Web-based Learning Media for Alphabet Recognition in Early Childhood using LeNet-5

Abstract

A child's ability to learn to read, write and count in their early years is an indicator of the developmental milestones that must be achieved, particularly in the areas of cognitive and language development. The process of acquiring reading ability entails the initial recognition of vowels and consonants, which are subsequently integrated to form syllables, and then words and simple sentences. The use of learning media such as posters, cards, letter puzzles and letter boxes has been used to help improve letter recognition skills. However, the disadvantages of this learning media are that the use of digital technology in learning media is not optimal and requires teacher support, while the school we observed has a limited number of teachers. These problems are solved by creating a web-based learning media for letter recognition, which can help students to learn how to write and read letters and to learn two languages (English and Indonesian). This learning media applies computer vision technology with the Convolutional Neural Network (CNN) method of LeNet-5 architecture. The optimal parameter is a learning rate of 0.1 with a maximum epoch of 100, resulting in a system accuracy of 100% and an application user acceptance test result of 92.8%.