Abstract
Owing to the current and upcoming extensive surveys studying the stellar variability, accurate and quicker methods are required for the astronomers to automate the classification of variable stars. The traditional approach of classification requires the calculation of the period of the observed light curve and assigning different variability patterns of phase folded light curves to different classes. However, applying these methods becomes difficult if the light curves are sparse or contain temporal gaps. Also, period finding algorithms start slowing down and become redundant in such scenarios. In this work, we present a new automated method, 1D CNN-LSTM, for classifying variable stars using a hybrid neural network of one-dimensional CNN and LSTM network which employs the raw time-series data from the variable stars. We apply the network to classify the time-series data obtained from the OGLE and the CRTS survey. We report the best average accuracy of 85% and F1 score of 0.71 for classifying five classes from the OGLE survey. We simultaneously apply other existing classification methods to our dataset and compare the results.
Highlights
Variable stars have served a pivotal role in expanding our knowledge about various aspects of the universe
While the 2D Convolutional Neural Network (CNN) model requires generating dm-dt mappings or bi-dimensional histograms, 1D CNN-Long Short-Term Memory (LSTM) does not require any pre-processing and is a step forward towards classifying light curves without providing engineered features or pre-processing
We find that the classification performance on the Catalina Real-Time Transient Survey (CRTS) dataset is suboptimal
Summary
Variable stars have served a pivotal role in expanding our knowledge about various aspects of the universe. We use a deep-learning framework called Long Short-Term Memory (LSTM; Hochreiter and Schmidhuber, 1997) networks which are designed for handling timeseries data and propagate learning from the data to the deeper layers These models are capable of learning long as well as shortterm temporal features and can accommodate the input light curves of varying length. CNN layers learn to generate the features efficiently and the LSTM part carries out the task of finding the correlations among different observations at varying timescales of an input light curve These are fed to a fully connected classification layer to predict the variability class. For training the classification model, we use the raw light curve data as an input to the network without any feature-extraction process This aspect gives this implementation an edge over the previous works.
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
