Research On Industrial Workpiece Character Recognition Based On Deep Learning

In the field of industrial production,many workpiece tables have characters describing the production situation,and most of these characters are manually identified and input into the computer for processing.This method is labor-consuming and inefficient.With the development of industrial economy,the classification of industrial accessories tends to be more and more refined and standardized.In terms of materials,industrial parts can be divided into metal materials and non-metal materials.From the connection mode,it can be divided into thread connection,spline,elastic ring connection,etc.From the maintenance point of view,the maintenance cycle of each workpiece is also different.The steel stamp code and label on the parts are their identity information,which is helpful for subsequent classification,storage,regular maintenance and traceability.Since the industrial field has entered the intelligent stage,the requirements for manual operation are gradually reduced,and the degree of mechanization is improved,which gradually tends to the intelligent industry.In recent years,the label of industrial workpiece has gradually become standardized and meticulous,which brings good development to the whole industry.At the same time,the manual identification of the label also sets obstacles for each stage,and the speed and accuracy of manual reading can not be guaranteed.In order to achieve the goal of high efficiency and high accuracy,with the help of computer image recognition technology,the label recognition link in industrial manufacturing is intelligent upgraded to realize automatic recognition processing.With the rapid development of computer field,intelligent coding recognition is the inevitable way.The workpiece is the most basic unit of industrial production,and the label identification effect on it is closely related to the performance of the workpiece.Therefore,it is necessary to develop a highly efficient automatic identification of the workpiece label technology.At present,the image processing algorithm has some limitations in the field of character processing,which has a negative impact on the accuracy of character recognition.Therefore,it is necessary to establish an image technology that can effectively separate characters with the same color as the background and accurately identify such characters,which is also the focus of research in this field.In this paper,we study the reflective metal workpiece whose characters are the same color as the background and discuss the character image recognition problem of this type of workpiece.And the comparative analysis found that the classical image processing methods can not meet the recognition requirements,so it is necessary to combine the new image acquisition,preprocessing,character segmentation and recognition technology,which can effectively and accurately recognize such characters.The research content and results of this paper are as follows:1)In order to solve the problem of reflection,white light source and plane array camera are selected for image collection based on the oblique upper illumination mode,which is conducive to improving the clarity of the image and the contrast is also significantly improved.2)Details are enhanced by Retinex algorithm during preprocessing,so as to better meet the contrast requirements.Then gaussian filtering and bilateral filtering are used to improve the image quality.On this basis,MSER algorithm is used to evaluate and determine the most stable region,and Graham algorithm is used to process and accurately determine the convex shell envelope of the character region,according to which character segmentation is carried out.Then the character image is corrected according to the inclination Angle of the edge.For the curved image,the correction purpose is achieved by model comparison.3)Combined with Wellner algorithm,a local threshold method is established to perform automatic binarization transformation on preprocessed images,and then segment images with strong single-peak features.After neighborhood search of all independent connected domains in the image,independent characters are divided,and then the image of standard size is formed after appropriate scaling.4)Collect industrial character training set and test set,use tesserACt-OCR traditional character recognition method to identify target characters,classify the labels of industrial artifacts,and conduct experiments in turn,judge the advantages and disadvantages of traditional character recognition method according to the accuracy of testing various characters,and seek better solutions.5)In order to meet the requirements of character recognition in this context,deep learning technology is introduced,modeling is carried out based on Pytorch framework,and two feature detection heads and prediction heads are appropriately combined under STR architecture,and then different types of neural networks are formed.Then,an adaptive residual correction network SKE modified ResNeSt is established to train on MJSynth data set to improve its feature extraction performance,better character recognition,and test its performance through the test set.The results show that the character recognition accuracy of the improved ResNeSt neural network is improved obviously,and the corresponding accuracy is 97.66%.Comparing the recognition accuracy of the proposed model with other recognition models,it is found that the recognition accuracy of the proposed model is significantly improved compared with other models of the same type.This recognition model has strong applicability and can recognize characters in various industrial situations efficiently and accurately,which can be further popularized and applied.FIG.40 Table 4 reference 59...