These findings highlight that fNIRS, with its portability and non-invasive nature, can be a powerful alternative to other neuroimaging techniques when paired with data analysis methods specifically tailored for it.

We combine the kernel matrices into a block diagonal matrix. Our method achieves 96% accuracy in binary classification and 65% in 8-class classification of mental imagery tasks.

We introduce a classification framework that leverages the dual nature of the fNIRS signal. This approach significantly outperforms traditional classification methods by applying separate kernel functions and regularizations to each chromophore.