• 0

Question:Can anyone help me troubleshoot my NDVI imagery?

abdul is asking a question about infragram: Subscribe to answer questions on this topic

abdul asked on October 21, 2016 22:28
294 | 4 answers | shortlink

Hello all,

I have assembled a Mobius camera with Public Lab's infrared lens, have focused the lens, and went outside to take some pictures. It is autumn, so many of the plants are more orange than green, but the trees still had a fair share of green on them.

However, when I upload the images to http://www.infragram.org I am not achieving the results that I expected. Are my expectations erroneous? Other opportunities for failure were:

  • The assembly
  • The config file
  • The plants not being green enough
  • A processing glitch
  • Who knows

Here is the base image that I have taken, without any digital editing.


This image is of course taken using the public lab NDVI lens installed into the mobius camera.

When I upload it to www.infragram.org and choose the basic present for the NDVI red filter images, I am returned a mostly black screen:


When I ask infragram.org to return a colorized analysis instead of a basic one, the analysis is still a bit strange:


Does anyone know what needs to be changed here so that I can get better photos? Does anything at all need to be changed? Please help out!

ndvi calibration infrared question infragram

question:infragram question:question question:multispectral-imaging


Log in to comment

4 Answers

You are correct that all of the things you list could be responsible for the failure to produce a meaningful NDVI image. It could also be due to the color of the ambient light being strange, or the custom white balance being wrong for that lens. Without some controlled tests it will be difficult to determine what is to blame.

The photo from the camera is not blue enough to be directly converted to NDVI (i.e., with the standard formula). The blue channel, which is used for NIR, is not much brighter than the red channel in pixels capturing green leaves. If you alter the photo by making it bluer, stretch both histograms, and then increase the value in all the pixels of the NDVI image by 15, you get this:


Making photos which can be converted directly to a facsimile of NDVI requires a custom white balance appropriate for the lens, the sensor, and the ambient light. It also requires a properly exposed photo. Satisfying all of these requirements for each photo taken of plants is probably not possible if the same white balance and automatic exposure are used for all photos regardless of ambient light.


Abdul - it looks like the config file you're using is not set up with a corrected white balance.

This page has the corrected WB required as Chris outlines above: https://publiclab.org/wiki/infragram-point-shoot#Installing+config+file

It also has a "good/bad" comparison photo to help people to know better if their images are properly white balanced. If there's a better place to feature that, that'd be great -- maybe on your recently posted activity?

@warren @cfastie

So I took a new photograph with the adjusted config file, as I had indeed uploaded the wrong one.

The photo looks like this when it is unadjusted IMAG0078.JPG

But after putting it into the infragram.org basic red filter analysis it comes out mostly black again:


And after the colorized analysis it comes out like this (which looks mostly correct, but not as brilliant as the one you posted @cfastie )


Should I update any of the information in the SYSCFG file in order to make photos analyzed on infragram.org more compatible? We recently switched IR lenses and I want to make sure the final product is reliable to use and am in the process of updating the instructions so I want to get it all as correct and clear as possible for everyone!

Also if anyone has any suggestions, please state them!

Abdul, It is possible that the new lens has different coatings which alter the color balance (red:NIR). This might require an adjustment to the custom white balance setting. It might also be good to check that the same sensor is being used in the current Mobius because that could have an even greater effect on color balance. The easiest way to find a better setting is to try different ones. Try a setting with more separation between the red and blue values. The current values are red 310, green 500, blue 700.

Your latest photo does not have enough separation between red and blue (red and NIR) to produce typical NDVI values for foliage pixels. Blue values for foliage are about 50% greater than red values:


In the photo below from this Claytonb note, blue values are an order of magnitude greater than red values. This photo produces NDVI values for foliage in the typical range (0.3 to 0.8).


There is nothing sacred about the current white balance settings (red 310, green 500, blue 700). These were found to work acceptably with the Mobius two and a half years ago but I am not aware of any tests to see if other settings produce better results (they probably do) or whether different settings are required for different lighting conditions (they probably are).


Sign up or Login to post an answer to this question.