(23rd-March-2020)
A typical non-invasive BMI are following.
Brain wave (EEG)
Magneto encephalography (MEG)
NIRS
FMRI
In non-invasive methods, research has been advanced because the risk of damaging the brain is small and research on humans is relatively easy. However, there is a problem that the spatial resolution is inferior to the invasive type.
Feature of extraction are following.
Time frequency distributions (TFD)
Fast fourier transform (FFT)
Wavelet transform (WT)
Eigenvector methods (EM)
Auto regressive method (ARM) (autoregressive model)
Using the feature vector obtained by the feature extraction, the brain wave is connected to the operation target (for example, the brain wave and the direction of the artificial hand). The method is
Independent component analysis / ICA
k-means clustering
Support Vector Machine / SVM
There are also methods such as convolutional neural networks that train end-to-end without explicit feature extraction.
Comentarios