PCA and MCR with 3D spectrum data

[liuchzzyy]

Hi, all
Could the 3D spectral dataset be processed using PCA and MCR-ALS in spectrochempy package now? For example, I have a dataset with x position, y position, and spectrum for each point.
I check the code of spectrochempy, it seems possible to do this, but not really after I tried it. right?
Thanks in advance ~
Cheng

[atravert]

Hi @liuchzzyy ,

MCR-ALS is intrinsically a 2-dimensional method, so to apply it to hyperspectral data (shape =n_x, n_y, n_wavenumbers) you have to:

1/ unfold the hyperspectral data in a 2-dimensional NDDataset D of the spectra (shape = n_x*n_y, n_wavenumbers). You can either keep the spatial informations as coordinates in the NDDataset, or separately

2/ carry out the MCR-ALS, that will yield the C (shape = nx*n_y, n_species) and St (shape = n_species, n_wavenumbers) matrices

3/ if needed, reshape the C matrix to (n_x, n_y, n_species)

The same, I think, holds for PCA.

Does this answer your question ?

Best regards
Arnaud

[liuchzzyy]

Hi, Arnaud
Thanks for your reply! Okay I would like to try it the way you mentioned, which is of great helps!
and actually Yesterday I got a similar result with tensorly.decomposition.Parafac2. Hope it useful.

Best,
Cheng

[atravert]

Yes, PARAFAC is a factor analysis method suitable for three-dimensional data sets. Please note that, like PCA, PARAFAC does not aim to recover spectral and concentration profiles (which MCR-ALS does).
Best regards,
Arnaud

[liuchzzyy]

Thanks for this info! and I am so fresh with PARAFAC.