Reconhecimento de caracteres utilizando redes neurais convolucionais para auxiliar nas correções do sistema multiprova

Abstract

The Multiprova software was widely used during remote teaching at the Federal University of Rio Grande do Norte. With the return of classroom classes, the exam correction process needed to be improved to facilitate the adaptation of the system to presential courses, one of these improvements was the automatic correction of student response cards. Optical character recognition is a recent technique widely used for machines to read text written by humans. This academic work aims to develop a handwritten character recognition process to improve and facilitate exam correction in Multiprova software. For this purpose, convolutional neural networks were used to perform this task. With the UFRN student’s images, different scenarios were analyzed and compared with the increment of network configurations to generate neural networks with the best accuracy rates. As a result, excellent levels of accuracy were obtained, allowing high reliability in the software and more security and ease in exam correction.