Deep learning based Intelligent Emotion Recognition and Classification System

Abstract

Deep learning techniques for automatic facial emotion recognition (ER) have recently received a lot of attention, however, the models that have been built are still unable to generalize properly due to a lack of large emotion datasets for deep learning. To solve this issue, in this paper we propose to use a MobileNetV2-based transfer learning approach to investigate how knowledge may be gathered inside a specific dataset and how information from ImageNet dataset (14,197,122 images) can be transferred into Kaggle emotion dataset (36082 images) to improve overall performance. To test the reliability of our system, we ran a series of tests using Kaggle emotion datasets containing seven different emotions. The experimental findings show that our emotion recognition system (ERS) system, which obtained 98.70% accuracy at a learning rate of 0.0001, leads to enhanced emotion detection performance and outperforms previous state-of-the-art methods that use fine-tuning and pre-trained techniques.