Good day Wouter,
[…] would it be sufficient to analyse the layers between two of the darker vertical streaks?
For example as indicated here:
Concerning classification, as proposed by Ellen, I doubt that the lower bound of the third layer can reliably be detected. A statistical analysis seems appropriate!
Classifiers appear to do magical processing, but their basis is pure maths, i.e. statistics. The results must be interpreted correspondingly, i.e. statistically. Furthermore, and this appears to be not commonly known, the learning sample must be related to the number of parameters that are to be adjusted. The minimum size is about 3 linear independent and class-characteristic sample signals/images per parameter.
(This is the reason why layered artificial neural network classifiers with up to many millions of adjustable synaptic weights need incredibly large numbers of samples …)