DeepShape: Radio weak-lensing shear measurements using deep learning

Context. The advent of upcoming radio surveys, such as those enabled by the SKA Observatory, will provide the desired sensitivity and resolution required for weak-lensing studies. However, current shape and shear measurement methods are mostly tailored for optical surveys. The development of novel techniques to facilitate weak-lensing measurements in the radio waveband is thus needed. Aims. There are a few algorithms for shape measurement in the radio waveband. However, these are either computationally intensive or fail to achieve the accuracy required for future surveys. In this work, we present a supervised deep learning framework, dubbed DeepShape, that measures galaxy shapes with the necessary precision while minimizing computational expenses. Methods. DeepShape is made of two modules. The first module is a plug-and-play (PnP) image reconstruction algorithm based on the half-quadratic splitting method (HQS), dubbed HQS-PnP, which reconstructs images of isolated radio galaxies. The second module is a measurement network that predicts galaxy shapes using the point spread function and reconstructed image pairs. Results. We test our framework on a simulated radio data set based on the SKA-MID AA4 array configuration. The HQS-PnP algorithm outperforms the standard multiscale CLEAN algorithm across several tested metrics. DeepShape recovers galaxy shapes with an accuracy comparable to the leading radio shape measurement method, RadioLensfit, while significantly reducing the prediction time from ~4 minutes to ~220 milliseconds. We also demonstrate DeepShape’s applicability to shear measurements and recover shear estimates with an additive bias meeting SKA-MID requirements. Although the multiplicative shear bias is an order of magnitude higher than the required level, it can be mitigated using a shear measurement calibration technique, such as applying quality cuts.