I have a large 23121x41169 stitched survey shot with a Xiaom Yi and an IR-PRO filter. I uploaded the straight camera files for stitching instead of post-processed ones from ImageJ because the stitching fails with the ImageJ images.
My plan was to then take the final stitched survey and run it through ImageJ to get an NDVI result. (I understand the limitations of this approach and that it isn't a true NDVI image I'm creating, but at least it lets the farmer see relative differences over time.) What happens when I process the entire image is that the results are just blue and green without any yellow or red components.
However, if I take one of the original, (component), images that went into the stitch and process that the same way, I get a result that has a range of blue, green, red, and yellow. I've done some further testing and extracted just a portion of the large stitched image, processed that, and still get the same blue and green result. Scaling it down to 10K pixels and 5K pixels width has the same result.
However, if I crop a more central portion that doesn't have any black around the edges, (the stitched result is an odd shape, so areas outside of the survey are black), then I get a result that is closer to what I got from the one original component image.
I can't quite wrap my brain around what is happening. The overall pixel dimensions don't seem to be an issue as much as excluding the black border, but why would the black border cause an issue. I'd post examples, but the stitched image is quite large. You can see it here on MapsMadeEasy: https://www.mapsmadeeasy.com/maps/public/a33930a59a5447eeaaae5830c144b7f5