Research On Steel Stamping Character Recognition Based On Deep Learning YOLOv2 Algorithm

In the production of metal industrial products such as steel artifacts,each production artifact will be marked with its own unique production label,thus facilitating the monitoring of its production,supporting management and quality tracking.The traditional solution of manually reading the production label on the artifact and then manually entering the computer management system is labor-intensive and time-consuming,and error-prone.The vigorous development of computer vision technology makes it spread in all walks of life,the automatic recognition of characters based on visual images can realize the real-time grasp of steel component production information and timely find and solve problems,thereby greatly improving production efficiency.According to the development trend of Industry 4.0,the factory's demand for intelligent industrial production management is very urgent.Therefore,the automatic recognition technology of steel embossed characters for industrial production lines has great application prospects.At present,the character recognition technology has matured,and the steel embossed characters are different from the general optical characters.The steel embossed characters are the concave and convex characters that are pressed bythe mold.Using traditional optical character recognition methods,this makes the automatic recognition of embossed characters in steel with certain difficulty.Therefore,this paper proposes the use of deep learning algorithms to realize the rapid and accurate recognition of embossed characters on steel for the problem of embossed character recognition on steel in the industrial production process.Specifically,the following aspects of research work have been carried out:1.This paper studies the preprocessing algorithm for steel imprinted character images,and proposes a preprocessing scheme suitable for steel imprinted character images.Aiming at the characteristics of the embossed characters of steel and the different problems of character images due to the influence of the production environment of the factory,the research uses corresponding preprocessing methods to improve the images.First,the Hoff transform method is used to correct the slanted character image,and then the gray-scale transformation is used to enhance the image contrast to highlight the character area.Finally,the median filter and bilateral filter are used to suppress the noise interference in the image while retaining the characters Edge details.In addition,the Poisson fusion method is used to expand the character samples,and at the same time,the number of samples of each character type is balanced.2.This paper proposes a character recognition scheme based on deep learning algorithm,adopts YOLOv2 algorithm for object detection for steel imprinted characters,converts the problem of character classification and recognition into character object detection,and realizes end-to-end steelimprinted character recognition.Based on the embossed character data set of steel,a YOLOv2 network is constructed,and k-means algorithm is used for cluster analysis to improve the parameters of the anchor frame in the YOLOv2 network,thereby improving the recognition performance of the network model.At the same time,the output of the YOLOv2 network is improved,and the steel embossed character recognition results are directly output in the order of production labels.Compared with other traditional character recognition algorithms,the trained network model has improved recognition accuracy and detection rate to achieve better recognition results and meet the needs of industrial production accuracy and real-time performance.3.Build a web-based steel embossed character recognition platform.HTML and PHP are used to make a web page.In the process of designing and building a web platform,the design process has been improved to improve the performance of the web platform.Finally,on this web page platform,remote one-click steel embossed character recognition can be realized,and the production label on the steel artifact can be obtained in real time to achieve the goal of system management.