| In the two major food and medical industries,Label of product packaging as an important part of the product,it is necessary to perform an online full review of its appearance,integrity,and internal information.Machine vision detection is stable,reliable,fast,and accurate,and can be used as an alternative to manual inspection of product label defects.Because the bottles are mostly cylinders,the labels of the cylindrical products are generally stuck on the cylindrical surface of the products,and it is more difficult to detect the appearance of the cylindrical packages than the flat packages.In this paper,the defects detection algorithm for cylindrical products is studied.Based on machine vision and machine learning technology,a complete set of solutions for defects detection of cylindrical products is designed.This article mainly includes the following aspects:(1)To the common defects detection requirements for cylindrical product labels,the difficulties of using machine vision to detect defects are analyzed.Based on the general flow of machine vision detection,the general idea of the label defects detection algorithm for cylindrical products is designed.(2)An image mosaic algorithm for cylindrical labels is designed.The cylindrical label original images are captured by four cameras from different angles and needs to be stitched into a complete label image.In this paper,a four-camera evaluation model,a 3Dcylindrical distortion correction model and a matching method are established to obtain a complete continuous normalized cylindrical product label unrolled diagram,which ensures the universality of the subsequent detection algorithm and facilitates the optimization of the detection algorithm speed.(3)The shape defects,color defects and content defects of the cylindrical product label are detected and judged by the feature threshold defects detection method,while the appearance edge defects are inspected and judged and classified by the SVM classifier defects detection method.(4)According to the actual needs of enterprises,an experimental platform was set up and an overall experimental scheme for the label defects detection system for cylindrical products was designed.The specific contents include the system hardware architecture,software architecture,and vision hardware selection.The cylinder product label defects detection algorithm studied in this paper is applied to the hardware system,which can realize online full-automatic visual inspection of label defects of cylindrical products.The experimental results show that with all the inspections enabled,the system test results omission rate is 0,the error rate is 0.2%,the appearance defects classification accuracy is higher than 97%,the average detection time per label is 480 ms,and the detection results accuracy and detection speed can meet the needs of the company. |
PDF Full Text Request