A Comparative Study of Window Size and Channel Arrangement on EEG-Emotion Recognition Using Deep CNN

Abstract

Emotion recognition based on electroencephalograms has become an active research area. Yet, identifying emotions using only brainwaves is still very challenging, especially the subject-independent task. Numerous studies have tried to propose methods to recognize emotions, including machine learning techniques like convolutional neural network (CNN). Since CNN has shown its potential in generalization to unseen subjects, manipulating CNN hyperparameters like the window size and electrode order might be beneficial. To our knowledge, this is the first work that extensively observed the parameter selection effect on the CNN. The temporal information in distinct window sizes was found to significantly affect the recognition performance, and CNN was found to be more responsive to changing window sizes than the support vector machine. Classifying the arousal achieved the best performance with a window size of ten seconds, obtaining 56.85% accuracy and a Matthews correlation coefficient (MCC) of 0.1369. Valence recognition had the best performance with a window length of eight seconds at 73.34% accuracy and an MCC value of 0.4669. Spatial information from varying the electrode orders had a small effect on the classification. Overall, valence results had a much more superior performance than arousal results, which were, perhaps, influenced by features related to brain activity asymmetry between the left and right hemispheres.

Image-Based Learning Using Gradient Class Activation Maps for Enhanced Physiological Interpretability of Motor Imagery Skills

“…The feature extraction procedures require several parameters to fix, affecting the CNN learning properties like discriminability and interpretability. However, the short-time window selected to encode the latency of brain responses must be adjusted to extract temporal EEG dynamics accurately [69]. For example, as [61] suggests, using shorter window values may increase the performance of poor-performing subjects.…”

Application of Electroencephalography-Based Machine Learning in Emotion Recognition: A Review

“…Generally, the purpose of setting the low cutoff frequency at about 4 Hz ( Özerdem and Polat, 2017 ; Chao et al, 2018 ; Pane et al, 2019 ; Yin et al, 2020 ) was to remove electrooculography (EOG) artifacts (0–4 Hz) and potential artifacts of respiration and body movements within 0–3 Hz. While some documents set the low cutoff frequency at about 1 Hz ( Yuvaraj et al, 2014 ; Bhatti et al, 2016 ; Liang et al, 2019 ; Hou et al, 2020 ; Keelawat et al, 2021 ), the purpose of which was to remove the baseline drift (DC component) in the EEG signal and the 1/f noise introduced by the acquire equipment. On the other hand, for high cutoff frequency, most researchers set it to about 45 Hz ( Kessous et al, 2010 ; Yuvaraj et al, 2014 ; Liang et al, 2019 ; Yin et al, 2020 ) to remove the other artifact noises at the high frequencies.…”

“…And the main contribution of Alhalaseh and Alasasfeh (2020) relied on using empirical mode decomposition/intrinsic mode functions (EMD/IMF) and variational mode decomposition (VMD) for signal processing purposes. Besides, Keelawat et al (2021) used EEGLAB, an open-source MATLAB environment for EEG processing, to remove contaminated artifacts based on ICA.…”

“…And Gupta et al (2020) used an HTC Vive VR display to enable participants to interact with immersive 360° videos in VR and collected EEG signals using a 16-channel OpenBCI EEG Cap with a 125 Hz sampling frequency. Later, Keelawat et al (2021) acquired EEG data based on a Waveguard EEG cap with a 250 Hz sampling rate from 12 students from Osaka University, to whom song samples were presented. What’s more, to effectively collect EEG signals, the attachment position of electrodes for EEG equipment in many studies follows the international 10–20 system ( Chai et al, 2016 ; Seo et al, 2019 ; Hou et al, 2020 ; Huang, 2021 ).…”

Review on Emotion Recognition Based on Electroencephalography

“…In recent years, CNN (Yea-Hoon et al, 2018;Dm et al, 2021;Keelawat et al, 2021), LSTM (Li et al, 2017;Liu et al, 2017;Sharma et al, 2020), Generative Adversarial Network (GAN) (Luo, 2018), and other network models have been widely used in EEG emotion recognition. In this paper, several selected keywords were used to search related literature in Elsevier and Springer and 645 published studies were retrieved.…”