Csp-nn: a Convolutional Neural Network Implementation of Common Spatial Patterns

In this paper we propose, describe, and evaluate a novel deep learning method for classifying binary motor imagery data. This model is designed to perform CSP-like feature extractions. It can be seen as a neural network with a specifically designed architecture where the latent space corresponds naturally to the features found in CSP methods. Our model allows for easy generalization from spatial filters to spatio-temporal filters. It also allows for the feature extraction and filtering stages to be optimized jointly with the classifier. This allows standard regularization methods to include the filtering stage. In addition the network provides the expressiveness and robustness of deep learning to improve upon the efficiency of CSP filtering methods.