Recognition of Human emotions using Semi-Supervised Technique with SHO based DTM

There are many different aspects to study in the relatively new field of facial expression recognition (FER). Credibility of these methodologies and their use in human-computer interaction for intelligent results have been bolstered by their applications in healthcare, security, safe driving, and so on. Computational FER is able to provide helpful indications that supplement speech in a manner reminiscent of human FER coding abilities. Similar FER approaches using deep learning and AI techniques have been built into edge modules to guarantee speed and accuracy. Many investigations have been conducted to learn more about FER. There is an immediate need to address the issue of insufficient generalizability of expression recognition models due to the small size of FER datasets, which hinders performance on the facial expression recognition job. We offer a semi-supervised learning approach for successful training of expression recognition models using unlabeled face data. To maximize the utility of the face recognition (FR) data for pseudo-label generation and hence enhance the model's capacity to predict facial emotions, we employ a dynamic threshold module (DTM) that may adaptively alter the threshold. Sea Horse Optimization (SHO) is used to find the best possible value for the confidence threshold. Validation results show that using a large number of unlabeled images helps boost facial expression recognition accuracy.