point pairs for the transformation matrix between images.Finally,the weight based on the Sigmoid function are used to build panorama for label inspection.(4)Corresponding detection algorithms are designed for common defects in cylindrical beverage labels.Specifically,it includes inspect whether a label is missed based on image gray features,an image gray projection algorithm are used to inspect slanted labels,and proposed an algorithm based on color space and morohological processing to inspect ruptured labels.(5)The software platform of cylinder label defect inspection system is developed,and the real-time capability and accuracy of the system is tested in the actual environment.The experimental results show that the proposed method in this thesis can fuse the label image collected by multi-station camera into a cylindrical label expansion image,and the defect inspection accuracy of cylindrical beverage packaging label is 0.99,which is helpful to improve the degree of automation of beverage label packaging production line.