An Analysis of Deep Learning Models in SSVEP-Based BCI: A Survey

Abstract

The brain–computer interface (BCI), which provides a new way for humans to directly communicate with robots without the involvement of the peripheral nervous system, has recently attracted much attention. Among all the BCI paradigms, BCIs based on steady-state visual evoked potentials (SSVEPs) have the highest information transfer rate (ITR) and the shortest training time. Meanwhile, deep learning has provided an effective and feasible solution for solving complex classification problems in many fields, and many researchers have started to apply deep learning to classify SSVEP signals. However, the designs of deep learning models vary drastically. There are many hyper-parameters that influence the performance of the model in an unpredictable way. This study surveyed 31 deep learning models (2011–2023) that were used to classify SSVEP signals and analyzed their design aspects including model input, model structure, performance measure, etc. Most of the studies that were surveyed in this paper were published in 2021 and 2022. This survey is an up-to-date design guide for researchers who are interested in using deep learning models to classify SSVEP signals.

A Convolutional Neural Network for SSVEP Identification by Using a Few-Channel EEG

“…Due to the distinct features of SSVEP signals in the frequency domain, along with the automated feature extraction capabilities of neural networks, transforming time-domain signals into the frequency domain can enhance SSVEP identification [ 27 ]. Furthermore, deep learning models utilizing frequency-domain inputs generally have a relatively simple structure [ 31 ]. Therefore, the EEG time series is transformed into its frequency-domain counterpart through an FFT.…”

“…In general, CNN-based methods tend to surpass the traditional methods. The convolutional layers in CNNs are considered to exploit the local spatial coherence inherent in SSVEP signals, making the models suitable for SSVEP analysis [ 31 ]. However, deep-learning-based methods typically require a lot of data for training and fine-tuning to achieve good results.…”

SSVEP-DAN: Cross-Domain Data Alignment for SSVEP-Based Brain–Computer Interfaces

“…This observed improvement can be attributed to several factors. The utilization of the hyperbolic tangent activation function is beneficial for training performance as it could mitigate signal drift or bias in the data by concentrating data near zero, offering an improvement in managing data variations [50], [51]. Furthermore, it aids in the normalization and constraint of high-amplitude EEG data, thereby enhancing network stability [50].…”

Assistance Device Based on SSVEP-BCI Online to Control a 6-DOF Robotic Arm

“…The effective use of such a system hinges on the patient maintaining a functional cognitive capacity to concentrate, comprehend the system’s operation, and direct their attention accordingly. In instances where these cognitive abilities are compromised, alternative approaches, such as motor imagination coupled with additional training, warrant consideration for ensuring successful interaction with a BCI system [ 9 , 10 , 11 , 12 ].…”

“…R 6 = (R 3 ) T R 6(10) Furthermore, the Euler angles denoted as ZXZ are regarded as observed in Equation(11) for the reference systems associated with the orientation, specifically in the final three links. The constituent elements of the 3 R 6 robot are established, thereby enabling the resolution of its inverse kinematics.3 R 6 = rot z (α) * rot x (γ) * rot z (β)(11) …”