Reprocessing the NEAT dataset with machine learning

The Near-Earth Asteroid Tracking survey pioneered asteroid discovery techniques. Operating from 1995 to 2007, it discovered 41,227 minor planets (Helin et al., 1997; Pravdo et al., 1999). We are reprocessing this dataset with technologies not available thirty years ago: fast processors, machine learning, and astronomy-specific software packages. Machine learning algorithms are used for star matching and artifact removal.

Image frames are being recalibrated, with astrometric accuracy to within a half arcsecond and photometric accuracy to ≲ 0.1 mag. This calibration information will be submitted to the Planetary Data System for future researchers. Asteroids are identified using machine learning. Tracklets are assembled via FindPOTATOs, a flexible linking algorithm written by N. Tan.

We will describe the algorithms used in reprocessing. When testing is concluded, the software for this project will be open source and freely available.