Motor imagery EEG decoding based on multi-loss fusion FBCNet

Brain-computer interfaces (BCI) enable direct communication with external equipment, using neural activity as the control signal. Electroencephalogram (EEG) signals are usually selected as the control signal. For EEG signals obtained from a given experimental paradigm, a superior algorithm for feature extraction and classification is very significant. As one of the representative algorithms of deep learning, the convolutional neural network (CNN) has been widely used in the field of BCI. In this work, we introduce the filter-bank convolutional network (FBCNet) and propose an improved method. It mainly improves the network performance by modifying the loss function. The single loss function in the network is improved to the multi-loss fusion functions. Various loss functions are added to the network, and the characteristics of different loss functions are used to train the network to improve the network classification performance. This method of multi-loss fusion functions is validated on a dataset of 11 healthy subjects and compared with the other three benchmark algorithms. The result shows that the improved FBCNet produces a four-classes accuracy of 78.5%, which is superior to other algorithms.