Abstract:The classification of mental or cognitive tasks in real time using single- or multi-channel EEG signals is an important field of research for neurofeedback and portable brain–computer interface systems. Exploring aspects of the electroencephalogram (EEG) data is an important research project because it sheds insight into the brain activity related to cognitive intents. CNNs have become a popular deep network design among the many others used for EEG data feature extraction, and this exploration offers considerable promise for this application. Due to many learnable parameters and a small number of EEG trials, most CNN-based techniques currently in use experience overfitting. This work presents the creation of a deep network model for classifying two distinct tasks from EEG data that is expressive, lightweight and efficient. It is based on the ResNet theory. In this work, we present a ConvResT architecture for two tasks classification, through the utilization of the conventional convolutional classification technique, which is enhanced by fusing a different residual connection and transformer mechanism, where the layers before the transformer are repeated once and twice. The Res block uses three different residual connections like normal, pre-activation and identity skip connection. Each Res connection leads to one model, which explored to three ConvResT models. The proposed models are evaluated on a real-world public dataset. Evaluation metrics like mean ROC and mean accuracy for loops 1 and 2 recalculated for individual subject. The highest subject independent classification of 96%, 97% and 98% obtained in loop 1 and 96%, 93% and 97% in loop 2 of mean ROC, and 91%, 92% and 93% in loops 1 and 2 of mean accuracy. In all the three ConvResT models, when compared with the other two models Res with identity skip connection achieved good results by an average mean ROC of 85.83% and 79.38% in loops 1 and 2 and an average mean accuracy of 78.22% and 78.19% in loops 1 and 2.

Deep Convolutional Neural Networks for Mental Load Classification Based on EEG Data

Convolutional Neural Networks with Large-Margin Softmax Loss Function for Cognitive Load Recognition

Learning Representations from EEG with Deep Recurrent-Convolutional Neural Networks

Mental Task Classification Using Electroencephalogram Signal

EEG-Based Mental Fatigue Detection Using CNN-LSTM

Cognitive Load Recognition Using Multi-channel Complex Network Method.

Pattern Recognition of Momentary Mental Workload Based on Multi-Channel Electrophysiological Data and Ensemble Convolutional Neural Networks

BRAIN2DEPTH: Lightweight CNN Model for Classification of Cognitive States from EEG Recordings

A Deep Transfer Convolutional Neural Network Framework for EEG Signal Classification

Using the Center Loss Function to Improve Deep Learning Performance for EEG Signal Classification

EEG-DCNet: A Fast and Accurate MI-EEG Dilated CNN Classification Method

Pattern Classification of Instantaneous Mental Workload Using Ensemble of Convolutional Neural Networks

Cognitive Load Recognition Using Multi-threshold United Complex Network

EEG-Based Hardware-Oriented Lightweight 1D-CNN Emotion Classifier

Multimodal Classification with Deep Convolutional-Recurrent Neural Networks for Electroencephalography

Deep Learning Decoding of Mental State in Non-invasive Brain Computer Interface

Cognitive Load Evaluation of Human-computer Interface Based on EEG Multi-dimensional Feature

Mental Pressure Recognition Method Based on CNN Model and EEG Signal under Cross Session

Attention-Based Parallel Multiscale Convolutional Neural Network for Visual Evoked Potentials EEG Classification.

Spectral and Temporal Feature Learning with Two-Stream Neural Networks for Mental Workload Assessment

EEG signal-based classification of mental tasks using a one-dimensional ConvResT model