This short note focuses on a pattern recognition layer behind a sensor system, particularly where the sensor gathers mixed and noisy data. Let's use a NIR detector system .8-1.8 microns. A distinguishing feature of this window is that it does not have the sharp resonance peaks seen at longer wavelengths. What is often seen is a back-end layer involvig a pattern recognition engine. This entire system has been used for discriminant analysis for say olive oil, even though the spectrum looks featureless.
Here is one simple analysis of how pattern cognition works on sparse, low rssolution data. The following doodle:
Suppose you divide the NIR band into 10 even regions. And each sub-band has only 32 or 2(exp5) levels of resolution. So the number of possible patterns is (2(exp5)exp10) or 2(exp40) different patterns. So a number of low resolution channels can achieve high multiplicative complexity. Other competing signals are less likely to correlate in this way.
j