To Identify Facial Emotions Using an Enhanced CNN-TCN Model

Abstract

A fascinating area of research, facial expression-based automatic emotion recognition has been introduced and applied in a variety of fields, such as safety, health, and human-machine interactions. A computer technique for recognizing different emotions is called Facial Emotion Recognition (FER), and it is crucial for artificial intelligence (AI) research. For FER classification using video frames, this study employs convolutional neural networks (CNNs) and temporal convolutional networks (TCNs) using the Nadam optimizer in conjunction with preprocessing techniques. The preprocessing procedures include shuffling, one-hot encoding, and video pixel normalization. These are first used in the video frames to identify emotions. In order to manage the temporal and spatial dimensions of the video input, the classifier first incorporates 3D Convolution layers. The result should then be reformatted to be fed into the 1D Convolution layers of the TCN section. This work illustrates the architecture, the databases used, and the contributions addressed by comparing the results with the proposed methods. The performance of the proposed model is compared with other pre-trained deep learning models, including GoogleNet, ResNet, and Inception v3. The suggested enhanced CNN-TCN model with Nadam optimizer had 98% accuracy, precision, recall, and F1 score.