Bangla Handwritten Character and Words Recognition-based on the YOLOv5 Algorithm

Abstract

Today, handwritten character recognition (HCR) is a significant research project issue in the Bangla language. One of the most well-known fundamental issues in Artificial Intelligence and it’s all sectors (ML, DL, IP) is considered preeminent difficult. Because of the wide range of different writing styles and organizational reflection amongst characters, it is a complicated motive. Bangla handwriting is characterized by many perplexing characters of handwritten and elaborate spontaneous writing style; despite progress in object identification technique, Handwritten Bangla Characters Recognition (HBCR) remains mostly unresolved. This paper has introduced a new approach based on Handwritten Bangla Character Recognition using Convolutional Neural Networks to enhance effectiveness. The real benefit of CNN methods is that they can isolate significant characteristics from given images, improving interpretation speed and efficiency and decreasing the model size. The proposed method was examined by employing a Convolutional Neural Network (CNN) and the You only look once (YOLO)v5 algorithm to detect Bangla handwritten continuous characters with higher recognition accuracy. For character recognition, the used dataset is comparably vast and robust. In this study, we have employed the Bangla-Lekha Isolated Dataset, which contains 50 primary characters, ten-character modifications, ten Bangla numerical digits, and 23 compound characters. Finally achieved 96% accuracy on Bangla character recognition and 91% on Bangla word recognition using the proposed approach, which is a substantially superior method.