Searching for Sub-Second Stellar Variability with Wide-Field Star Trails and Deep Learning

TL;DR

This paper introduces a novel method combining star trail imaging and deep learning to detect sub-second stellar variability across wide fields, enabling rapid transient detection with high time resolution.

Contribution

It presents a new operational technique and neural network approach for identifying millisecond-scale stellar variability in wide-field images, using simulated data for training.

Findings

Deep neural network detects 10 ms transient bursts.

Star trail images serve as effective light curves for fast variability.

Method enhances the capability of ground-based telescopes for rapid transient detection.

Abstract

We present a method that enables wide field ground-based telescopes to scan the sky for sub-second stellar variability. The method has operational and image processing components. The operational component is to take star trail images. Each trail serves as a light curve for its corresponding source and facilitates sub-exposure photometry. We train a deep neural network to identify stellar variability in wide-field star trail images. We use the Large Synoptic Survey Telescope (LSST) Photon Simulator to generate simulated star trail images and include transient bursts as a proxy for variability. The network identifies transient bursts on timescales down to 10 milliseconds. We argue that there are multiple fields of astrophysics that can be advanced by the unique combination of time resolution and observing throughput that our method offers.