| Developers: | LETI St. Petersburg State Electrotechnical University, Siberian Federal University of FSAEI HPE |
| Date of the premiere of the system: | July 2022 |
| Branches: | Information Technology |
2022: Announcement of a neural network that recognizes handwritten text in Russian with 99 percent accuracy
At the end of July 2022, it became known about the creation in Russia of a neural network that recognizes handwritten text in Russian. This is the development of specialists from the Siberian Federal University (SFU) and the St. Petersburg State Electrotechnical University "LETI."
We are talking about the so-called convolutional neural network (CNN). It is capable of reading the manuscript in Russian with an accuracy of 99%, the developers say. According to them, the algorithm is oriented regardless of handwriting, protected from information leakage and does not require an Internet connection.
Neural network training was carried out using pre-processed data from the CoMNIST repository, a well-known database containing samples of handwritten spelling of letters in Latin and Cyrillic. First of all, scientists created a new set of data with a marked image for 33 letters of the Russian alphabet, then developed a new CNN architecture for detecting handwritten letters and compared it with existing models. After that, they laid out a complete description of the convoluted neural network and the source code so that other researchers could reproduce this data. The language Python and interactive development environment of Jupyter was chosen for programming.
According to Anastasia Safonova, Associate Professor of Artificial Intelligence Systems at SFU, the data set contains 13,299 photographs of uppercase, printed and italicized letters. Approximately 85% of these images were learned by the neural network (CNN) to recognize the letters of the Russian alphabet, and another 15% were tested for learned "knowledge." The whole training took 3 hours. The prediction accuracy of the model was up to 95.83%.
The developers of the neural network downloaded the project to the GitHub platform so that everyone could train the model on their own data set.[1]
Notes
| The site content is translated by machine translation software powered by PROMT. The machine-translated articles are not always perfect and may contain errors in vocabulary, syntax or grammar. Read original article If you find inaccuracies or errors in the results of machine translation, please write to editor@tadviser.ru. We will make every effort to correct them as soon as possible. |