ИСТИНА |
Войти в систему Регистрация |
|
ИПМех РАН |
||
Today, astronomers are faced with the challenge of handling vast volumes of data, as modern instruments are capable of generating terabytes of data in a single night. One such instrument is the Zwicky Transient Facility, an automated sky survey that can detect approximately a million candidate astrophysical objects among the observed regions of the night sky in a single night. However, a significant portion of the detected objects turn out to be artifacts, i.e., phenomena with non-astrophysical origins. Therefore, specialists must invest time in manually classifying these objects, as there is currently no efficient method that can perform this task without human intervention. The goal of this work is the development of an algorithm to predict whether the light curve from the Zwicky Transient Facility data releases has a bogus nature or not based on the sequence of frames. A labeled dataset provided by experts was utilized, comprising 2230 frames series. Due to the substantial size of the frame sequences, the application of a variational autoencoder was deemed necessary for mapping the images into lower-dimensional vectors. For the task of binary classification based on sequences of compressed frame vectors, a recurrent neural network was employed. Several neural network models were considered, and the quality metrics were assessed using k-fold cross-validation. The final performance metrics, including ROC-AUC=0.869 and Accuracy=0.804, suggest that the model has practical utility. The code implementing the algorithm is available on GitHub.