Optimal Spatial Attention Network based Convolutional Neural Network for Facial Emotion Recognition

Abstract

Emotions create an innate and significant aspect of human behavior that colours the way of human communication. Automated emotion recognition related to facial expression becomes an interesting research domain that can be applied and presented in numerous areas like health, safety, and human-machine interfaces. Research scholars in this domain were concerned with formulating methods for interpreting, coding facial expressions, and extracting such features to have a superior forecast by computer. The analysis of human face characteristics and the recognition of their emotional conditions are very difficult tasks. With the outstanding success of deep learning (DL), various kinds of infrastructures of this method were used for obtaining superior performance. The main aim of this study was to create study on existing works on automatic facial emotion recognition (FER) through DL. In this aspect, this paper projects a novel enhanced seagull optimization with transfer learning based FER (ESGOTL-FER) model. The proposed ESGOTL-FER technique aims to detect the face and identify the emotions accurately and automatically. The proposed technique encompasses a three state process: face detection, emotion classification, and hyperparameter tuning. In addition, YOLOv5 was utilized for face detection. For emotion classification, spatial attention network with convolutional neural network (SAN-CNN) was utilized. For improving the performance of the SAN-CNN approach, the ESGO algorithm was employed as hyperparameter optimizer. A comprehensive group of simulations are implemented on benchmark datasets and outcome was examined in many aspects. The experimental results stated that the control of the proposed technique over the other existing approaches interms of several performance measures.