Brain activity recognition with a wearable fNIRS using neural networks

Functional Near-Infrared Spectroscopy (fNIRS) has taken the focus in the domain of Brain Computer Interfaces (BCI) in the recent years. However, there is not yet a defined standard for the treatment of data obtained through fNIRS. This study aims at evaluating the performance of deep neural networks and convolutional neural networks in the classification of different stimulus in the prefrontal cortex, caused by predefined activities: subtractions, word generation, and rest. After optimizing both types of networks, experimental results suggest deep neural networks to be more precise, but convolutional neural networks to be faster to train. It also shows that neither type of network is able to distinguish subtraction from word generation.