Research And Realization Of Workpiece Warehousing System Based On Number Visual Recognition

With the improvement of my country’s industrial production level,industrial production automation has become one of the most important technologies in the field of modern manufacturing.The realization of automation is a major way to significantly improve the production efficiency of enterprises.Nowadays,most factories still use manual identification number characters for the management of workpieces,and then enter the information into the system.Traditional Optical Character Recognition(OCR)is mainly used for character recognition with simple background and clear text,such as documents.Although the technology is relatively mature,the recognition performance of workpiece number characters is generally poor.In recent years,deep learning techniques have achieved remarkable results in the direction of computer vision.By combining deep learning technology,the recognition of workpiece number images in industrial scenarios will play a role in promoting the development of industrial automation in my country.The main research contents of this thesis include the detection and recognition of workpiece numbers.By optimizing the speed of the text detection network and enhancing the character characteristics of the text recognition network,the problems of misdetection and adhesion fracture of characters in industrial scenarios that lead to recognition failure are solved.Aiming at the problem of handwritten handwriting pollution in the industrial environment,which is easy to cause wrong positioning,this paper determines the text detection network PSENet based on semantic segmentation as the benchmark method for the experiment.In order to meet the real-time use requirements of the workpiece storage and retrieval system,a lightweight feature extraction network is designed for PSENet,and the upsampling feature enhancement is carried out using the idea of depth separability;the ASPP module is introduced to solve the problem that the numbered area occupies a small proportion in the image.Aiming at the small amount of data in the self-built text detection data set of images collected by the factory,the transfer learning method is adopted to reduce the dependence on training data.Experiments show that the improved PSENet in this paper has a detection accuracy of 98.5%on the self-built data set,which is 0.4% higher than the original network F-measure index,and the detection speed reaches 79 ms per image.In terms of character recognition,in view of the characteristics of character breakage and adhesion that are difficult to segment in industrial scenarios,this thesis selects the CRNN deep learning method without segmentation as the part number character recognition algorithm.Dense Net is used to improve the feature extraction network to improve the expressive ability of character features,bidirectional LSTM is used to predict character sequences according to the correlation between characters,and CTC loss function is used as the extension connection of LSTM for the problem of variable numbering length.On the self-built industrial character recognition dataset,different data enhancement methods are used to make the recognition model more robust.The experimental results show that the improved CRNN network has an accuracy rate of 92.9% on the dataset,which is better than other feature extraction network algorithms.This thesis improves the number detection and recognition effect through experiments in industrial scenarios.In addition,based on the B/S architecture development on the We Chat developer platform,the MINA framework and the Flask framework are used to complete the front-end and back-end communication,and the My SQL database is used for data storage,realizing a complete workpiece storage and retrieval system.