Achieving Handwritten Number Recognition through Deep Learning Strategies based on Conventional Neuron Networks

Abstract

Recognizing handwritten numbers is a common function of modern electronic devices that aids us with great convenience in aspects such as typing and vehicle identification. Aiming to simulate this function, this paper trains an algorithm that is based on Convolutional Neuron Network structure, a prevailing method for picture classification in Deep Learning, and identifies pictures of handwritten numbers by converging them to matrices. Later trials proves that such a model achieves a reliably high accuracy of over 98% out of the dataset for test use, and is thus capable of accomplishing normal number classification.