If you're using a handheld camera, or camera on a moving vehicle, where the field of view changes drastically, then I thin it's best for you to use imfreehand() to manually find them. A demo is attached.
If you're using a mounted camera with a fixed field of view then you can just predefine a template in advance.
Trying to automatically find rows would be difficult, especially since I think you're doing this on growing crops and the "rows" are not so obvious in situations where the plants are very young (just specks in a rectangular grid), or very mature (in which case the rows may even come in contact with each other).
Another option would be deep learning but even that may not be perfect.
It would be easier in your other photos where all the plants are the same color and are in straight "rows", not curved, and are large enough that the intra-row spacing is less than the row-to-row spacing.
Another option would be to try the radon transform. I attach a demo.
