Research On Character Location And Recognition Technology In Intelligent Labeling System Of Steel Plant

With the continuous emergence of more and more complex artificial intelligence systems,the iron and steel industry not only expands the production scale,but also urgently needs to improve its own production process.The rise of the industry 4.0wave has brought valuable opportunities,and the iron and steel industry has made great efforts to seize the opportunity and make every effort to build an iron and steel plant with fully intelligent production.However,the method of manual detection to identify the spray code on the surface of steel plate in iron and steel plants has the disadvantages of low efficiency,high risk,high misdetection and high cost,which brings difficulties to detection and identification,and it is difficult to meet the requirements of fully intelligent production.Therefore,the development of an intelligent labeling system based on machine vision can not only improve the work efficiency of the steel plate production line,but also realize the fully automatic production of the iron and steel plant.The advantages of machine vision technology are fast and accurate detection,high recognition rate,high stability,good adaptability and robustness.This paper has done a lot of research and application on three kinds of character location and recognition technology of steel plate upper surface steel seal,side surface spray printing and upper surface spray printing in iron and steel works.The main research contents and work are summarized as follows:First of all,the hardware structure design of intelligent labeling system is introduced in detail.In order to meet the imaging requirements of steel plate movement,a method of capturing video stream in large area by the combination of multiple area array cameras and high-power LED light source is proposed.In order to reduce data transmission,multi-sensor signal fusion software is used to control asynchronous trigger,which solves the problems that may exist in the process of steel plate movement,such as direction judgment,reciprocating movement,arbitrary stop,re-spray printing and large area snapping in different positions of steel plate.In order to improve the imaging quality of the captured image,high-power LED light source and different lighting methods are adopted to solve the difficulties such as dark indoor environment,natural light interference and steel plate self-reflection in the industrial production line.The connection between the whole hardware system is simple,reliable and effective,the signal and data transmission is fast and not offline,the imaging system is stable,and the imaging quality of the captured image is very good.Then,according to the different characteristics of three kinds of characters,an improved method based on FAST corner detection feature is proposed,which solves the problems of uneven illumination and motion distortion of steel printing and surface spray characters.The method based on image moment feature is used to solve the difficulties such as uneven gray distribution of side spray characters,complex background and external pollution.In view of the complex pollution and reflection of steel plate,a strategy from coarse to fine is proposed to locate the character region accurately,the center of gravity coarse location algorithm to find the bounding layer of the label area,and the projection extreme point fine location algorithm to accurately locate the label.it solves the problem of character region location in complex production environment.Aiming at the steel characters with leaking points and motion blur,a minimum bounding region algorithm based on experience guidance is proposed,which successfully solves the segmentation of dot matrix characters.On the other hand,face spray and side spray characters propose a multi-directional linear array scanning algorithm,which successfully solves the difficulties such as tilt,adhesion and splitting of continuous characters.In order to improve the accuracy of character recognition,BP neural network and support vector machine are compared.Finally,support vector machine is used to recognize characters,which solves the problems of large scale of large sample training model,slow training speed,low recognition rate of steel plate characters and difficult recognition of similar characters.Finally,according to the practical application requirements of steel plate production line,an intelligent labeling software system is designed.The system includes image acquisition software,detection and recognition software,database management software and human-computer interaction software.A large number of experiments show that the system can realize real-time recognition of three kinds of labels in different positions on the steel plate,and the recognition rate is more than99%.Missing recognition or error recognition mainly comes from seriously degraded images or incomplete characters.The system has been installed and operated in a steel plant for more than one year,and its stability has been verified.