After wavelet decomposition and calculating the entropy of each sub band, it is important not to simply sum the entropy values into a single feature for the classifier. Each subband, such as the approximation and detail bands (LL, LH, HL, HH), provides distinct information about the signal or image at different frequency ranges and orientations. Combining these entropies into one value would cause a loss of valuable and discriminative information. A better approach is to form a feature vector where each element represents the entropy of a specific subband.
For example, [ELL,ELH,EHL,EHH,… ], depending on the number of subbands. This feature vector can then be fed into the classifier, allowing it to effectively learn patterns from the individual contributions of each subband and improve classification performance.
For more details, look into the following documentation:
Hope this helps!
