Reflectance/Radiometric Calibration on MicaSense Sequoia

[gtalckmin]

Hello everyone,

Congratulations on adding the "Calibrate Reflectance" feature on the latest version of Agisoft.

A couple of questions about these features:
- When the workflow will be documented?
- After adding the calibration images (as a separate folder) + "Calibrate Reflectance Option" , does Photoscan reads the QR codes/identifies the reflectance values for each channel? Or should user have to insert the values provided by Micasense ?
- As per the calibration images, does the software masks out the non relevant part of the imagery (i.e. everything that is not the panels). Should we do it manually?
- The button "Locate Panels", locate panels+mask within these images?
- If it states that is missing the calibration file (eg. RP02-1603178-SC.csv) where should I source this file or what is the information that should be contained in it?
- At the moment, I am running the workflow, but I cannot notice any changes on pixel/reflectance values.

Kind regards & once again: congratulations.

[gtalckmin]

Also, when I export the orthomosaic, each band has values that are within 0 and 1. How can that be sorted?
Cheers

[mdc]

Hi gtalckmin,

I am trying to figure out how to use the 'Calibrate Reflectance' option with a Sequoia in 1.4.1 too. There were a couple of earlier posts about this, most notably by Sean, but these tend to relate to 1.4.0:

http://www.agisoft.com/forum/index.php?topic=8362.0

Somewhere along the line Alexey (Agisoft Technical Support) produced a procedure which I've attached.

I'm using the Airinov QR reflectance targets, but these aren't getting automatically recognized either, although the images are automatically put into a Calibration folder, presumably because of some information in the metadata. This isn't such a problem, as the procedure tells you how to mask out the non-target area for each band, and manually enter the calibration values. You can then see if things have been successful if the 'normalize band sensitivity' option has been enabled in the Camera Calibration dialog.

However, having gone through the procedure I end up with almost totally white images, which seem to be over-exposed. This also seems to have been encountered by a couple of other users (Nenad  and school):

http://www.agisoft.com/forum/index.php?topic=7730.msg38070#msg38070

I'm now going down the road of trying to correct the images outside of Photoscan using python and some information provided by Micasense, but this is a challenge:

https://micasense.github.io/imageprocessing/MicaSense%20Image%20Processing%20Tutorial%201.html

Hopefully someone will be able to explain where we are going wrong, or if there is an issue with the software.

Cheers

[Alexey Pasumansky]

Hello,

Updated reflectance calibration tutorial is available from the following page:
http://www.agisoft.com/support/tutorials/intermediate-level/

[mdc]

Hi Alexey,

Many thanks for the updated procedure.

I'm still having a few issues though. Adding the photos, the manual masks and reflectance values is fine, and when I hit OK on the Calibrate Reflectance dialog it seems to do something, so that the 'Normalize band sensitivity' option in the Camera Calibration dialog is automatically enabled. But, my images all seem to then turn white.

Any ideas on what might be causing this?

A few images are attached.

Thanks
Mark

[selimerguden]

Hey Mark,
Same thing happens to me as well. If you want to be able view your images as normal just change the brightness setting to %1000 or there is also an option to do auto adjust brightness. Changing the brightness shouldn't affect your orthomosaic result, it is just so you can view the images, which is needed especially if you are doing GCP's.

[mdc]

Hi selimerguden,

Thanks for your reply, I really appreciate people trying to help me out. I've tried your tip with the brightness setting, but this doesn't really seem to help. Instead I went through the normal processing flow, but left the Reflectance Calibration until just before making the orthomosaic. This makes working with the GCPs easier, but I still end up with very white images/orthomosaic. If I go into the Raster Calculator and just graph the B1 (green) band, then you can see that rather than being between 0 and 1, the values are huge (around 65500). It seems like something is going wrong in the calculation somewhere, but I'm stuck as the workflow is pretty straightforward. This seems like the same thing the gtalckmin was saying in the original post.

Any thoughts?

Cheers
Mark

[gtalckmin]

Hello guys,

I will try tutorial in the coming days and provide an update.

