Ponente
Descripción
The PAU Survey (PAUS) is an imaging survey using a 40 narrow-band filter camera, named PAUCam. Images obtained with the PAUCam suffer from scattered light: an optical effect where light appears where it is not intended to be. Scattered light is not a random effect, it can be predicted and corrected for. Nevertheless, currently, around 8% of the PAUS flux measurements are flagged as scattered light affected and removed. Moreover, failures to flag scattered light result in photometry and photo-z outliers. With the aim of understanding and predicting scattered light, we have built BKGnet, a deep neural network for background prediction. BKGnet is trained with 120x120 pixel stamps and their corresponding positions on the CCD. To benchmark the BKGnet performance, we have developed an skyflat correcting method to remove the effect of scattered light on images. On PAUCam images on the COSMOS field, we get a 28% improvement on average with BKGnet compared with the skyflat correction method.
