1. Quantitative interpretation of WV3 imagery over
selected exploration targets
Neil Pendock
TerraCore
neil.pendock@gmail.com
Andy Lloyd
Geologist
Hanna Mazus
Geo Data Design
2. World View 3 SWIR
Good spatial
resolution, 8 band
positions
Bad cost
Ugly tiff vs. hdf5
calibration issues
8. Quantitative interpretation: linear mixing model
Image = library x abundances
Problem with linear algebra: for linear independence
# endmembers <= NB
Not enough in case of WV3
Solution: overcomplete basis
Many machine vision algorithms
9. USGS resampled library as training set
representation against a
library of 100 “spectra”
from the image
iron sulphide
Supervised
10. Random forest classifier
100 different
classification trees
randomly choose 7
features to construct
each tree, to avoid
overfitting
choose the mode of
predictions for 100 trees
22. Big guns weigh-in
“WV-3 mineral mapping results were promising, establishing WV-3’s potential as valuable new tool
for geologic and alteration mapping—better than any other commercial multispectral sensor
currently in orbit.
WV-3 sensor is performing as expected and predicted. Extracted spectra are remarkably similar to
the simulated spectra with typically less than 5% error.
Overall, the on-orbit WV-3 SWIR data closely match expectations for mineral mapping as predicted
by the simulation, and WV-3’s carefully selected eight SWIR bands and 7.5-m resolution provide
extensive new mineral mapping capabilities not available from other spaceborne multispectral
systems.”
