Public Lab Research note


Mobius Actioncam with WRATTEN 25a filter NDVI analysis on PhotoMonitoring Plugin

by silvanhi | August 22, 2014 12:07 22 Aug 12:07 | #11072 | #11072

Hello everybody. I really appreciate this useful and open-minded communtiy. I spent the last couple weeks reading many articles on NDVI photography and made some great progress with all the information provided on this site. Now I'm at a point at which I need some specific help from experts.

What I want to do

I'd like to make a big NDVI Map with my quadrocopter. The Idea is to support the farmers around me with useful information.

My attempt and results

I did make some progess with a proper white balancing and re-focusing the mobius action Cam. I managed to produce relatively accurate NDVI photos with the Infragram Sandbox-Tool and a custom HSV-Setting. The H-Value is [-(R-B)/(R+B)*4] ; S=V=1. You can see these results in the attachement.

Questions

Even though I've got some good results with the Infragram Sandbox, the resolution of my images gets really decreased. So I installed Ned Hornings PhotoMonitoring Plugin on Fiji to get the same result but with a much better resolution. Unfortunately I'm not able to reproduce these results. I can't include my custom HSV-Formula i used on the sandbox. I'm also not sure which .lut file and ColorIndex values I have to use to get the same results like on the sandbox. The best approximation i got is in the attachement.

Additionally I can't understand why in my picture with the sandbox high photosynthesis areas are gree represented and not red. I would be really thankful if somebody can support me.

Thanks.

Here you can see the scenery (taken with an iPhone).

photo.JPG

Here is the picture I get in he plugin with the questioned settings. Screen_Shot_2014-08-22_at_13.34.26.png

Screen_Shot_2014-08-22_at_13.34.43.png

This should be the result: 42.jpeg

A quick follow up question: The defautl .lut in the infragram sandbox is the following i think. Screen_Shot_2014-08-17_at_20.34.37.png

Even tough a high NDVI value (red) represents high photosyntesis, it does exactly the opposite in my picture. green is high photosynthesis and blue/red is low as you can see. Why is that?


10 Comments

Hi Silvanhi,

That's a gorgeous landscape. And a great subject for ndvi images. There are a couple of things I don't understand about your results.

  1. I don't know what color look-up table you used in the Fiji plugin -- I don't know what infraBlueNDVI.lut is.
  2. The NDVI color key you include above is for the lut called NDVIBlu2Red.lut. That lut can be downloaded here. That lut does not include magenta, which is present in the NDVI image above it. So that color key does not belong with that NDVI image.
  3. You are correct that your results from Infragram.org do not look correct because hotter colors (reds) are representing non plants. I don't know which of several likely explanations is responsible. I think it's possible to apply multiple operations on a photo at the infragram sandbox and produce a meaningless image. I'm not sure there is a way to revert to the original uploaded photo.

If you include one of the super-red photos from your Wratten 25A Mobius, we might be able to do better troubleshooting.

In Fiji, right after you produce an NDVI image, if you add about 25 to the value in each pixel (process/math/add) you can slide all the colors in the NDVI image to the right on the NDVI color scale. If you have to do that to get an NDVI image that is meaningful, then the super-red photo from the Mobius might need a different white balance setting for that type of scene.

Chris

Reply to this comment...


Hi Chris, Thanks for your fast answer.

1.) Here is the infraBlueNDVI.lut; I think its quite the same as the NDVIBlu2red.lut Screen_Shot_2014-08-23_at_11.03.47.png

Screen_Shot_2014-08-23_at_11.25.36.png

2.) Ok. I use from now on the NDVIBlu2red.lut in the Fiji app. What do you mean by not having magenta in it? How can it be that an InfragramSandbox image with the NDVIBlu2red.lut Color Key does produce magenta? The only possible answer is that the HSV-formula [-(R-B)/(R+B)*4] ; S=V=1 ; does that. I'm a bit confused since the NDVI it produces does quite a good job representing high/low photosynthesis areas.

3.) thanks for your thoughts on that.

Here is my original superred photo taken with Mobius and Wratten 25a filter: IMAG0099.jpg

I processed this foto in PhotoMonitoring with the NDVIBlu2red.lut and added afterwards a value of 40 to it. now it looks like that: 99_-1_1_40.tiff

Do you think it represents good NDVI values? There are quite a few questionmarks e.g. if you look at the small house at the lower left corner. The wall is red in the NDVI which represents a high NDVI value. But there is no photosynthesis on that. Even tough I have to say that it is painted dark-greenish. Can I improve things like that? An other example would be the tree trunk. It is also red but doesnt make any photosynthesis.

