Getting Hyper(spectral)

As an optical engineer, I have a professional interest in all sorts of optical systems. I currently don't do optical design on a regular basis myself, but sometimes I have to dig into some particular aspect of optical design to develop a presentation or prepare for a customer visit. It is such a circumstance that is leading me to dig a bit into hyperspectral imaging systems. These are cool optical systems that also connect nicely with my interest in space exploration. The picture above is the spectrometer part of a sample system I've been playing with today (designed by a colleague for demonstration purposes).

Spectrometers are pretty well known in astronomy and other technical fields (e.g., chemistry). A spectrometer uses a prism, or more commonly a diffraction grating, to spread incoming light into its component wavelengths (corresponding to colors for visible light). Spectra can be collected from the sun, from stars, from flames, light bulbs, etc. and when properly captured, recorded, and analyzed, they can reveal information about materials present in the light source (and possibly also of a medium in between the source and the instrument, e.g., the Earth's atmosphere, which selectively passes or absorbs certain wavelengths emitted by the sun). Spectrometers can also look at reflected or scattered light, and if the properties of the source are known, changes in the spectrum recorded from the reflected light can provide information on what chemical substances must have been present in the object that reflected or scattered the light.

Imaging spectrometers also do this, but they do it in a way that captures both an image of the source ("spatial data") and spectral data from specific parts of the image. So instead of recording a spectrum of the whole scene or object (which would have to be some sort of average of everything in the scene), an imaging spectrometer records a separate "rainbow" for each pixel in the image.
If you scale this up to where you are collecting such pixel-by-pixel spectra for many narrow wavelength bands across a wide range of wavelengths (maybe ultraviolet to infrared), you are now in the hyperspectral imaging domain, and this is especially exciting for remote sensing of the Earth and other planets. Instruments of this type are carried on the Mars Reconnaissance Orbiter (CRISM) and on various Earth remote sensing satellites (and aircraft) to help map the locations and types of minerals and also vegetation (at least on Earth). The amount of data that is collected can be huge if you have high spatial resolution (many megapixels) and high spectral resolution (many wavelength channels). This data can be depicted visually as a "datacube," as in the AVIRIS example at left (the 224 contiguous wavelength bands form the "depth" of this cube). This article is a good brief overview of hyperspectral imaging for remote sensing.