In this work, we propose to use convolutional neural networks to detect contaminants in astronomical images. Each contaminant is treated in a one vs all fashion. Once trained, our networks are able to detect various contaminants such as cosmic rays, hot and bad pixels defaults, saturated pixels, diffraction spikes, nebulosities, persistence effects, satellite trails, residual fringe patterns, or tracking errors in images, encompassing a broad range of ambient conditions, PSF sampling, detectors, optics and stellar density. The CNN is performing semantic segmentation: it can output a probability map, assigning to each pixel its probability to belong to the contaminant or the background class, except for tracking errors where another convolutional neural network can assign to a whole focal plane the probability that it is affected by tracking error. Training and testing data have been gathered from real data originating from various modern CCD and near-IR cameras or simulated.
Link to PDF (may not be available yet): P1-10.pdf