Determining the Emotional State of a Student-Child in a Digital Educational Environment From Static Images Using a Neural Network

Abstract

The article describes research on the possibility of recognizing children's emotions based on convolutional neural networks for use in digital educational environments such as, PiktoMir. Alternative solutions are considered and the experience of creating your own neural network trained on an experimental data set is presented. The results of testing the system using various metrics based on the validation data set and comparing the results with the most popular emotion recognition algorithms are presented. Conclusions are drawn on the direction of further research, in particular, the need to expand the training data set, further steps are proposed for refinement in order to integrate into the PiktoMir educational environment.