Research On The Key Technology Of Mathematical Formula Recognition And Its Application In Automatic Paper Judgment

With the continuous development of wisdom education,automatic judgment of theses has become a trend in the future,and as the entrance of wisdom education,OCR recognition has also been paid more and more attention by researchers.Among them,the handwritten mathematical formula recognition in the OCR recognition system has always been slow due to its complex and diverse two-dimensional structure.This thesis takes the two-dimensional structure of mathematical formulas as a starting point to study the recognition of mathematical formulas,and further studies the related problems of judging the consistency of expressions,and finally builds an automatic judging system for filling in the blanks in junior middle school mathematics.The main research contents are as follows:For mathematical formula recognition,this thesis uses the encoder-decoder frame-work to construct an end-to-end recognition model.In order not to limit the size of the input image,and to better extract character features under different sizes,this thesis pro-poses a multi-scale automatic selection of dense networks as the encoder.Aiming at the problem of complex structure in mathematical formulas,this thesis proposes an attention mechanism based on structure enhancement that can pay attention to the relative spatial position relationship in the structure,and for the problem that the nesting of structures complicates mathematical formulas,this thesis proposes the use of stack structure Man-age global information containing structural information.Finally,this thesis establishes a real handwritten mathematical formula data set to enhance the effect of the model in prac-tical applications.The mathematical formula model proposed in this thesis has a formula accuracy rate of 47.50% on the CROHME2014 test set,and a formula accuracy rate of69.71% on the self-built real mathematical formula data set.Aiming at the problem of the diversity of writing forms in the mathematics fill-in-the-blank test thesis,this article classifies the answer types to convert various answer types into judgment questions and consistency judgment questions.At the same time,this the-sis proposes an equality discriminator based on symbolic calculation and an expression consistency discriminator based on expression trees.Based on the above mathematical formula recognition model and expression consistency judging system,this thesis builds an automatic judging system for filling-in-the-blank questions in junior high school mathematics.Finally,the system conducts detailed experiments on the test data of 550 fill in the blank questions,and the recognition of mathematical formulas is accurate.The rate is 89.09%,and the judgment rate of the overall system is 90%.The experimental results show that the mathematical formula recognition model used in this article can accurately identify the answer,and the expression consistency discriminator can not only solve the problem of automatically judging thesiss by filling in the blanks,but also the core module of the future junior middle school math steplevel thesis judgment.