lifeandlight / band-adjustment

Spectral Band Adjustments for Remote Sensing Reflectance Spectra in Coastal/Inland Waters

Spectral Band Adjustment

About

This repository contains source code for the paper:

"Spectral Band Adjustments for Remote Sensing Reflectance Spectra in Coastal/Inland Waters". N. Pahlevan, B. Smith, C. Binding, D. M. O'Donnell. Opt. Express 25, 28650-28667 (2017).

Data used in the paper to train the networks can be found here.

Usage

To create the heatmap comparison, run python3 heatmap.py

If the predictions were not yet generated, the script will create all necessary prediction files via the supplied network builds; a summary of the result statistics can then be found in the Results folder.

Once all band adjustments are made, the resulting image should appear as:

A command line interface is also available via cli.py:

$ python3 cli.py -h
usage: cli.py [-h] [-s SOURCE] [-t TARGET] [--filename FILENAME]
              [--datadir DATADIR] [--preddir PREDDIR] [--builddir BUILDDIR]
              [--gridsearch GRIDSEARCH] [--trainfmt TRAINFMT]
              [--testfmt TESTFMT]

optional arguments:
  -h, --help            show this help message and exit
  -s SOURCE, --source SOURCE
                        Source sensor [VI, MSI, OLI, OLCI, AER]
  -t TARGET, --target TARGET
                        Target sensor [VI, MSI, OLI, OLCI, AER]
  --filename FILENAME   Name of file to convert
  --datadir DATADIR     Directory data is located in
  --preddir PREDDIR     Directory predictions should go in
  --builddir BUILDDIR   Directory DNN build is located in
  --gridsearch GRIDSEARCH
                        Flag to turn on hyperparameter gridsearch
  --trainfmt TRAINFMT   Format of training file(s), with %s identifying the
                        source/target name (i.e. Rrs_LUT_%s)
  --testfmt TESTFMT     Format of file(s) to be converted, with %s identifying
                        the source/target name (i.e. Rrs_LUT_%s)

For example, to create an adjustment from AERONET -> VIIRS using the supplied in situ file, you can run:

python3 cli.py --source AER --target VI --filename "Data/In Situ/Rrs_insitu_AER"

which then creates the file "Predictions/AER_to_VI_DNN.csv".

To train a new network, use the --datadir flag to point towards the training data location, and --trainfmt to specify the file name format.

For example:

python3 cli.py --source AER --target VI --datadir "Data" --trainfmt "Rrs_LUT_%s"

The "%s" represents the sensor abbreviation, which in the above case would mean there should be two files available:

./Data/Rrs_LUT_AER

./Data/Rrs_LUT_VI

The models will then be created which learn the mapping between bands of Rrs_LUT_AER and Rrs_LUT_VI.