New exoplanet candidates? Deep learning exploration of the SHINE high-contrast imaging survey

In the last decade, the field of high contrast imaging (HCI) has allowed the detection and characterization of massive exoplanets. This is the result of combining different technologies, such as adaptive optics, dedicated coronagraphs for HCI, and powerful image post-processing techniques. In this context, supervised machine learning has demonstrated to enhance detection performance of state-of-the-art post-processing techniques for the detection task. This is the case of NA-SODINN, a novel detection algorithm that relies on convolutional-LSTM neural networks to distinguish between image noise and the planetary signature in noise regimes. In this poster, we employ NA-SODINN to analyze the complete F150 sample of the SHINE survey using SPHERE high-contrast imager at the VLT. We observe improvements in the detection limits compared to standard post-processing algorithms, and we identify new exoplanet candidates that require follow-up to investigate common proper motion.