Hi all,
Is there any standard to read the surveys made with multispectral cameras for precision farming?
I mean that a coloured map with a NDVI index tells pretty nothing if everyone can understand the map without a reference.
Even the ground truthing gives a non-standardised reference.
Any informations about this matter?
Since big players like Topcon or Trimble have a huge background, I think that some kind of reference should exist.
Replies
@ActionDrone - This is a great question/interest and a strong indicator that you're taking the right approach.
The best source of information that I've found is our local ARS, University extension offices, crop consultants with experience in Infrared/NIR and (surprisingly) the national inter agency fire center.
A standardized interpretation method that is not regional (or hardware) specific doesn't exist that *I* know of, but I'm aware of two software suites in the works that are supposed to do this. Both of them are predicting ~2 years out, which is also a good sign that they're verifying with trended data.
Since absolute reflectance values are so hard to obtain in the field, most of the processing seems to be comparative with another known indicator in the scene. Basically, every process I've seen that has an acceptable level of reliability still requires hands-on work. This isn't ideal and it's not how the process will work 5 years from now, but I believe that this is still where we are at.
Just my opinion.
Good answer, thanks @Mario
Can you please tell me which those SW are? I have experience with Photoscan Pro and Pix4D, nothing else yet.
I've read about ground truthing, that is probably the same as "on field reflectance check", so we should stay on this topic in order to find out a good solution.
I'm guessing the "on field reflectance check" means taking multiple readings throughout the field with a spectrometer. If that's the case, those points would be used similar to how GCPS are used to align or confirm/deny the accuracy of the project, only you would be doing it for reflectance.
For this to be a widespread and useful technology, the use of a spectrometer needs to NOT be a day to day part of the work flow. They should be used more often then they are for development purposes and periodically for QC, but my guess would be a quick death for UAS based multispectral imaging, if it requires a ground based spectrometer assessment for each flight.
Once again, this is just my opinion and based on what I've learned in the last 2 years. I am not an expert, but I do get to rub shoulders with some brainy men and women pretty often.
Hello, of course there are well stablished references. The most important is to consider Known Point for your local geodetic grid, normally supplied by the government or CORS stations if working with RTK GNSS receivers.
In case of CORS stations, the reference comes from the correction signal and there is little else to do for referral. BTW this has nothing to do with any Brand, it depends on the geodetic projection that every country uses (i.e. WGS84, etc) however any GPS uses the Geoidal positioning that has no loss by projection (Long, Lat, H)
If you need millimetric accuracy for your UAV for professional applications, you can review our Systen on Module receiver, specific for surveying-grade integration: http://northsurveying.com/index.php/instruments/gnss-rtk-receiver.
Best!
www.northsurveying.com
Hello Bernardo,
I think I've not been sufficiently clear. I'm aware of RTK systems but I'm not talking about geographic precision, I'm talking about a reference for the post elaboration data.
If I make a NDVI map and also make ground truthing, it's not enough to compare the same kind of crop in two different continents. Especially if I use different sensors.
Am I wrong?
It is my understanding that you will not be able to do this. Reasons such as different types of the same plant thrive in different ground/water situations.
From what I have gathered so far on the topic, ground truthing is the only for sure way to prove your data shows what is actually happening and that it is not reversed, inaccurate etc...
Please do share if you find anything useful!
Yes, indeed this is the case. The Instrument-result callibration has to be done on a per-case basis as there is no direct all-purpouse detection algorithm.