Fine-grained EEG classification using convolution neural network

Abstract

Brain-computer interface (BCI) is a technology that enables direct communication with machines through brain signals. As BCI technology evolves into new applications, the need for robust feature extraction technology will only continue to increase. In BCI tasks with small amplitude variations, such as low-contrast oddball classification, classification and recognition of EEG signals are challenging. Inspired by fine-grained classification in the field of image classification, this study innovatively uses and integrates some fine-grained classification strategies based on convolutional neural networks to improve the classification performance of the system through feature learning and feature fusion at part-level and multi-scale. Ten subjects were recruited to perform the subthreshold low-contrast Oddball task. The results showed that Fine-grained EEG CNN had a better performance in small-difference EEG signal classification compared with the classical EEG convolution neural network. Therefore, we provide a valuable new strategy for improving the classification performance of small-difference EEG signals.