A group of Australian scientists has designed a system based on artificial intelligence to classify galaxies, which allows the process to be carried out in a very short time.
To do this, they used a convolutional neural network, networks that are designed mainly for classifications and image analysis. Until now, astronomers and scientists performed the classification process manually; basically, observing the images obtained, which implies a considerable time.
But with the new convolutional neural network (CNN) system proposed in the study, a classification that can take months, can now be done in just a few seconds. For example, by using an ordinary video card, 14,000 galaxies can be classified.
And that helps to solve one of the main challenges that many astronomers are facing in recent times is the fact that the number of galaxies that are observed and classified has grown. According to Mitchell Cavanagh, one of the study’s authors, this has led to what they call “citizen scientists” had to be recruited to help with the classification of galaxies.
Even so, the use of a neural network for this process entails some difficulties, which have to do with the precision that these networks achieve when analyzing an image.
However, because it is machine learning, the algorithm is constantly evolving; these are currently 80 percent accurate and, in specific cases, can be as high as 97 percent accurate.
The team of researchers hopes that, in addition to helping to classify galaxies, the developed system can also be used in other fields that require large-scale visual analysis.