Identification of problematic epochs in astronomical time series through transfer learning

Aims. We present a novel method for detecting outliers in astronomical time series based on the combination of a deep neural network and a k-nearest neighbor algorithm with the aim of identifying and removing problematic epochs in the light curves of astronomical objects. Methods. We used an EfficientNet network pretrained on ImageNet as a feature extractor and performed a k-nearest neighbor search in the resulting feature space to measure the distance from the first neighbor for each image. If the distance was above the one obtained for a stacked image, we flagged the image as a potential outlier. Results. We applied our method to a time series obtained from the VLT Survey Telescope monitoring campaign of the Deep Drilling Fields of the Vera C. Rubin Legacy Survey of Space and Time. We show that our method can effectively identify and remove artifacts from the VST time series and improve the quality and reliability of the data. This approach may prove very useful in light of the amount of data that will be provided by the LSST, which will prevent the inspection of individual light curves. We also discuss the advantages and limitations of our method and suggest possible directions for future work.