Real-time EEG-based Emotion Recognition using Discrete Wavelet Transforms on Full and Reduced Channel Signals

Abstract: Real-time EEG-based Emotion Recognition (EEG-ER) with consumer-grade EEG devices involves classification of emotions using a reduced number of channels. These devices typically provide only four or five channels, unlike the high number of channels (32 or more) typically used in most current state-of-the-art research. In this work we propose to use Discrete Wavelet Transforms (DWT) to extract time-frequency domain features, and we use time-windows of a few seconds to perform EEG-ER classification. This technique can be used in real-time, as opposed to post-hoc on the full session data. We also apply baseline removal preprocessing, developed in prior research, to our proposed DWT Entropy and Energy features, which improves classification accuracy significantly. We consider two different classifier architectures, a 3D Convolutional Neural Network (3D CNN) and a Support Vector Machine (SVM). We evaluate both models on subject-independent and subject dependent setups to classify the Valence and Arousal dimensions of an individual's emotional state. We test them on both the full 32-channel data provided by the DEAP dataset, and also a reduced 5-channel extract of the same dataset. The SVM model performs best on all the presented scenarios, achieving an accuracy of 95.32% on Valence and 95.68% on Arousal for the full 32-channel subject-dependent case, beating prior real-time EEG-ER subject-dependent benchmarks. On the subject-independent case an accuracy of 80.70% on Valence and 81.41% on Arousal was also obtained. Reducing the input data to 5 channels only degrades the accuracy by an average of 3.54% across all scenarios, making this model appropriate for use with more accessible low-end EEG devices.

• The paper presents a novel EEG emotion recognition system leveraging discrete wavelet transforms to extract time-frequency features from both full and reduced channel signals.
• It compares traditional 32-channel setups with a practical 5-channel configuration using consumer-grade EEG devices, showing a performance drop of under 4%.
• The study employs SVM and 3D CNN classifiers with baseline removal techniques, reaching over 95% subject-dependent accuracy.

Real-time EEG-based Emotion Recognition using Discrete Wavelet Transforms on Full and Reduced Channel Signals

Introduction

Recent advancements in electroencephalogram (EEG)-based emotion recognition have shown promise for both affective computing and psychiatric applications. The study titled "Real-time EEG-based Emotion Recognition using Discrete Wavelet Transforms on Full and Reduced Channel Signals" (2110.05635) introduces novel methods for emotion recognition using consumer-grade EEG devices with limited channels. Traditional setups often depend on extensive channel arrays, such as those found in medical-grade devices, whereas this study explores the viability of reduced, more practical configurations.

EEG Devices and Channel Configurations

This research focuses on a five-channel consumer-grade EEG device, the Emotiv Insight, known for its ease of use and affordability in brain-computer interface (BCI) applications. The study investigates both full 32-channel setups and reduced 5-channel configurations derived from the DEAP database, analyzing their potential in real-time emotion recognition systems.

Figure 1: The 5-channel Emotiv Insight EEG Device.

In a foundational part of this research, the authors discuss Russell's Circumplex Model of Affect, which classifies emotional states based on valence and arousal dimensions—key in constructing the emotion recognition framework.

Figure 2: Russell's Circumplex Model of Affect. \cite{Russell1980AAffect

Methodology and Feature Extraction

The authors propose a preprocessing technique utilizing Discrete Wavelet Transforms (DWT) to extract time-frequency domain features from the EEG signals. DWT has been employed to handle nonstationary signals and enhance data representation by capturing both temporal and spectral characteristics.

The EEG data from the DEAP dataset serves a dual purpose as a benchmark for evaluating the efficacy of these features. By applying DWT on various frequency bands (theta, alpha, beta, and gamma), the study aims to achieve emotion recognition with substantial accuracy using baseline removal preprocessing techniques.

Figure 3: The selected sensors for the reduced 5-channel scenario mapped onto the International 10-20 system.

Classifier Architectures

Two primary models are evaluated: a 3D CNN and an SVM with RBF kernel. The SVM showcased superior performance in most classification tasks, utilizing wavelet entropy and energy as feature inputs. The research also experimented with integrating baseline removal techniques, which significantly improved classification outcomes, particularly in subject-dependent scenarios. *Figure 4: Continuous Convolutional Neural Network with 3D DWT entropy input. $C \in {5,9*

The study further explores a chained classification approach where valence and arousal predictions are interdependent, leveraging one dimension's outcome to augment the accuracy of the other.

Results and Analysis

The study reports notable accuracy rates exceeding 95% in subject-dependent contexts for both full and reduced channel setups, demonstrating the effectiveness of baseline removal preprocessing. Furthermore, the loss of accuracy when transitioning from 32-channel to 5-channel configurations remains under 4%, reinforcing the potential of low-channel devices for practical applications.

Figure 5: Comparison of the best accuracy for Valence and Arousal. These were all achieved by the proposed SVM model.

Discussion

This research underlines the feasibility of using low-channel, consumer-grade EEG devices for emotion recognition tasks, particularly when coupled with advanced signal processing techniques such as DWT. The implications for BCI systems are vast, potentially transforming both entertainment and therapeutic domains with real-time, emotion-adaptive interactions.

The authors suggest future validation on diverse datasets and consumer experiments, considering further expansion into dimensional space through additional emotion features and more sophisticated preprocessing approaches.

Conclusions

In conclusion, this paper advances the capability of real-time EEG-based emotion recognition using reduced channel setups without significant sacrifice in accuracy. The findings present promising prospects for the deployment of affordable, practical emotion recognition systems in everyday applications, pioneering diverse innovations in affective computing and mental health diagnostics.