Data centric AI in problems with high labeling cost

Particularly in machine learning from satellite images for agricultural crop recognition, labeling is reserved for experts and obtaining ground truth can be too expensive. How to do data centric AI considering this difficulty?