I am asking a general question, using specifics as examples. For instance, If I sampled flours – wheat, rye, millet, buckwheat,…. and created a good categorical model that could identify each. If I then mixed two different flours together, how can I create a model that will say ” 50% rye, 50% millet”? I think right now the models just try to find the best categorical fit and does not try to match linear combinations of categories, right?
Thanks, Ayelet. I think that is a problem. If I had 5 different grains and mixed only some, I’d need to create and check the mixture against 5 models to determine composition? If each grain had a unique spectral signature, I would expect a that single good model where each grain is well characterized should suffice to determine the composition of a mixture. Think of a simple mixture, say baking soda and wheat flour. Or ethanol and water. Each should be clearly distinguishable in a spectrum. The respective fraction of each component should be determinable from a single analysis by the relative peak (or peaks) areas.
I cannot believe I am the only person interested in such a capability, and I suspect the modeling is not too complicated. Can Consumer Physics consider working on this?