Finding asteroid trails in HST data with AutoML

We are using Google’s AutoML algorithm to search for asteroid streaks on HST/ACS and HST/WFC3 images using labelled asteroid and artificial satellite streaks identified by citizens in the same datasets at our Zooniverse Hubble Asteroid Hunter project, which was successfully concluded on July 2020. The results have been fantastic and show incredible potential for the application of the technique to the search of other types of objects in images. We have presented the work in a conference talk in October 2020 and are working now on three papers with the main results, but while the papers come out, we have found over a thousand new asteroid streaks and over two thousand satellite streaks, a large fraction of which are new to the literature.

Update Jan. 2022 : The first paper just came out :

Hubble Asteroid Hunter: I. Identifying asteroid trails in Hubble Space Telescope images

More info in our Machine Learning group website.