Feature extraction using DWT and WPT
45 views (last 30 days)
Can anyone confirm if my method of feature extraction is correct or not? I have used DWT and later WPT to decompose and extract features from vibration signals.
For DWT, I used the following MATLAB functions:
plot(cA1); title('Level-1 Approximation Coefficients')
plot(cD1); title('Level-1 Detail Coefficients')
k1=kurtosis(cA1) %E.g of one parameter used to extract some features%Approximation%
And other statistical parameters.
My extracted features, after being used as inputs for the ANN classifier, showed that approximations are better than details, and performed with a very high classification rate.
I haven't found a reason to reconstruct the signal or using any filters (i.e. Low and high filters)?!
MATLAB staff and experts, can you confirm my method or correct me please?
Bernhard Suhm on 25 Mar 2018
You could try using wavelets, the basic documentation is here: https://www.mathworks.com/help/wavelet/ref/dwtfilterbank.wavelets.html. That requires the Wavelet toolbox though, which you could get via a trial.