Research On Defect Detection Of Code Character On Beverage Package Based On Deep Learning

Industrial automation inspection is an important way for enterprises to improve production efficiency and ensure product quality.In the beverage industry,the quality of the inkjet code such as the production date and product model on the beverage package directly affects consumers' desire to buy and the profit and reputation of the company.There will be various defects of code characters in the actual production process such as repeated printing,lost printing,missing printing and pollution due to some reasons,and defect inspections need to be completed in time before products leave the factory.Traditional machine vision technology can better detect defects of code characters on simple background or high contrast,but cannot meet the inspection requirements under harsh conditions,such as detecting defects of code characters on complex backgrounds.A defect detection method based on deep learning designed in this paper for code characters on the complex background,the experimental results show that its speed and accuracy are better,and it can meet the code detection requirements in actual industrial field.The research content of this paper is as follows:(1)According to the actual needs of code inspection,the hardware equipment of the inspection system is selected and designed,and an experimental platform for code inspection is constructed.After collecting the code data set and deploying the algorithm model,the experimental platform finally achieved the defect detection task of the code characters on the complex background.(2)Aiming at the problem of the small number in the original code data set and the imbalance of positive and negative samples,processing algorithms such as morphological operations are used to generate a large number of defective samples to balance the positive and negative samples.Combining multiple data expansion methods to further expand the code data set,including data expansion for code characters and their labels,and combined data expansion,etc.The above data preprocessing operations lay the foundation for the realization of the detection algorithm.(3)The efficient object detection network called BBE is designed to detect the code characters on the complex background.The network is divided into feature extraction network BUNet,feature fusion network BWNet,classification and regression network.The feature extraction network is based on the core module of the Efficient Net network,which is lightweight and has strong feature extraction capabilities.The accuracy of the training model under the Tensorflow framework reached 0.9961,and the actual test accuracy reached 0.9985.The detection time of a single code image is about 72 ms under the hardware environment of the Intel Core i5 CPU.The detection method in this paper is more accurate and faster compared to the classic object detection networks and their lightweight models.(4)The software of the code character defect detection system is designed based on Python and Py Qt5.The main interface of the software is friendly to human-computer interaction,which can integrate the detection algorithm model of this paper to test the product quality.This paper proposes to use the BBE object detection network to achieve the defect detection of the code characters on the complex background.The experimental results demonstrate that the lightweight detection method proposed in this paper has high detection accuracy(0.9985)and speed(50,000 bottles/hour),which meets the code detection needs in actual industrial field.There is good practical value and reference significance for the beverage production industry.