Abstract:OBJECTIVE: Brain-computer interfaces (BCI) enable direct communication with a computer, using neural activity as the control signal. This neural signal is generally chosen from a variety of well-studied electroencephalogram (EEG) signals. For a given BCI paradigm, feature extractors and classifiers are tailored to the distinct characteristics of its expected EEG control signal, limiting its application to that specific signal. Convolutional neural networks (CNNs), which have been used in computer vision and speech recognition to perform automatic feature extraction and classification, have successfully been applied to EEG-based BCIs; however, they have mainly been applied to single BCI paradigms and thus it remains unclear how these architectures generalize to other paradigms. Here, we ask if we can design a single CNN architecture to accurately classify EEG signals from different BCI paradigms, while simultaneously being as compact as possible.APPROACH: In this work we introduce EEGNet, a compact convolutional neural network for EEG-based BCIs. We introduce the use of depthwise and separable convolutions to construct an EEG-specific model which encapsulates well-known EEG feature extraction concepts for BCI. We compare EEGNet, both for within-subject and cross-subject classification, to current state-of-the-art approaches across four BCI paradigms: P300 visual-evoked potentials, error-related negativity responses (ERN), movement-related cortical potentials (MRCP), and sensory motor rhythms (SMR).MAIN RESULTS: We show that EEGNet generalizes across paradigms better than, and achieves comparably high performance to, the reference algorithms when only limited training data is available across all tested paradigms. In addition, we demonstrate three different approaches to visualize the contents of a trained EEGNet model to enable interpretation of the learned features.SIGNIFICANCE: Our results suggest that EEGNet is robust enough to learn a wide variety of interpretable features over a range of BCI tasks. Our models can be found at: https://github.com/vlawhern/arl-eegmodels.

Next Generation Imaging in Consumer Technology for ERP Detection based EEG Cross-Subject Visual Object Recognition

Rapid Face Recognition Based On Single-Trial Event-Related Potential Detection Over Multiple Brains

An Iterative Approach for EEG based Rapid Face Search: A Refined Retrieval by Brain Computer Interfaces

A Rapid Face Recognition Bci System Using Single-Trial Erp

ERP Component Analysis for Rapid Image Searching in Finer Categories

Decoding Visual Recognition of Objects from EEG Signals based on Attention-Driven Convolutional Neural Network

Decoding Brain Signals from Rapid-Event EEG for Visual Analysis Using Deep Learning

EEGNet: a compact convolutional neural network for EEG-based brain–computer interfaces

Brain State Decoding for Rapid Image Retrieval

Applying advanced machine learning models to classify electro-physiological activity of human brain for use in biometric identification

EEG-NeXt: A Modernized ConvNet for The Classification of Cognitive Activity from EEG

A Contrastive Learning Based Convolutional Neural Network for ERP Brain-Computer Interfaces

Event-related brain potential markers of visual and auditory perception: A useful tool for brain computer interface systems

A Robust 3D-Convolutional Neural Network-Based Electroencephalogram Decoding Model for the Intra-Individual Difference

EEGNet: A Compact Convolutional Network for EEG-based Brain-Computer Interfaces

A Multi-Scale EEGNet for Cross-Subject RSVP-based BCI System.

Towards Fast Single-Trial Online ERP based Brain-Computer Interface using dry EEG electrodes and neural networks: a pilot study

Spatial Attention-Enhanced EEG Analysis for Profiling Consumer Choices

A large and rich EEG dataset for modeling human visual object recognition

A Spatiotemporal Approach to N170 Detection with Application to Brain-Computer Interfaces.

Classifying four-category visual objects using multiple ERP components in single-trial ERP