I got curious about the zero-padding of EEG signals, if I have two events that mark an area of interest, and I cut these specific areas of interest, they differ in their length by a few samples.
So, if I wanted to make all the epochs the same length, can I pad them with zeros or will that affect the overall results?
To be more specific, if the sampling rate is 500Hz and the area of interest is supposed to be 2 seconds long, so I expect to have 1000 samples, but I end up with 970 or 1100 samples, what would be the best approach to make them all the same length?
Thank you in advance.
Padding means the addition of bits. Hence, padding may affect the classification.
I think it depends, with my latest experiments on my paper, I equated the length of the samples by padding with zeros or pruning at the start and the end of each row data to have the same length for all samples.
In my case and based on my analysis it didn't have any significant effect.
For your reference, you can check page 122866 and Figure 14.
https://ieeexplore.ieee.org/iel7/6287639/9312710/09522131.pdf
Note: I applied on Speech signal samples.
Take care that padding may result in sparse matrix which is not easy to handle and requires deep mathematical analysis
Theoretically adding additional zeros will not provide any additional information to the signal. But for the consistency in feature dimensions, it is a practical approach. If you add zeros to some of the features of a particular class and not to others, you may end up with a biased training model. So better add zeros consistently to all features or truncate or window all the signals with the same window length.
Hope this comment may help.
- Badges
- Science topic