Tri-model classifiers for EEG based mental task classification: hybrid optimization assisted framework.

The commercial adoption of BCI technologies for both clinical and non-clinical applications is drawing scientists to the creation of wearable devices for daily living. Emotions are essential to human existence and have a significant impact on thinking. Emotion is frequently linked to rational decision-making, perception, interpersonal interaction, and even basic human intellect. The requirement for trustworthy and implementable methods for the detection of individual emotional responses is needed with rising attention of the scientific community towards the establishment of some significant emotional connections among people and computers. This work introduces EEG recognition model, where the input signal is pre-processed using band pass filter. Then, the features like discrete wavelet transform (DWT), band power, spectral flatness, and improved Entropy are extracted. Further, for recognition, tri-classifiers like long short term memory (LSTM), improved deep belief network (DBN) and recurrent neural network (RNN) are used. Also to enhance tri-model classifier performance, the weights of LSTM, improved DBN, and RNN are tuned by model named as shark smell updated BES optimization (SSU-BES). Finally, the perfection of SSU-BES is demonstrated over diverse metrics.

Deep Neural Network for EEG Signal-Based Subject-Independent Imaginary Mental Task Classification.

BACKGROUND: Mental task identification using electroencephalography (EEG) signals is required for patients with limited or no motor movements. A subject-independent mental task classification framework can be applied to identify the mental task of a subject with no available training statistics. Deep learning frameworks are popular among researchers for analyzing both spatial and time series data, making them well-suited for classifying EEG signals. METHOD: In this paper, a deep neural network model is proposed for mental task classification for an imagined task from EEG signal data. Pre-computed features of EEG signals were obtained after raw EEG signals acquired from the subjects were spatially filtered by applying the Laplacian surface. To handle high-dimensional data, principal component analysis (PCA) was performed which helps in the extraction of most discriminating features from input vectors. RESULT: The proposed model is non-invasive and aims to extract mental task-specific features from EEG data acquired from a particular subject. The training was performed on the average combined Power Spectrum Density (PSD) values of all but one subject. The performance of the proposed model based on a deep neural network (DNN) was evaluated using a benchmark dataset. We achieved 77.62% accuracy. CONCLUSION: The performance and comparison analysis with the related existing works validated that the proposed cross-subject classification framework outperforms the state-of-the-art algorithm in terms of performing an accurate mental task from EEG signals.