Optimizing Deep Learning Based CNNs for Effective Handwritten Character Image Recognition of Devanagari Script using R Language

The field of handwritten character recognition research has grown during the past few years. Its primary contributions to computerization and the creation of language-specific apps have earned it praise. Nonetheless, not much has been accomplished in the field of Devanagari script. The technique that has shown the most success in this area is optical character recognition, or OCR. The term OCR describes a machine's capacity to identify printed or handwritten text. The method of optical character recognition can be used to extract text from printed documents. However, retrieving data that is outdated, damaged, or otherwise specific to a given object presents procedural difficulties. Around the world, a large number of OCRs have been created for various languages. Recently, Indian languages such as Sanskrit, Hindi, Pali, Bangla, Telugu, etc. have also gained popularity due to their usage in development of domain specific applications. As a result, the application of Convolution Neural Networks (CNN) as a machine-learning instrument for Devanagari character classification is suggested in this research paper. The experimentation involves various CNN architectures with different depths and structures. The suggested models training_set and validation_set accuracies are discovered to be in range of 97.52%-98.85% and 98.12%-99.06% respectively.