Research On The Formula Recognition Of Photographic Intelligent Paper-revising Machine Based On Deep Learning

With the rapid development of science and technology,computer online marking system in the college entrance examination,secondary examination,education bureau unified examination and school daily examination application examination is becoming more and more popular.However,the cursor marking machine is currently widely used in schools.For example,multiple-choice questions are used to mark papers.With the advent of the era of big data in recent years,the field of artificial intelligence has developed rapidly,and has demonstrated strong capabilities in machine translation,speech recognition,and semantic analysis.Based on the research status of the marking machine at home and abroad,this paper studies the design of the mechanism of the marking machine and the identification of elementary mathematical formulas as key technologies.At present,the cursor marking machine used in the education industry mainly uses answer cards as the medium,and printers,plotters and other equipment use printing paper and drawing paper as the medium.Therefore,a set of test papers is designed to take the test papers as the analysis object.The device for taking pictures has a lot of meaning.In the mechanical design part of this subject,according to the test paper size,friction coefficient and other physical properties,the paper analyzes the mechanics of the paper during the paper feeding process.The calculation selects and designs the drive part of the marking machine,and calculates the transmission gear,drive shaft,etc.By calculating the selection and design of the transmission part of the roller,and the key parts such as transmission gear and transmission shaft are evaluated in accordance with the corresponding strength theory.In addition,in consideration of the working characteristics of the marking machine and the compactness,safety,and debuggability of the marking device,the mechanism of pinching,twisting,feeding,and winding was designed,and three-dimensional modeling was performed using Solidworks software.Through the continuous optimization and improvement of the interference,layout and other issues between various institutions,the mechanical design of the final marking device was completed.The choice of elementary mathematical formula recognition as another key technical study is mainly because mathematical formulas are more complex than sequences of text recognition,and have more complex two-dimensional structures than formulas and equations of other science subjects,and There are problems of different sizes of handwritten symbols,which have certain research difficulty and research value.Based on deep learning technology,this topic expounds the collection and preprocessing of handwritten mathematical formula data sets.In order to solve the problem of different sizes of handwritten symbols,different receptive fields are used to enhance feature extraction.Bidirectional GRU effectively captures the contextual relationship between the encoding feature vector and the character table vector,and closely links the semantic relationship of complex formulas.It also has good predictive ability for long formulas and improves the accuracy of model prediction.In the end,the hyperparameter settings of the model training are described,and the training results and loss changes of the model on the real handwritten data set are given.The realization of the handwriting formula recognition function is achieved,and the accuracy formula recognition rate is well obtained.