A Deep Learning Approach for Bangla Handwritten Character Recognition

For many purposes, such as document digitalization and linguistic study, handwritten character recognition is essential. This experiment offers a novel method for reading handwritten characters in Bangla using the modern EfficientNetV2B2 model. The complex structure of the Bangla alphabet, with its range of writing styles and variants, poses a significant difficulty. The innovative method makes use of the modern EfficientNetV2B2 model’s efficiency and superior performance, which have been adjusted on the dataset to produce notable increases in recognition accuracy, particularly for complex character types and styles. Analyzed the model’s interpretability and provided insights into the recognition procedure. Also discussed the wider societal effects of using technology to advance literacy and safeguard cultural heritage. On the two datasets, the experiment utilized had pretty excellent accuracy. Using the first dataset, EfficientNetV2B2 model was able to reach 100% training accuracy, 99.33% validation accuracy, and 98.67% test accuracy. On the second dataset, the model got 95.65% training accuracy, 97.75% validation accuracy, and 94.95% test accuracy.