Development of Technology for Creating High-Quality Abstracts

This paper focuses on creating high-quality notes. It is based on a modified version of the LeNet5 model, which allows to recognize individual characters, with the function of loading your own datasets, which can improve the accuracy of recognition, due to the characteristics of each handwriting. A neural network containing recurrent, convolutional, and fully connected layers was used to define words. To select a single word, the CTPN machine learning model was used. During the research, it was found that this solution for recognizing handwritten text is the most optimal for the task at hand. Also, an equally important part is to define the characteristic objects of the notes, such as headings, definitions, and formulas. The solution to this problem also used machine-learning algorithms. At the final stage, the main methods for working with file graphics were used to create the visual component of the note.