The dataset consists of a large number (~600) of publicly available images of AEGIS-equipped ships, with their SPY arrays and VLS blocks highlighted. The presumption is that this is being used to train AI/ML image-recognition tools. They speculate that the author could be anything from a military or university research team to a motivated undergrad using it as a showcase for employers. It's about an order of magnitude smaller than typical ML training datasets.
To me, it looks like this could be a first step toward a ship recognition/identification tool. There are pretty well-developed ML ship detection tools to find ships in overhead imagery but not too many (in public) that do specific ship-class identification. Training one specifically to recognize only AEGIS ships seems like an odd side-alley to go down unless it was just a proof-of-concept for a more general capability to identify specific ship classes.