Recognition of latin letters of the kazakh alphabet based on a neural network

Abstract

This article discusses the recognition of Latin letters of the Kazakh alphabet based on a neural network using the Tensorflow framework. The paper provides an overview of existing recognition methods, including methods based on machine learning. To do this, a database of Kazakh letters of the Latin alphabet is being created using the Turtle graphic library. A computational algorithm for recognizing Kazakh letters of the Latin alphabet has been developed. The algorithm uses operations on labels entered for letters to predict the target variable. The model is trained on the prepared data and then evaluated based on its performance on the validation dataset. The overall accuracy of the model is approximately 93.90%. This means that about 93.90% of the model's predictions were correct. The paper uses two metrics for the accuracy Precision and completeness Recall classes. These metrics show whether the models work well by class, show information about the performance of the model. The experiments have shown that the proposed algorithm provides high accuracy of recognition of letters of the Latin Kazakh alphabet. The results are illustrated graphically. In the discussion of the results, a multiclass ROC curve is presented, which is a graphical representation of the performance of the classification model at all classification thresholds. The performance of the model as a whole indicates the high performance of the classification model. The algorithm proposed in the article for recognizing Latin letters of the Kazakh alphabet can be used in various applications, such as optical character recognition (OCR) systems, automated verification of entered texts using mobile devices.