OGSSL: A Semi-Supervised Classification Model Coupled With Optimal Graph Learning for EEG Emotion Recognition

Abstract

Electroencephalogram (EEG) signals are generated from central nervous system which are difficult to disguise, leading to its popularity in emotion recognition. Recently, semisupervised learning exhibits promising emotion recognition performance by involving unlabeled EEG data into model training. However, if we first build a graph to characterize the sample similarities and then perform label propagation on this graph, these two steps cannot well collaborate with each other. In this paper, we propose an Optimal Graph coupled Semi-Supervised Learning (OGSSL) model for EEG emotion recognition by unifying the adaptive graph learning and emotion recognition into a single objective. Besides, we improve the label indicator matrix of unlabeled samples in order to directly obtain their emotional states. Moreover, the key EEG frequency bands and brain regions in emotion expression are automatically recognized by the projection matrix of OGSSL. Experimental results on the SEED-IV data set demonstrate that 1) OGSSL achieves excellent average accuracies of 76.51%, 77.08% and 81.29% in three crosssession emotion recognition tasks, 2) OGSSL is competent for discriminative EEG feature selection in emotion recognition, and 3) the Gamma frequency band, the left/right temporal, prefrontal, and (central) parietal lobes are identified to be more correlated with the occurrence of emotions.

A novel semi-supervised deep learning method for enhancing discriminability and diversity in EEG-based emotion recognition task

“…5.4.5. Limited SSL Some recent EEG-SSL [25][26][27] methods are specifically proposed to label the unlabeled EEG segments that are available in the training phase. Thus, the classification results show the quality of generated pseudo-labels for the unsupervised set.…”

“…Thus, the classification results show the quality of generated pseudo-labels for the unsupervised set. In this experiment, we compare the proposed method with OGSSL 1 [27] in this setting on the DEAP dataset for both valence and arousal recognition tasks, and the SEED dataset while the number of labeled data is limited (num labeled 200, 1000 { } Î…”

“…Another improvement of FixMatch is FlexMatch [24] that considers a different threshold for any class in the given dataset. Some methods [25][26][27] used self-supervised learning in the case of scarce labeled data. However, they train a shallow model using handcrafted features.…”

Encrypt with Your Mind: Reliable and Revocable Brain Biometrics via Multidimensional Gaussian Fitted Bit Allocation

“…Currently, most studies utilize EEG as the main source for biometric construction due to the convenience of signal acquisition [ 19 , 20 , 21 , 22 , 23 , 24 ]. However, a major problem with EEG is its low SNR because the brain’s electrical signals decay significantly while passing through the skull and scalp.…”

“…However, the keys generated by behavioral biometrics are generally very short, e.g., approximately 40 bits [ 16 ], which may be insufficient security for applications with high security requirements. Recent studies have demonstrated that the human brain can provide superior revocable biometric features [ 19 , 20 , 21 , 22 , 23 , 24 ]. Brain waves are continuous physiological signals with variable characteristics that change in real time; thus, it is extremely difficult to steal or imitate brain signals.…”