Research On Character Recognition Method Of Continuous Ink Jet Printers

With the development of social economy and the continuous improvement of people's living standards,the Party and the government pay more and more attention to the safety of food and medicine.As an important reference standard for food and drug safety,production date has become a necessary inspection object when people choose to buy goods.Because of its low cost and simple operation,spray printing has become a common method of marking production date in food and pharmaceutical packaging enterprises.In order to test the correctness of the production date and other information of the jet printer,an experimental platform of the jet printer character recognition system is constructed,and a user-friendly interface for the jet printer character recognition is designed.The platform includes hardware system,MV_View,HALCON,PyCharm and PyQt5.The recognition method of jet-coded characters is studied by using the built platform.The recognition technology of jet-coded characters is improved,and the end-to-end non-segmented jet-coded character recognition is realized.The main research work completed in this paper is as follows:1.Design of the experimental platform for the jet code character recognition system.The experimental platform consists of hardware system and software system.The hardware system includes light source,industrial camera,lens,Mechatronics transmission device and external trigger device.The software system includes image acquisition module and image processing module.The image acquisition module includes format control,acquisition control,digital I/O control and continuous image storage plug-in.The image processing module includes image reading module,image preprocessing module,character segmentation module and character recognition module.2.Research on the recognition method of jet code character based on HALCON12.0.In this experiment,a single character image segmentation method based on connected domain segmentation and equal interval segmentation is studied,and the HALCON OCR(Optical Character Recognition)model is trained by using Support Vector Machine(SVM).The data set of this experiment is 150 dairy packaging box images with jet coded character areas.The HALCON OCR model is trained with 100 images.The off-line test of another 50 images iscarried out with the trained model.The recognition accuracy reaches 98%,and the average time of recognizing an image is 0.25 s.3.Research on the recognition method of jet code character based on TensorFlow.After studying the traditional machine learning methods,a Convolutional Neural Network(CNN)model is built using TensorFlow framework,which realizes the end-to-end non-segmented date code recognition of jet-coded characters.The data set used in this experiment is a self-made set.The convolution neural network is trained by reading 3650 sample images.At the same time,random expansion and corrosion algorithms are added to improve the robustness of the system.365 images out of the sample set were tested offline using the trained model.363 images were correctly recognized,the accuracy rate was 99.45%,and the average time of recognizing one character image was 25.2 ms.4.Design of Jet Code Character Recognition Interface Based on PyQt5 Platform.In order to more intuitively feel the role of each step in image processing,this paper designs a character recognition interface based on PyQt5 platform in PyCharm software by configuring computer vision library python-opencv and GUI development platform PyQt5.The interface is designed with modularization idea,which is divided into reading image module,image preprocessing module,image segmentation module and image recognition module.In the image preprocessing module,two image enhancement methods,two threshold segmentation methods and two basic morphological operations are provided for users to choose.Through template matching method,the jet code character image is recognized,and good results are achieved.It shows that the jet code character recognition interface designed can complete the work of jet code character recognition.