Readers involved in astronomical imaging will be familiar with the technique of taking multiple images and then stacking them together to improve signal strength and yield better images. Taking this technique further, many research projects have longer time frames than simply observing date nights of the same subject. This data is usually taken from different places and under different conditions. There is a problem with matching observations across all these survey runs. Researchers share a new approach to quantifying whether separate images of the same object provide additional signals or create useless noise.
Usually the images combined in astrophotographs are taken with the same telescope, so the instruments capturing the data and conditions are the same. To date, capturing data using multiple telescopes from different locations to create a single image has been an unconventional, impractical approach.
A team of researchers from the Johns Hopkins Institute identified a major problem in evaluating images from sky surveys taken over many years from different telescopes in different locations under different conditions. Matching observations of the same objects is challenging and is even more challenging when the observations are in close proximity. Existing tools for crossmatch data are available from various catalogs such as TopCat, CDS Match and Aspects, but to date, these are seeded and have higher than desired failure rates.
The team has developed a new data science approach, known as 'mixed integer quadratic constrained programming’ or MIQCP for short, which focuses on assigning a score to each pair of observations from different surveys. The assigned score measures the likelihood that the observations are the same and the score increases when the observations are closer and decreases if they are further apart.
Using their new technique, they can take observations from different studies and match objects to eliminate the task of sorting through all possible pairs. This not only speeds up the matching process, but also allows handling large amounts of data. In tests, the results are very promising. While previous approaches were still fast, but did not allow for all possible matches limiting the chance of success, something improved in this new technique.
The team emphasizes that the studies are important for understanding many mechanisms throughout the universe, not just at the macro level, but also at the particle level. Their new technique opens up new possibilities for processing image data and will further improve the way the team handles large datasets. Already the team can handle 100 lists, the current 10 has been upgraded to 20 using existing methods.
„Oddany rozwiązywacz problemów. Przyjazny hipsterom praktykant bekonu. Miłośnik kawy. Nieuleczalny introwertyk. Student.