EAD-OHMER:Research On Online Handwritten Mathematical Expression Recognition Based On Encoder-Decoder

With the rapid development of Internet and artificial intelligence technology,the pace of education informatization is accelerating,the concept of smart education has begun to influence and change the traditional way of education,and modern electronic products and mobile terminals have been covered in all aspects of education.Handwriting recognition technology plays an indispensable role in the teaching process.At present,handwriting recognition mainly focuses on the study of Chinese characters,English characters and numbers,and has achieved very good results in these fields,but these recognitions are limited to one-dimensional space,and the recognition of two-dimensional structure distributions such as mathematics,physics and chemical formulas is not enough.Since the data of the two-dimensional structure of the mathematical formula is not only to consider the category of each symbol,but also to consider the structural positional relationship between the symbols,the requirements for handwriting technology are very high,so the research in this field still in the early stages,and needs to be constantly explored.Based on this,this paper will analyze the recognition technology of online handwritten mathematical expression,propose a data chain construction method based on behavior chain,and combine the deep learning model LSTM to analyze the sequential relationship between data,and help to better analyze the composition of symbols in the formula and the relationship between the symbols,and use the framework of the encoder-decoder and attention model to construct the entire online handwritten mathematical expression recognition model(EAD-OHMER).The main work done is as follows:1.Based on the Long Short Term Memory Network(LSTM),which is currently popular in processing sequential problems to construct online handwritten mathematical expression recognition model,which can directly use handwritten data as input and the formula symbol can be directly recognized without symbol segmentation.It is more convenient and accurate than the traditional identification method.2.Proposed a data construction method based on behavior chain,which focuses on the abstract description of the relationship between symbols in the formula.It aims to reflect the correlation between symbols,and avoid discarding the correlation between symbols and only consider the category of a symbol separately during analysis,and also reduces data redundancy in the input network model.3.Proposed an online handwritten mathematical expression recognition model based on encoder-decoder(EAD-OHMER),and the attention mechanism is added to solve the shortcomings of the original encoder-decoder when the input sequence is too long,which will cause the previous information to be covered.Compared with the traditional mathematical expression recognition,the methods of separate research in several links are more simple and accurate.4.Based on the proposed EAD-OHMER model,the implementation methods of each module,data preprocessing and feature extraction,behavior chain construction and encoder and decoder model construction algorithm and training process are introduced in detail.The input of the model is(8836,50,210)and the output is a series of mathematical symbol.Based on the method described in this paper,it was verified on different datasets and compared with other systems under the same dataset.It was found that EAD-OHMER had better recognition effect under the same conditions.