کلمات کلیدی:

شبکه عصبی پرسپترون چند لایه – تفکیک ستاره از کهکشان – داده­های طیفی – انتقال به سرخ

Abstract:

There are complexity in science and calculations of normal human brain computing systems associated with fault and spend a lot of time. Today, artificial intelligence and neural networks in the world of science, as a solution to the problem is complex. This complexity is not only common knowledge of astronomy, but also one of the most complex and difficult issues that are science-world problems. In this thesis, the problem of segregation in the field of cosmology Galaxy star of spectral data has been astronomical. The problem with this research is trying to solve it, distinguish objects, stars and galaxies that are too high because they are seen as galaxies, usually in large astronomical catalogs, wrongly diagnosed. In this study, the separation of the stars of the galaxy redshifts of their spectral data and property was carried out. Property redshift of celestial bodies, is created when objects are within the scope of the universe apart. Since the universe is expanding, then usually the objects in the universe than we are redshifted. The source for data extraction redshift objects, the whole of them. The aim of this study was to detect and correct separation between floors of stars and galaxies using MLP neural network and data spectral redshift objects, objects related information. In this research with used of exact mathematical equation, the weight of neural network connection, regulated in such a way which obtained result from this network in compare with previous method, will be more accurate with less percentage of error. Basic strategy for star/galaxy separation is measurement the amount of movement between graphs of spectral lines objects with Hydrogen laboratory spectral line. In this neural network, the set of actual training samples used for training of neural network, that with preservation of accuracy in distinction and separation of objects, covering wide range of stars and galaxies. At the end of this research and with considering the results, we can find out that with presence of less error in neural network in compare with other method, eventually more exact separation has been achieved.

Keyword: