Multispectral Accuracy

Good Day,

I am posting because I am not feeling confident in the Plant Health report of WebODM. I really want it to be accurate but I am having doubts. The problem that I have is that when I run my multispectral dataset from my Micasense Rededge MX I do not get results that are similar or consistent with the results from DroneDeploy, Pix4D Mapper or Fields. I have data sets from June 22 and July 6 that I processed on all three programs for comparison. I have one from a couple days ago, July 21 that I only ran in WebODM. The July 21 data is suggesting that the gravel road is alive and well and in a state of photosynthesis.

Some examples from June 22, 2022:
This Image is a screen shot of Pix4D Fields. I think the thing to keep an eye on is the histogram results defining values returned. As you see it is an NDVI map.

This is the WebODM using the same data. The histogram suggests a totally different set of values. On this map, the gravel road surface is returning an NDVI value of about .4 or .5. That just does not make sense to me.

In this next set from July 6, 2022, I am able to set the WebODM histogram values to match the default values selected by DroneDeploy, from -0.11 - 0.62. The WebODM is a little better this time, but the concentration of values on the histogram is quite different than the DD results. The DroneDeploy NDVI map really was bang on for this flight I think. A great representation of the field. The first picture is the WebODM Ortho of the flight using the RGB of the Micasense.

And finally a couple days ago, the result that I do not feel at all good about. July 21 flight. I will attach the ortho first for help to understand the NDVI. The Ortho is from An X4S camera on the same flight. The WebODM NDVI just does not make sense to me. I am going to mention the road surface again, it is represented as being alive. I have just tuned the histogram to concentration of values returned.

To be honest, I am really hoping that I am making an error in some way of my interpretation. If that is not the case, what do others think about what I am seeing?

I think there’s likely some misinterpretation happening here plus perhaps the opportunity for some improvements to how we deal with multispectral datasets.

The web interface, for good and bad is assuming a particular band order that may not apply to your dataset. This can provide some challenges in interpretation and use of the correct ratio.

First thing’s first though: did you process your multispectral data in or your RGB data?

The multi spectral images I posted are processed using only the Micasense Rededge MX. The only image posted using an RGB camera was the ortho that I identified as being from the X4S.

I’m not sure what you mean by misinterpreted. It’s an NDVI formula. It would be reasonable to expect NDVI values as all I am interpreting. Sound fair? The only thing I really pointed out as an example was what the grid road was returning. It matched the spectrum/values of a positive number.

Like I said, I hope I am making a mistake or misunderstanding something. I just can’t find it yet.

Cool. Just wanted to make sure. I both assume that everyone knows what they are doing, but also ask lots of questions just in case, kind of like trust but verify. We expect everyone to arrive with some expertise, but also that there’s lots to learn for all of us.

Side note: I usually leave multi-spectral questions to another here, but he is taking some time off, so I’m pinch hitting a bit and researching as I go. We’ll be learning a bit together here.

Did you check the band ordering of your NDVI? You’ve got two things controlling the creation of NDVI in WebODM: Algorithm and Filter. You want to make sure your NIR (N) and Red (R) channels are correct in the Filter drop down (the green doesn’t matter, in this case).

image

If those are right, and you are still getting wrong values, then we need to go deeper.

You are on to something here.

From Micasense:

With Filter order corrected:

This is looking like it is bringing values closer, could be exact or close, I am not sure yet. I am going to need a good look. Wouldn’t it be nice if any of these had a mouse over value display of a location.
So you brought this right to a head, thanks. I was making an error. Thought I would update.

1 Like

Awesome! Glad we got you there. It’s reasonable to expect that as OpenDroneMap as an engine can handle your sensor, that this would be automatically set. But as the WebODM interface is lightly coupled, it does not yet in this case.

If you get a chance, it might be useful to review docs.OpenDroneMap.org and if this isn’t described or addressed well in them, create a pull request at GitHub - OpenDroneMap/docs: 🎉 Contribute to OpenDroneMap's documentation! Read how below! 🎉 describing this idiosyncrasy. I am happy to help with that pull request if needed.

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.