On closer inspection you’ll see that it is actually one small herd of cows repeated between the images and that the images do not align properly with vegetation features common in both scenes.
The software is looking for features throughout each scene that are common in both images. The cows are obviously in different places (legs…right?) in the two scenes and the feature-matching process, which looks for distinct features, struggled to find the overlapping features. It is likely that the strong contrast differences of the cows caused the software to identify and place a higher confidence on those features verses the more subtle vegetation features.
Since most of the drone use in agriculture focuses on crops and not cows, this phenomenon is fairly rare, but can occur in the case of a flock of birds within the scenes or the movement of a center pivot irrigation arm. Sometimes the difference is subtle and takes some patience to spot the errors in the processing.
We have seen this more often construction sites (just replace cows with bulldozers!)
So, what to do? Consider removing some of the overlapping images to have the herd (or dozer) in just one image and re-run the process to see if you can still create your mosaic with the reduced overlap in that area.
If your mission is to actually count cows (a growing utility for drones!) be aware that if you have a mosaic as your source, look for signs of obvious concern (in this case the thin overlap of the images).
Fly safely out there and keep finding new uses for those drones!