Hyperspectral Remote Sensing

Question 1

Distinguish hyperspectral from multispectral imagery - 3 things

Answer 1

multispectral instruments capture imagery in less than a dozen bands, hyperspectral instruments capture hundreds to thousands of different bands

hyperspectral instrument makes measurements across a region that may only be 10 nanometers wide

allows for a much more detailed construction of spectral signatures

Question 2

List a few unique applications of hyperspectral sensing

Answer 2

may reveal mineral composition of soils and rocks, or the biochemical composition of plants

mapping of individual plant species

agricultural applications help growers identify plant growth stages, nutritional information, and different types of plant disease and stress

determination of biochemical leaf traits

plant functional and biochemical traits such as chlorophyll content, leaf level concentrations of minerals like calcium, phosphorous, and potassium, as well as lignin and phenolic compounds

used by geologists to map the mineral composition of soils and rocks

Question 3

Describe what PCA and MNF seek to accomplish

Answer 3

they are dimensionality reduction techniques (creates new synthetic components are created as linear combinations of all of the original bands):

creates synthetic components by linear combinations of the input bands such that we can represent almost all of the original information but usually with vastly fewer individual features

The MNF transform is very similar to PCA but seeks to maximize the signal to noise ratio rather than purely decorrelate

1. PCA (Principal Components Analysis)
   - driven by the statistical properties of a particular set of data
   - construct a new set of mutually orthogonal axes through the data such that the first axis goes in the direction of highest data variance
   - minimal amount of correlation and thus contain no redundant information
2. MNF (Maximum Noise Fraction)
   seeks to maximize the signal to noise ratio rather than purely decorrelate

Question 4

Explain, in general, how SMA (Spectral Mixture Analysis) works

Answer 4

hyperspectral analyses typically seek to map much more specific features such as plant species or particular mineral types - these analyses are thus affected to a much greater degree by the mixed pixel problem

There is a suite of techniques called spectral mixture analysis that seeks to unscramble the omelet of a mixed spectral signature and provide the analyst with a measure of the fraction of each of many different constituent quantities is within the pixel

uses pure spectral endmember signatures along with some simple linear algebra to solve for the fractions of each component present in each pixel

linear mixing - radiation leaving the surface from each component remains separate before reaching the sensor

non-linear mixing - the radiation leaving the surface is mixed before it reaches the sensor

Question 5

Provide examples of existing hyperspectral sensors

Answer 5

AVIRIS and AVIRIS-NG
EO-1 Hyperion