Design Of Printing Inspection System For Beer Box Based On Machine Vision

Beer is a daily drink with a great consumption.The printing quality of beer box as their packaging greatly affected the sales volume and brand image of the product.So,it was of great significance for enterprises to inspect the printing quality of beer box.And for beer boxes surface paper such picture information rich offset printing quality inspection was also of great academic significance.Therefore,according to the actual demand,designed a set of printing inspection system for beer box.In this paper,the common defects in the process of beer box printing were divided into global defects and local defects.According to the nature of global defects,a relatively broad contrast threshold was constructed to complete the detection of obvious global defects.According to the properties of local defects,a template matching detection scheme based on the variation model was proposed firstly,and a variation model was made by extracting the shape and gray value information from the template image,and then the new image to be detected was compared with the variation model to achieve the purpose of local defect detection.In order to solve the conflict between the detection accuracy and the poor anti-interference ability of traditional machine vision technology,a local defect detection scheme based on the semantic segmentation model of Pyramid Scene Parsing Network was proposed.Through the multiscale feature fusion,more context information was obtained,and the local defect detection was completed.The two local defect detection schemes were tested respectively.The detection accuracy of the template matching scheme based on the variation model reached 96%,and the average time was 10ms.The detection accuracy was high but the anti-interference ability was poor.The accuracy of the scheme based on semantic segmentation model is more than 71%,and the average time was in 50ms,but the anti-interference ability is strong.Experimental results prove the feasibility of two options,used at this stage is more stable and efficient local defect detection scheme based on differences between the model and global defect detection of complete system implemented in the form of software platform,experiment results prove the design of the system can quickly and accurately for beer box printing inspection.