Devanagari Script Digit Classification using modified AlexNet with Transfer Learning
Ch. Rajendra Prasad, Sreedhar Kollem, Srinivas Samala, P. Ramchandar Rao, Srikanth Yalabaka, A. Chakradhar · 2022
Digitization of ancient scripts is more important to make available for future generations. Devanagari script is the most popular ancient Indian literature of India. Over the last decade the researcher focusing on Devanagari script digit classification. This paper presents Devanagari script digit classification using modified AlexNet with transfer learning. The pre-trained AlexNet 23rd and 25th layers modified for the requirement of Devanagari script digit data. The proposed model employs a dataset from the Kaggle and Stochastic Gradient Descent with a momentum optimizer employed for training. The proposed model shows impressive results in terms of accuracy of 93.64%.