FullExpression Using Transfer Learning in the Classification of Human Emotions

Abstract

During human evolution emotion expression became an important social tool that contributed to the complexification of societies. Human-computer interaction is commonly present in our daily life, and the industry is struggling for solutions that can analyze human emotions, to improve workers safety and security, as well as processes optimization. In this work we present a software built using the transfer-learning technique on a deep learning model, and conclude about how it can classify human emotions through facial expression analysis. A Convolutional Neuronal Network model was trained and used in a web application. Several tools were created to facilitate the software development process, including the training and validation processes. Data was collected by combining several facial expression emotion databases. Software evaluation revealed an accuracy in identifying the correct emotions close to 80% .