Handwriting Numerals Recognition Using Convolutional Neural Network Implemented on NVIDIA’s Jetson Nano

Abstract

An efficient handwriting numerals recognition structure based on Convolutional Neural Network (CNN) with RMSProp optimizer algorithm and Adam optimizer algorithm is presented in this paper. The experiment is implemented on NIVIDIA’s Jetson Nano platform, where we compare the performance of CNN models with two different optimizer algorithms. Experimental results show that the training accuracy of the model using the Adam optimization algorithm is better than that of the model with the RMSProp optimization algorithm. The training accuracy is 98.25%. Adam algorithm has fast convergence speed and RMSProp algorithm.