NASA SOHO Comet Search

Helping NASA Discover Brand New Comets Through Innovative Data Science Approaches and Algorithms.

The Challenge

NASA wanted to use Artificial Intelligence (AI)/Machine Learning (ML) to automatically detect comets in data recorded by the Large Angle and Spectrometric Coronagraph (LASCO) telescope on the European Space Agency (ESA) and Solar and Heliospheric Observatory (SOHO) at NASA Goddard Space Flight Center. SOHO’s Large Angle and Spectrometric Coronagraph, or LASCO, is the instrument that provides most of the imagery, with two coronagraph telescopes designed to block direct blinding sunlight and observe the much fainter solar corona and solar outflows. As an unintended consequence of the instrument’s sensitivity, LASCO also detects large numbers of previously unknown sungrazing comets. Comets in SOHO/LASCO data are dynamic and morphologically diverse objects, and thus computationally highly complex to detect and track.

NASA challenged the Topcoder community to develop solutions that would reduce background noise in the LASCO data, in order to enhance the visibility of comets, and facilitate comet discovery and tracking.

The Solution

Topcoder hosted a Marathon Match Challenge that attracted top Data Scientists in the community. The challenge resulted in seven quality solutions with different AI/Ml approaches for image processing and comet detection. The top two challenge winners both have over a decade of experience in software engineering, and the third place winner has a PhD and is working in a field that has similar detection needs.

Topcoder also developed a tool to compare the output files of several algorithms for cross-checking “false positives” that may actually be previously unidentified comets, and the algorithms output images to aid human verification.

The Impact

NASA was thrilled with the results. They love that Topcoder’s community used different approaches to identify possible comets. Through the use of the crowd-crafted algorithms, two previously unidentified comets were discovered, including a difficult-to-detect non-group comet! They found that some solutions were also sensitive to other solar outflows, which may be useful in other heliospheric domains.

The solutions Topcoder developed modernize and improve NASA’S older image processing tool. While comets were the use case specific to this challenge, NASA has expressed interest in expanding upon this technology for other celestial object identification. These algorithms will be helpful in detecting comets that are difficult to detect by humans, adding a powerful tool to NASA’s study of our solar system and beyond.

Highlights

  • 596

    Contestants
    From 74 Countries

  • 27

    Unique Submitters with
    212 Submissions

Winners

1st place: ceres1
2nd place: gardn999
3rd place: selim_sef
4th place: cannab
5th place: ZFTurbo
6th place: ap31
7th place: kruntuid