Mineral Classification

Classification of minerals for COAL is done via a spectral angles algorithm using Spectral Python. The first step of classification is to calculate the spectral angle between each pixel and each class in the spectral library that is being passed in. The returned NumPy array can then be used to classify each pixel by using the argmin function to get the class that has the smallest angle compared to each pixel.

The given spectral library is resampled with Spectral Python's BandResampler in order to detect and scale units, as any inconsistency in units will give wildly inaccurate classifications.

pycoal has been tested with the USGS Digital Spectral Library 06. A preliminary implementation of ASTER support has been added to version 0.5, but it is yet to be tested to the same degree.

AVIRIS images contain many pixels with no data on the outside of the flightlines, as well as occasionally inside the flightline in a few places. These are identified by a values in the spectra for each pixel. A pixel with no data in an AVIRIS-C image will have a value of .05 in the spectra and a pixel with no data in an AVIRIS-NG image will have a value of -50 in the spectra. These are given a class name of No data in the final classified image.