Apologies if this questions is not directly code related but it might be relevant considering some classification examples that I saw in your docs. In real-valued NNs we often use data pre-processing (e.g., skleanr's StandardScaler ). In case we have some complex-valued data, e.g., backscatter parameter such as S11 from a VNA (because I think your familiarized with it 😅), should we also normalize it (despite it already being between 0-1)? Also how do we normalize the complex input in these cases, is it the same as using complex batch norm layer before inputting onto the network?
I previously read your excellent paper (Impact of PolSAR Pre-Processing...) but I dont think it discusses this aspect (I'll re-read it still).
Hi again Jose,
Apologies if this questions is not directly code related but it might be relevant considering some classification examples that I saw in your docs. In real-valued NNs we often use data pre-processing (e.g., skleanr's StandardScaler ). In case we have some complex-valued data, e.g., backscatter parameter such as S11 from a VNA (because I think your familiarized with it 😅), should we also normalize it (despite it already being between 0-1)? Also how do we normalize the complex input in these cases, is it the same as using complex batch norm layer before inputting onto the network?
I previously read your excellent paper (Impact of PolSAR Pre-Processing...) but I dont think it discusses this aspect (I'll re-read it still).
Thanks in advance! Tomás