Here are my white balancing settings for the Mobius. I did over one hundred trial-and-error-tests to get the best possible values with the InfragramSandbox and the upper HSV formula. Do I have to make the same tests again with the fiji app? Screen_Shot_2014-08-23_at_11.52.52.png

Silvan

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Silvan,

  1. Okay, infraBlueNDVI.lut is the same as NDVIBlu2red.lut
  2. Infragram does not use the luts that are used in Fiji. Infragram uses its own look up tables, and the default is one based on NDVIBlu2Red, but I guess it has magenta in it. Maybe someone who understands infragram.org knows why you got those unhelpful results.
  3. Your super-red photo does not look blue enough, and it looks over exposed in the bright areas. The super-red photos that produce the best NDVI images have vegetation with a nice blue hue like the ones in this note and this note. Did you try the Mobius settings in the screenshot in this note?
  4. Your super-red photo does not have much dynamic range, and the resulting NDVI does not have much dynamic range. So all the foliage in your scene has about the same color and ends up with the same value for NDVI. Many non-foliage things in the photo also have a similar color and end up with similar NDVI values.
  5. The wall and tree trunk in your photo are very dark and produce high NDVI values. That is a common artifact. It just happens that some dark areas have very low values for red (e.g., 4) and slightly higher values for blue (e.g., 20) so NDVI computes to a high value. Very dark and very bright areas of super-red or infrablue photos rarely produce reliable NDVI information.
  6. In general, single camera NDVI photos must be taken carefully and strategically to get meaningful NDVI results. Exposure, white balance, and scene contrast all have critical impacts on NDVI.
  7. When you find a good white balance setting for the Mobius, it should work in Fiji and any other program you use. You should not need separate white balance settings for different analysis protocols. It might be that slightly different white balance settings produce better results under different lighting conditions.

Chris

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Thanks Chris for your information.

  1. I tried the values from your note initially for the HSV formula but they didn't work out so good so I tried my own. I thought they are also the best for the Fiji plugin but they aren't apparently. So I filled in the white balance settings from your note again and they work much better.
  2. Thanks for that info!

6./7. Alright. I also thought on having a light sensor at the same time each photo is taken. And with a propper script the information from the sensor can produce a better white balancing and/or useful information for post-processing. What do you think on this approach?

Here are my results. What are your thoughts on these? In my optinion they are pretty accurate. :)

IMAG0110.JPG

IMAG0110_NDVI_Color.tiff

An other thing I have in mind is to use a double camera setup (two mobius: one with filter and one without). Do you think that I will get even better results with such a setup?

Silvan

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Silvan,

Your new results look really good. You can probably improve on those settings for any particular scene, but it will require more trial and error. If you had a sensor that could tell you how to adjust the settings, that would be a really smart sensor. If you had a sensor that smart, you probably would not need the camera! And the Mobius might not be the best camera for this type of automation. That seems like a job for the Raspberry Pi camera.

If you put some targets of known spectral reflectivity in your photo scene, you might be able to calibrate the NDVI values. But so far that has been done most successfully with cameras that record RAW image data (e.g., http://publiclab.org/notes/nedhorning/07-10-2014/using-a-raw-image-to-calibrate-a-jpeg-image).

Two-camera systems have the potential to produce much more meaningful NDVI information. For flying cameras, the shutters must be synchronized so the photo pairs can be aligned. The Mobius may not be the best camera for that type of control. I am not sure how you could get perfect shutter synchrony with Mobius cameras.

The Mobius lens is also very wide, so stitching maps with overlapping nadir photos will be problematical. Mobius aerial photos might be better suited for structure from motion production of three dimensional topographic models (discussion here: http://publiclab.org/notes/code4maine/06-25-2014/kap-test-for-invasives-monitoring-project).

Chris

Reply to this comment...


Here is an NDVI image of your latest scene using a new lut I have been trying.

SilvanNDVI_VGYRM.JPG

I'll post a research note about this lut soon.

Reply to this comment...


really intersting. i dont no how to make a lut but i'd love to have one. My idea is to mirror the values in the NDVIBlu2red.lut from NDVI 0 to 1 and values <0 rest as they are now. so areas of high photosynthesis should be green and low photosynthesis is red. is something like that possible? In my opinion it is more intuitive that healthy areas are green and not red..

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


I posted a note about the new lut. There is a tab delimited file of the colormap in a comment there. I will put the lut file for use in Fiji there when I figure out how.

It's easy to make your own lut in Fiji. With any lut active (e.g., make a new NDVI image), select Image/Color/Edit lut. Click on any of the 256 blocks to change it's color, or select any range of blocks. Then you will be prompted to select a color for the first and last block of your selection. It will fill in the gradient between those two colors. When you have the look up table you want, save it.

For discussing our NDVI work in the Public Lab community, it would be good if everyone was using the same lut, or maybe a small selection of luts. In any case, always include a key to the colors when sharing any color mapped image.

Reply to this comment...


Hello. how to calculate the chlorophyll content in terms of NDVI?

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


I am doing about the same thing as you: use small quadcopters with small cameras with red/blue filters to analyze NDVI for small farmers. I posted my first two notes NDVI with a key chain camera to use with a mini-drone - First steps and NDVI with a key chain camera to use with a mini-drone - Second steps with subjects filled with more or less similar vegetation and too close to them. Processed images were not so resolved or detailed. I am using instead two small 1080p cameras from the type used in small spy key chain cameras. Not the best but an inexpensive beginning. I am still figuring out how to use and compare results from Infragram Sandbox and Fiji/PhotoMonitoting plugin. I really appreciate the work you are doing, it is helping me to sort some of my problems.

Reply to this comment...


Login to comment.