Network Architecture Influence on Facial Emotion Recognition

Abstract Artificial Intelligence has been blending into daily life by means of many useful applications from voice command to facial recognition. One therapeutic application be supported AI solutions is treatment people with Autism Spectrum Disorder. A closed loop feedback system planned in conjunction a novel reward that will encourage the user express emotions and rewarded for it virtual environment. In this work five popular neural network architectures VGG16, ResNet50, GoogleNet, ShuffleNet EfficientNetb0 are studied compared, aim finding relation between accuracy developed features based on architecture, Facial Emotion Recognition (FER). Three datasets were used, OULU-CASIA training validation, alongside FACES JAFFE robustness analysis. The images first preprocessed eliminate background noise. performance model was true positive predictions Grad-CAM prediction visualizations visualize focus networks making decisions classification. Results showed deep high parameter space performed best, architecture design showing more influence region than classification results. This attributed different layer combinations as well parameters used feature extraction. Shallow depth better low FER application.