A printed formula recognition method based on the "Encoder-Decoder" deep learning model

With the popularization of computers and the rapid development of internet technology, electronic books and literature have become important information carriers. Therefore, digitization and informatization of scientific and technological books and literature containing a large number of mathematical formulas has become an important task. Although OCR (Optical Character Recognition) technology has high efficiency and accuracy in text recognition processing, it provides a convenient method for digitization and informatization of scientific and technological books, literature, etc. However, due to the complex two-dimensional structure of mathematical formulas and the diverse logical relationships and varying sizes between characters, OCR technology is not ideal for recognizing mathematical formulas. This paper proposes a deep learning model of "encoder-decoder" to solve the recognition problem of printed mathematical formulas. Utilize the powerful expression and sequence modeling capabilities of Vision Transformer and Transformer architecture to achieve accurate recognition of printed mathematical formulas. Through experimental verification, the model has achieved the expected recognition results on the publicly available printed mathematical formula datasets, with high recognition accuracy, fast running speed, and automatic generation of editable formats. This indicates that the model has potential in this field and provides new directions and choices for further improvement and application.