McWhirter, PR, Steele, IA, Al-Jumeily, D and Hussain, A (2017) The Classification of Periodic Light Curves from non-survey optimized observational data through Automated Extraction of Phase-based Visual Features. In: Neural Networks (IJCNN) . (2017 International Joint Conference on Neural Networks (IJCNN 2017), 14 May 2017 - 19 May 2017, Anchorage, Alaska, USA).

Preview

Text paper_ijcnn_final.pdf - Accepted Version Download (621kB) | Preview

Abstract

We implement two hidden-layer feedforward networks to classify 3011 variable star light curves. These light curves are generated from a reduction of non-survey optimized observational images gathered by wide-field cameras mounted on the Liverpool Telescope. We extract 16 features found to be highly informative in previous studies but achieve only 19.82% accuracy on a 30% test set, 5.56% above a random model. Noise and sampling defects present in these light curves poison these features primarily by reducing our Periodogram period match rate to fewer than 5%. We propose using an automated visual feature extraction technique by transforming the phase-folded light curves into image based representations. This eliminates much of the noise and the missing phase data, due to sampling defects, should have a less destructive effect on these shape features as they still remain at least partially present. We produced a set of scaled images with pixels turned either on or off based on a threshold of data points in each pixel defined as at minimum one fifth of those of the most populated pixel for each light curve. Training on the same feedforward network, we achieve 29.13% accuracy, a 13.16% improvement over a random model and we also show this technique scales with an improvement to 33.51% accuracy by increasing the number of hidden layer neurons. We concede that this improvement is not yet sufficient to allow these light curves to be used for automated classification and in conclusion we discuss a new pipeline currently being developed that simultaneously incorporates period estimation and classification. This method is inspired by approximating the manual methods employed by astronomers.

Item Type:	Conference or Workshop Item (Paper)
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QB Astronomy Q Science > QC Physics
Divisions:	Astrophysics Research Institute Computer Science & Mathematics
Publisher:	IEEE
Date Deposited:	07 Feb 2017 10:06
Last Modified:	13 Apr 2022 15:15
URI:	https://researchonline.ljmu.ac.uk/id/eprint/5453

View Item