EEG-Based Emotion Recognition Using Optimized Deep-Learning Techniques

Emotions are crucial in identifying the current mental state. Electroencephalography (EEG) signals can accurately determine the current mental state. The automatic identification and analysis of human emotions using EEG data is vital in the treatment of psychiatric diseases due to the influence that emotions have on interactions, interpretations, and decisions. However, one major problem with EEG recorders is their low spatial resolution. EEG signals are nonlinear and complicated. Advanced signal processing techniques are required to analyze and extract useful characteristics. Emotion recognition using EEG can improve human-machine interactions. A convolutional neural network (CNN) is suggested in this article for automatic feature extraction and categorization of emotions. First, EEG signals are converted into pictures by applying time-frequency representation techniques. Following this, these pictures are fed into CNN and optimized CNN models for training. The optimized CNN model produces better accuracy for classifying emotions. Deep learning (DL) techniques are popular for determining complex patterns. Deep learning methods have shown good results in EEG-based emotion identification. It can automatically extract high-level information. DL's performance depends on hyperparameters. However, optimizing the performance of a deep-learning model requires accurate hyperparameter tuning. The manual hyperparameter tuning method is tiresome, costly, computationally expensive, and time-consuming. As a result, an automated method is required to determine the best hyperparameters to maximize DL's efficacy. To automate the hyperparameter optimization process, the proposed work presents a novel framework based on Bayesian optimization. Bayesian optimization (BO) is used to automate hyperparameter selection to predict emotional states.