Recognition of Bengali Handwritten Digits Using Spiking Neural Network Architecture

Abstract

Bengali Handwritten Digit Recognition (BHDR) has extensive applications in OCR, voting machines, postal mail sorting, security systems, robotics, and many other fields. BHDR can be performed using various popular machine learning models and deep neural network architectures among which Spiking Neural Network (SNN) is getting increasing attention in recent works. SNN is an emerging machine learning model which mimics the natural processing mechanism of actual neurons of the brain. In this paper, SNN is applied for the recognition of Bangla Handwritten Digits using a popular dataset called ‘NumtaDB’. The images have been brought through various preprocessing operations for the SNN model so that it could better interpret the digits. The performance is analyzed for different values of the parameters of SNN. By systematically changing the parameters, the best combination of the values is selected for getting optimal accuracy. The model gives an accuracy of 91.36% with a comparatively faster training time using fewer computational resources relative to other machine learning models.