I also had the white or very dark (blueish) resulting orthomosaics (as well as white input imagery)
Something that I did different from the tutorial was that I actually ran the "reflectance calibration" as the final step (tutorial indicates it should be the first) - MDC did the same, I guess.

I will play around and provide additional input in this topic/thread.

Cheers,

[mdc]

Hi gtalckmin,

Yes, I've tried two different scenarios, running the reflectance calibration both at the start of the processing workflow, and also at the end, just before making the orthomosaic. I think that Alexey said that you can actually do this at any stage, as long as it was done before the orthomosaic. In both cases though I get the same result. It's as if the DN numbers haven't been changed at all. Hopefully Alexey can perhaps check this out for us?

Cheers
Mark

[Rexion]

Hello Everyone,

I too am getting very dark images after running calibration and following the tutorial.  Additionally, the pixel values seem like radiance values, not reflectance values (0-1).  Any insight is appreciated.

Thanks

[selimerguden]

Hey Mark and Rexion,

The values that you are seeing 65500 is the DN number not the reflectance values. You need to normalize the data in order to get the reflectance values between 0 and 1. Based on your histogram divide your max value with each band in Raster Calculator, so it would be something like B1/max, B2/max, B3/max and B4/max. As an example in your case for the Green band it would be B1/65500 in Raster Calculator.

[selimerguden]

Hey Rexion,

 If you want to be able view your images as normal under tools change the brightness settings to %1000 or there is also an option to do auto adjust brightness. This won't affect your orthomosaic but you would be able to view the input images as normal.

I have processed over 200 multi-spectral projects in Agisoft Photoscan including MicaSense RedEdge and Parrot Sequoia and I personally prefer doing all the calibration settings at the very beginning.

[mdc]

Hi selimerguden,

Thanks for the info. I am a bit confused though, as I thought that the whole point of going through this process of calibrating the reflectance (using the target) was to end up with reflectance values. I also assumed that these results would be normalised, and that we wouldn't have to use the raster calculator to do this. But then I'm still trying to find my feet with the software.

I really appreciate any insight that you can give us though, as it sounds like you're someone who knows what they're talking about!

Cheers
Mark

[mdc]

Hi selimerguden,

Sorry, a further extension to my last post:

If the image values are still in DN units (even if calibrated), then then workflow from MicaSense suggests than the values from the reflectance panel haven't been used yet:

https://micasense.github.io/imageprocessing/MicaSense%20Image%20Processing%20Tutorial%201.html

As the suggested workflow is:

• Calibrated digital numbers are converted to radiance units by B1/(2^16) (for 16-bit images)
• The radiance image is converted into reflectance using the radiance values of the panel image of known reflectance to determine a scale factor between radiance and reflectance. With the final values usually in the range between 0 and 1.

If this is the case, then what is Photoscan actually giving us?

I think i'm even more confused!

Cheers
Mark

[gtalckmin]

Hi selimerguden,

Sorry, a further extension to my last post:

If the image values are still in DN units (even if calibrated), then then workflow from MicaSense suggests than the values from the reflectance panel haven't been used yet:

https://micasense.github.io/imageprocessing/MicaSense%20Image%20Processing%20Tutorial%201.html

As the suggested workflow is:

• Calibrated digital numbers are converted to radiance units by B1/(2^16) (for 16-bit images)
• The radiance image is converted into reflectance using the radiance values of the panel image of known reflectance to determine a scale factor between radiance and reflectance. With the final values usually in the range between 0 and 1.

If this is the case, then what is Photoscan actually giving us?

I think i'm even more confused!

Cheers
Mark

Hello Mark,

You are right on your considerations. If pixel values are still in DN, no reflectance transformation / radiometric calibration was performed. 
Dividing by the max value of each band (2^16) assumes that the max value is equal to the irradiance; which is most likely not the case.  Also, this would not use any of the EXIF parameter, such as irradiance meter, cosine effect etc. Thus, much of the fine tuning of Mica Sense sensors would be rendered irrelevant.

Mark, you do have the reflectance values of your  AirInov targets?
Has someone actually succeed in going through this process?

Cheers,
Gustavo