Feature Extraction of EEG signals

💬 Support & Discussions
1

Hi all,

How do I calculate the power spectral density (PSD) of eeg signal using FFT method, autoregressive models and using wavelet decomposition.

  • operating system: Windows 10

From the features generated going to try and classify them for detection of drowsiness.
Looking to extract any frequency-domain features from the original EEG signal.
Would like to know how to perform the above task.

Thanks

2

Have a look at the tutorials

https://mne.tools/stable/auto_tutorials/time-freq/20_sensors_time_frequency.html

Generally, you should be able to achieve this as follows:

  1. Subdivide into epochs of interest
  2. Create spectrogram calculation using one the functions of Time-Frequency — MNE 1.0.3 documentation (e.g. using psd_welch, which is using several overlapping FFT windows for more stable results)

The function returns psds, freqs which will give you the power spectrum for given frequencies.

Additionally, you can of course extract the raw data with epochs.get_data() and perform any calculation on them as well, e.g. autoregressive models (which afaik do not come out of the box in MNE), etc.

1 Like