|
Interferometry delivers the highest angular resolution. It is being used extensively in radio astronomy and, since about a decade, it is becoming an important player in infrared astronomy. However, infrared interferometry is restricted to sparse arrays and no full-phase information is recovered. While imaging is arguably the most intuitive way to analyze interferometric data, recovering images from sparsely sampled visibilities is an “ill-posed" problem. The current algorithms work under the framework of using regularized minimization techniques. These algorithms strongly depend on the priors and hyperparameters pre-defined. This gives rise to ambiguities and artifacts in the interpretation of the images and limits their accuracy/precision as well as their signal-to-noise ratio if the priors/regularizers are not well-defined. Also, it means that imaging is the domain of a handful of highly experienced astronomers, thus keeping the interferometric community small. CASSINI-AUTOMAP aims at disrupting this situation by creating a novel framework for interferometric image reconstruction. This project is based on the exploitation of the compressibility of a signal (following the principles of theory of Compressed Sensing) with a novel optimization scheme supported by Neural Networks. In particular, we focus our efforts in designing a Neural Network with adaptive activation functions to find an optimal mapping system between the infrared interferometric data and the reconstructed images. The online adaptability of the Neural Network frees us from having to rely on strong priors, making the reconstructions more accurate and less dependent on users' inputs. Our preliminary network architecture has been tested with Sparse Aperture Masking (SAM) data taken with the infrared camera NACO at the Very Large Telescope and it demonstrates the potential and reliability of the algorithm by recovering the interferometric observables. Future improvements on the software aims at analyzing data from instruments like GRAVITY at the Very Large Telescope Interferometer or the Sparse Aperture Masking mode of the James Webb Space Telescope.
|
