Search academic texts

Information × Registration Number 0221U103702, 0119U102565 , R & D reports Title Machine learning methods for binary morphological classification of big datasets of galaxies popup.stage_title Head Dobrycheva Daria V., Кандидат фізико-математичних наук Registration Date 24-02-2021 Organization Main Astronomical Observatory of the National Academy of Sciences of Ukraine popup.description2 Using the training sample of galaxies created during our work (N = 6163) and their photometric parameters, the classifiers were trained: Naive Bayes, Random Forest, Supporting Vector Machines, Logistic Regression, and K-Nearest Neighbors. Training of the Supporting Vector Machines method gave the highest accuracy of 96.4% (early morphological types of galaxies - 96.1%, late types - 96.9%). The best classifier has been approximated to a total target sample of about 316031 objects (141211 early galaxies and 174820 later). A cross-check of the target sample with data in the Galaxy Zoo 2 project (GZ2) was performed. It turned out that ~ 170000 target galaxies are in the GZ2 catalog. The target sample of ~ 316000 SDSS DR9 catalog objects was divided into a training sample (~ 170000 galaxies from Galaxy Zoo 2) and a target sample (~ 146000 galaxies with unknown morphological types). A training sample of N ~ 1800 available images of galaxies from the Galaxy Zoo 2 catalog was formed, which is more than 70% similar to the galaxies in the target sample. The program was written in the Python programming language, which used a convolutional neural network of Xception architecture to classify images of galaxies into elliptical and spiral using a training sample (N ~ 18000), and obtained an accuracy of 91.14%. An approximation of deep learning to the total number of unclassified galaxies was performed; the result is that the target sample of ~ 14600 galaxies contains ~ 68000 early and ~ 69000 late galaxy types; however, it was not possible to classify ~ 9000 galaxies that turned out to be artifacts.  Product Description popup.authors Vasylenko Maksym Yu. Dobrycheva Daria V. popup.nrat_date 2021-02-24 Close