The Realization of Handwritten Mathematical Formula Calculation System Based on BG-Net
In view of the difficulty and high computation of handwritten mathematical formula recogni-tion,a new method based on BG-Net neural network for handwritten mathematical formula recognition and automatic calculation is proposed.The problem of incomplete segmentation of mathematical formula is solved by combining the improved minimum outer border and projection method.On the basis of tradition-al VGGNet network,the zero filling layer is introduced to control the size of feature map,which effective-ly solves the problem that the model can not obtain the edge key information.The global pooling layer is used to replace the full connection layer to reduce the network parameters.It improves the training speed of the network,adds batch standardized processing after convolution operation,uses smaller convolution kernel for feature extraction,realizes formula mapping according to the two-dimensional spatial charac-teristics and grammatical logic of various types of mathematical formulas,and realizes the calculation a-mong four operations,definite integral,equation,matrix,differential and mixed operation by using the idea of"group tree".The results show that the overall recognition rate of BG-Net model on the CROHME 2016 dataset is 94.68％,the calculation accuracy of each module of mathematical formula is more than 95％,and the accuracy of hybrid operation is 94.7％,which verifies the effectiveness of the method.
handwritten formula recognitionBG-Net modelspatial analysismathematical formula mappingmodulecalculation
NETL
NSTL
万方数据
