Striping artifacts when orthorectifying drone imagery

[colbyrand]

Hello, I am curious if anyone else has encountered any striping artifacts when orthorectifying drone imagery, where parallel flight lines alternate between relatively darker to relatively lighter reflectance. These artifacts are affecting my results when applying spectral indices to the orthomosaic to classify various surface features.

For reference, I am working with multispectral imagery acquired by a Micasense RedEdge-MX Dual Camera system, which was mounted to a DJI Matrice 350 RTK. The mapping survey took place over the Lemon Creek Glacier in Southeast Alaska, and the images were acquired over a three day span since we were mapping such a large area. As such, there were differing lighting conditions encountered during the course of the survey.

My first thought was that differing sun angles were causing the striping artifacts when the drone flew in a westward direction vs an eastward direction. However, I also captured imagery with a DJI Mavic 3 Mutlispectral over the same area and time span (see the second attached image), and this orthomosaic does not show any striping artifacts at all. Therefore, I am not sure why the striping artifacts would show up in the Micasense orthomosaic, but not the Mavic 3M orthomosaic. Could this be an issue related to the Micasense camera system?

If anyone knows what might be causing these striping artifacts, please let me know!

Thanks,
Colby

[dsmapping]

Sorry this happened, super frustrating when this stuff happens.  I dont know anything about multispectral but could the homogenous landscape combined with the lighting have impact? What settings did you use when you imported and aligned the images?

[colbyrand]

If the homogenous landscape was an issue, I'm not sure why the Mavic 3 Multispectral orthomosaic does not show the same striping artifacts as the Micasense orthomosaic, despite showing the same area. I imported the multispectral images as a multi-camera system, and then I aligned the photos using high accuracy, generic preselection, a key point limit of 40,000, a tie point limit of 4,000, and adaptive camera model fitting.

[James]

Were cameras gimbal mounted on both drones?

I guess the Mavic 3 would have been, allowing it to maintain a perfect nadir angle, whereas if the Micasense used a fixed mount then if it was anything other than straight down as it flew Eastward, it would be off by the same amount in the opposite direction on the return Westward (assuming that the drone rotates 180 to return, and that the pitch of the drone itself is negligible? - I'm not a drone person!)

So if the Mavic maintained a perfect nadir angle, and the Matrice/Micasense combo didn't, then you would expect different reflections on alternating stripes with the latter which wouldn't be evident in the former.

Did you try calibrate colors, or check if the images on alternating stripes were visibly different side by side?

[jmgc]

Interesting point.
As James suspects, I guess the problem lies in the photos.
I deal with this same issues for a long time (vegetation indices).
Please see the image attached. Captured with an M300/Micasense Altum, no gimbal, 180 turns each stripe.
It's very clear the differences in color/light/shadow effect from side to side.
Image estimated orientation: Yaw = 1.032, Pitch = -5.591, Roll = 2.294

Also attached two NDVI images: mosaic and average blending.

I believe the gimbal itself can't solve this problem. I still see the same effects on images with Pitch ~ 0, Roll < 2.

Solutions post-capture:
Average blending (Metashape), Full blending (Pix4DFields) are ways to soften the problem, with high costs in some situation (adulterated radiometric values, loss of resolution/detail).

Would be great to have more investigation on this matter, preferably to avoid this to happen to the images during capture.