TSception: Capturing Temporal Dynamics and Spatial Asymmetry From EEG for Emotion Recognition

Ding, Yi; Robinson, Neethu; Zhang, Su; Zeng, Qiuhao; Guan, Cuntai

doi:10.1109/taffc.2022.3169001

Cited by 77 publications

(67 citation statements)

References 58 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We used the same dataset and EEG signals in our comparison experiments with other methods. Ding et al [ 24 ] proposed a multiscale convolutional neural network to achieve the classification of emotions in EEG by learning discriminative representations on temporal and channel dimensions. Ullah et al [ 25 ] proposed an ensemble learning algorithm that uses a kernel representation to describe the EEG channel and performs internal emotion recognition by solving an objective function.…”

Section: Experiments and Resultsmentioning

confidence: 99%

Deep Sparse Autoencoder and Recursive Neural Network for EEG Emotion Recognition

Liu

Shang

et al. 2022

Entropy

View full text Add to dashboard Cite

Recently, emotional electroencephalography (EEG) has been of great importance in brain–computer interfaces, and it is more urgent to realize automatic emotion recognition. The EEG signal has the disadvantages of being non-smooth, non-linear, stochastic, and susceptible to background noise. Additionally, EEG signal processing network models have the disadvantages of a large number of parameters and long training time. To address the above issues, a novel model is presented in this paper. Initially, a deep sparse autoencoder network (DSAE) was used to remove redundant information from the EEG signal and reconstruct its underlying features. Further, combining a convolutional neural network (CNN) with long short-term memory (LSTM) can extract relevant features from task-related features, mine the correlation between the 32 channels of the EEG signal, and integrate contextual information from these frames. The proposed DSAE + CNN + LSTM (DCRNN) model was experimented with on the public dataset DEAP. The classification accuracies of valence and arousal reached 76.70% and 81.43%, respectively. Meanwhile, we conducted experiments with other comparative methods to further demonstrate the effectiveness of the DCRNN method.

show abstract

Section: Experiments and Resultsmentioning

confidence: 99%

Deep Sparse Autoencoder and Recursive Neural Network for EEG Emotion Recognition

Liu

Shang

et al. 2022

Entropy

View full text Add to dashboard Cite

show abstract

“…BiDCNN [10] 68.1% 63.9% DBN [11] 88.3% 88.9% CDCN [12] 92.4% 92.9% TSception CNN [13] 63.7% 62.3% CNN-HRNN [14] 78.5% 78.5%…”

Section: Methodsmentioning

confidence: 99%

“…They used 1D dense block to find channel correlation. Ding et al [13] developed TSception, a multi-scale CNN, to seize EEG temporal dynamics and spatial asymmetry. The high-level fusion layer of TSception is used to fuse spatial relations for better accuracy.…”

Section: Introductionmentioning

confidence: 99%

EEG Emotion Realization through Hybrid Network

Bhandari¹,

Jain²

2022

IJETAE

View full text Add to dashboard Cite

Emotions are essential in interpreting the actions and relations of people. Nowadays, EEG signals are commonly used in emotion realization, so neurologists and psychiatrists can use a brain-computer interface to diagnose and cure emotional disorders. Another benefit of the BCI system is recognizing emotions without a clinical exam. A significant issue in automatic emotion realization is the appropriate extraction and selection of features. This paper aims to develop a state-of-the-art method to automatically classify four types of emotions from PSD and DE features. Features are fed to a convolutional neural network and long short-term memory network (CNN-LSTM). The 1DCNN network has been combined with the LSTM network to get a more stable system. The accuracy of subject-independent emotion realization with PSD and DE features is 84% and 93%, respectively. We have also compared our proposed CNN-LSTM model with other deep learning models. Our results are promising compared to similar approaches and have a better F1 score value. The experimental results also show that human emotions are better recognized by beta and gamma band

show abstract

“…According to the findings of this research, Ding et al [ 21 ] used scalp EEG data for primary interpretation. Deep learning approaches have been utilised, and they have been compared to more classic linear methods.…”

Section: Background Analysismentioning

confidence: 99%

Wearable Sensors with Internet of Things (IoT) and Vocabulary-Based Acoustic Signal Processing for Monitoring Children’s Health

Nagwanshi

Noonia

Tiwari³

et al. 2022

Computational Intelligence and Neuroscience

View full text Add to dashboard Cite

The brain is the most complex organ in the human body, and it is also the most complex organ in the whole biological system, making it the most complex organ on the planet. According to the findings of current studies, modern study that properly characterises the EEG data signal provides a clear classification accuracy of human activities which is distinct from previous research. Various brain wave patterns related to common activities such as sleeping, reading, and watching a movie may be found in the Electroencephalography (EEG) data that has been collected. As a consequence of these activities, we accumulate numerous sorts of emotion signals in our brains, including the Delta, Theta, and Alpha bands. These bands will provide different types of emotion signals in our brain as a result of these activities. As a consequence of the nonstationary nature of EEG recordings, time-frequency-domain techniques, on the other hand, are more likely to provide good findings. The ability to identify different neural rhythm scales using time-frequency representation has also been shown to be a legitimate EEG marker; this ability has also been demonstrated to be a powerful tool for investigating small-scale neural brain oscillations. This paper presents the first time that a frequency analysis of EEG dynamics has been undertaken. An augmenting decomposition consisting of the “Versatile Inspiring Wavelet Transform” and the “Adaptive Wavelet Transform” is used in conjunction with the EEG rhythms that were gathered to provide adequate temporal and spectral resolutions. Children’s wearable sensors are being used to collect data from a number of sources, including the Internet. The signal is conveyed over the Internet of Things (IoT). Specifically, the suggested approach is assessed on two EEG datasets, one of which was obtained in a noisy (i.e., nonshielded) environment and the other was recorded in a shielded environment. The results illustrate the resilience of the proposed training strategy. Therefore, our method contributes to the identification of specific brain activity in children who are taking part in the research as a result of their participation. On the basis of several parameters such as filtering response, accuracy, precision, recall, and F-measure, the MATLAB simulation software was used to evaluate the performance of the proposed system.

show abstract

TSception: Capturing Temporal Dynamics and Spatial Asymmetry From EEG for Emotion Recognition

Cited by 77 publications

References 58 publications

Deep Sparse Autoencoder and Recursive Neural Network for EEG Emotion Recognition

Deep Sparse Autoencoder and Recursive Neural Network for EEG Emotion Recognition

EEG Emotion Realization through Hybrid Network

Wearable Sensors with Internet of Things (IoT) and Vocabulary-Based Acoustic Signal Processing for Monitoring Children’s Health

Contact Info

Product

Resources

About