Radiometric Correction for MicaSense Multispectral Cameras

This Python module provides a simplified alternative for converting raw images from MicaSense multispectral cameras into reflectance, or into radiance for a calibration made afterwards (for example with targets in the field), using the reference panel and, optionally, the Downwelling Light Sensor (DLS) data to correct changes in illumination. It was written for the RedEdge-P and tested only with it (see What has been tested). It can be used through a user interface (UI), opened with a launcher on Windows, from the command line (the workflow runner) or as a Python library.

Contents

• What has been tested
• Quick start
• Documentation
  • Installing and launching
  • The user interface (UI)
  • The reference panel: reflectance and two panel captures
  • Checking, accepting and drawing the panel regions
  • Optional DLS correction
  • How the reflectance is calculated
  • Python library and workflow runner
  • Other cameras and other numbers of bands
  • Development notes
• Background and credits
• License and citation

What has been tested

The program is prepared for other MicaSense cameras (RedEdge-M/MX, Altum, Altum-PT, dual-camera systems): the bands, the file names and the bands that are left out are configurable, and the bands can be read from the images themselves (see Other cameras). For now, however, it has been tested only with images of the MicaSense RedEdge-P. The presets of the other models come from the documentation of the manufacturer, not from images, and cameras of other makers (for example Sentera) have not been tried. Use “Detect from capture” and compare the result with your own data before relying on them.

The reflectance calibration of Agisoft Metashape 2.3.1 was compared with this program on a RedEdge-P flight: with one panel capture before and one after the flight, Metashape interpolates them in time, which is “Before and after, weighted by time” here; with several, it uses only the last capture before each image and the first after it, which is “Closest single capture before and after each image, by time”. The panel factors of the two programs differ by 0.6 % (see Metashape).

Quick start

1. Install Python 3.10 or newer (with Tkinter) and, in a dedicated environment, the packages of requirements.txt (or run conda env create -f environment.yml), plus ExifTool, a separate program that reads the metadata of the images. On Windows, the QR code reader (pyzbar) also needs the Microsoft Visual C++ 2013 Redistributable (x64). The details, and what to check on a new computer, are in Installing and launching.
2. Open the UI: double-click open_radiometric_ui.cmd (Windows) or run python -m radiometric_app window --root . in the project folder. It asks which Python to use, checks the packages and ExifTool, and opens.
3. Correct the example images in images/raw: IMG_0005 and IMG_0422 show the reference panel, before and after the flight, and IMG_0010 is the capture to correct. IMG_0003 and IMG_0009 (before the flight) and IMG_0423 (after it) are more photos of the panel, to try several panel captures (the example images are six captures of a longer flight; the complete flight is not included). Go through the pages of the UI, described in The user interface:
   • Inputs: the images folder (or several), the results folder, the captures that show the panel (with “Check panel captures…” to see how they agree) and the captures to correct.
   • Reference panel: the reflectance of the panel (typed, or computed from its spectral response file; the one of the example panel is in reference_panel_spectral_response), then “Check / draw panel regions…” to accept the region of the panel in each band.
   • DLS (optional): corrects the changes in illumination between the panel capture and each image.
   • Run and results: “Check inputs”, then “Run”.
4. Read the results in the results folder: a reflectance image (32-bit float values) for each band, with the name of its source file (IMG_0010_3.tif is the Red band before and after the correction), a <capture>_provenance.json for each capture with what was used to make it, and radiometric_result.json with the status of the run, the flags of the checks and the settings.

The same steps run without the UI, with the workflow runner (a command line that reads a JSON configuration) or with the Python library.

Documentation

The same pages are also published as a website, with a side bar and a search: mhdf.github.io/micasense_radiometric_correction.

Page	What is in it
Installing and launching	What the program needs (Python, packages, ExifTool), choosing the Python, the check of the computer and the Windows launchers
The user interface (UI)	The four pages with screenshots, the saved settings (configuration and profiles) and where things are saved
The reference panel	The reflectance of the panel, typed or computed from its spectral response file; one or two panel captures and how they are combined (by time or by illumination)
Panel regions	The tool that checks, accepts or draws the region of the panel in each band
Optional DLS correction	What the DLS ratio is, the checks of the DLS data and of the geometry, how the flags of a run are counted and the summary at its end
How the reflectance is calculated	The steps from the raw values to radiance (dark level, vignetting, row gradient, scale) and to reflectance, and the geometry of the images
Python library and workflow runner	The example in Python, the checks and the calibration record, and the command line
Other cameras	The band table, detecting the bands from a capture, camera presets, file names, and what the program assumes about the camera
Development notes	Project layout, how the UI is built and the tests

Background and credits

All functions are derived from the original MicaSense image processing repository but have been streamlined to mitigate issues caused by Exif metadata inconsistencies. These issues were encountered while pre-processing images captured with the camera firmware v1.4.5 in preparation for geometric correction using photogrammetry software.

For more details about the steps required for pre-processing MicaSense multispectral images see the tutorials (e.g. ‘Tutorial 1’) provided by MicaSense in their repository.

License and citation

The program is distributed under the MIT License, which also keeps the copyright notice of the MicaSense code it is derived from. To cite it, use “Cite this repository” on the GitHub page of the repository, which gives the reference from CITATION.cff, and give the version used (the provenance record of each corrected capture also keeps a fingerprint of the code that produced it).