I am doing EMG classification and i process my signal via discrete wavelet transform with 4 decomposition level. Then is that i can just use the decompose wavelet for feature extraction? Or i need to do reconstruction first then proceed with feature extraction?
Hello dear
You can refer to the following link:
http://gwyddion.net/documentation/user-guide-en/wavelet-transform.html
Regards;
Zeyad Safaa Younus
Most of the DWT feature extraction methods just use decomposed wavelet components as target domain for feature extraction. It is because DWT is a frequency_pixel domain and we apply it to the signal to make a new descriptive of signal based on frequency. Hence just apply your feature extraction method to the decomposed wavelet version of signal directly if you have plan to make DWT features. Although there are many methods that after DWT decomposition they make many further filtering steps, like reconstruction on just some sub_domain ( Like HH, VH , .. ) and then extract features of that specific reconstructed version of signal, you don't need to do that. They do this just to present a filtered version of signal based on DWT domain. It is named DWT based signal filtering. Hence, If you are going to do feature extraction of DWT components directly, just make DTW decomposition and apply feature extraction to them directly.
Basically DWT based decomposition is dividing signal in sub band frequencies and features are extracted in sub bands. If desired feature is mixed frequency band then we have to filter out the undesired bands by forcing those coefficient to zero and reconstruct the desired features.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts