2025A&A...694A.293V

Aims. To prepare for the analyses of the future PLATO light curves, we develop a deep learning model, PANOPTICON, to detect transits in high precision photometric light curves. Since PLATO's main objective is the detection of temperate Earth-sized planets around solar-type stars, the code is designed to detect individual transit events. The filtering step, required by conventional detection methods, can affect the transit, which could be an issue for long and shallow transits. To protect the transit shape and depth, the code is also designed to work on unfiltered light curves.
Methods. The PANOPTICON model is based on the Unet family architecture, but it is able to more efficiently extract and combine features of various length scale, leading to a more robust detection scheme. We trained the model on a set of simulated PLATO light curves in which we injected, at the pixel level, planetary, eclipsing binary, or background eclipsing binary signals. We also included a variety of noises in our data, such as granulation, stellar spots, and cosmic rays. We then assessed the capacity of PANOPTICON to detect transits in a separate dataset.
Results. The approach is able to recover 90% of our test population, including more than 25% in the Earth-analog regime, directly in unfiltered light curves. We report that the model also recovers transits irrespective of the orbital period, and it is therefore able to reliably retrieve transits on a single event basis. These figures were obtained when accepting a false alarm rate of 1%. When keeping the false alarm rate low (<0.01%), PANOPTICON is still able to recover more than 85% of the transit signals. Any transit deeper than ∼180 ppm is essentially guaranteed to be recovered.
Conclusions. This method is able to recover transits on a single event basis, and it does so with a low false alarm rate. Due to the nature of machine learning, the inference time is minimal, around 0.2 s per light curve of 126 720 points. Thanks to light curves being one dimensional, the model training is also fast, on the order of a few hours per model. This speed in training and inference, coupled with the recovery effectiveness and precision of the model, make this approach an ideal tool to complement or be used ahead of classical approaches.