Research On Character Recognition And Detection Of Smooth Metal Workpiece Surface Based On OCR Technology

With the development of OCR technology,many character recognition algorithms such as SVM and CNN are gradually applied to character detection and recognition.The characters on the smooth metal surface are used to identify important information such as manufacturer,product serial number,and so on.At present,traditional manual visual inspection is used for character detection on smooth metal surfaces.This method not only has low detection efficiency,but also easily misses detection,resulting in a high error rate.Therefore,it is of great scientific value and practical significance to carry out research on the automatic detection technology of characters on smooth metal surfaces.Firstly,an optical system consisting of a low-angle ring light source illumination,a CCD area camera,and a telecentric lens for clear vision is designed for easy reflection on smooth metal surfaces.In the optical system design stage,the light reflection analysis of smooth metal surfaces was first performed using COMSOL multiphysics simulation software.Then,the hardware components of the optical system were analyzed and selected according to the detection index requirements.Finally,an optical system for character detection on smooth metal surfaces is designed.Secondly,a character detection algorithm for smooth metal surface images is proposed according to the distribution characteristics of characters on smooth metal surfaces.In the character detection algorithm design stage,the preprocessing analysis of the smooth metal workpiece image is first performed according to the characteristics of the smooth metal surface image.Then,a segmentation algorithm combining the horizontal projection method and the improved connected domain algorithm is proposed for segmentation of characters such as sticking,missing or breaking.Finally,a smooth metal surface character detection algorithm is proposed,and the algorithm flow design and character detection result analysis are performed.Finally,in view of the characteristics of high recognition rate and fast recognition speed in smooth metal surface character recognition,an improved CNN algorithm model is proposed,which can effectively improve the complexity of the network and increase the number of neurons in the output layer.The recognition accuracy rate and recognition speed of the algorithm model are given.Then the improved network is trained,tested and analyzed for character recognition results.