Deconvolution of SDO/HMI intensity and the vector magnetic field to achieve Hinode/SOT-SP data quality

Context. The SDO/HMI instrument provides continuous full-disk observations with a high temporal cadence but moderate spatial resolution. In contrast, Hinode/SOT-SP observes the Sun with high spatial resolution but at a low temporal cadence and within a limited field of view. Aims. This study seeks to enhance SDO/HMI observations by applying deconvolution techniques, specifically to the continuum intensity and full vector magnetic field in a local reference frame. Hinode datum, used as a training set, helps the model improve the resolution of HMI data to a Hinode-like level. Methods. We use deep residual convolutional neural networks trained on HMI-like and SOT-SP observations. The HMI-like observations utilise the SOT-SP observations and the exact HMI point-spread function and noise realisation. Results. The trained model successfully deconvolves both the intensity and the vector magnetic field. The model predictions and SOT-SP observations show the same level of detail. The root mean square difference between the two is 0.02 quiet-Sun level for intensity and between 40 and 45 G for vector magnetic field that is below the disambiguation noise level. Conclusions. Our model effectively addresses disambiguation noise and enables the deconvolution of the full vector magnetic field. Additionally, the modelled deconvolution provides enhanced results compared to the standard HMI pipeline products.