Method For Identifying Complex Surface Characters Of Industrial Workpieces

In industrial manufecturing,most workpiece surfaces have different types of characters to express production information.Most workpiece surface character information is often controlled by manual reading and input into computer.This method is labor-intensive and inefficient.At present?image processing algorithms are not suitable for segmentation of workpiece images with the same color of characters and background,which leads to the subsequent failure to recognize characters correctly.Therefore,the problem of image segmentation and character recognition of workpiece with the same color of characters and background needs to be solved urgently.When the problem of recognizing such characters is solved,the goal of intelligent industrial production of such industrial characters as the same color of characters and background can be achieved.The object of this project is metal workpiece image with reflective phenomena on the surface of workpiece and the same color of characters and background.By analyzing the gray histogram and image contrast and clarity of the character image of the target workpiece,it is found that the traditional image processing methods are difficult to achieve the purpose of recognition.Therefore,it is necessary to design corresponding image capture system?image preprocessing,character image segmentation and character recognition algorithm according to the characteristics of the object.The four parts of image processing technology are organically combined to achieve the recognition of target workpiece characters.The results of each part are as follows:(1)In the phase of image capture,we choose the way of illumination above oblique,and use LED white light source to cooperate with industrial lens and area array industrial Camera to collect character images,which can effectively solve the problem of image reflection,distinguish the characters with the same color as the background?and improve the uniformity and contrast of image illumination.(2)In the image preprocessing stage,firstly,Retinex algorithm is used to enhance the image character details and enhance the image contrast.Then the image is smoothed by Gauss filtering and bilateral filtering to suppress the interference information in the image.Based on MSER algorithm,the maximum stable region is found,and the convex hull envelope of the character region is found by Graham algorithm,then the character region is located accurately.The tilt angle of the edge of the character area is used to correct the tilt of the character image according to the size of the tilt angle.For the image with curved surface,the image with curved surface can be corrected by establishing a mathematical model.(3)A local threshold method based on Wellner algorithm and the idea of "center-periphery"is proposed to realize adaptive binarization of images and to segment images with obvious single peak features.Search the 8 adjacent area pixels in the image,all the independent connected areas in the image are segmented into a single character,and then the size of the single character image is scaled to a unified 28x28 pixel size.(4)The industrial character training set and test set have been collected.Next,an eight-layer convolution neural network model is designed and built.The trained convolution neural network model is used to recognize the target character.The final recognition accuracy is over 97%according to the test.The validity and robustness of the image segmentation algorithm are tested by selecting a variety of industrial character types.The results show that the image segmentation algorithm has good segmentation results for a variety of industrial characters.Compared with other recognition models,the recognition accuracy of this model is higher than that of other recognition methods The character recognition method in this thesis can be applied to many industrial occasions and meet the needs of many industrial character recognition.It has engineering application value.