Large sparse aperture telescope wavefront sensing and control via pretrained neural network with attention module

The ability to detect pistons with high accuracy over a wide range is paramount to the co-phasing of sparse aperture optical systems. This paper proposes a global piston error modulation method for sparse aperture mirrors based on convolutional neural networks. The efficacy of this approach is demonstrated by the introduction of a convolutional block attention module (CBAM) with a data generalization mechanism, which facilitates the rapid and accurate learning of key features from actual co-phasing sensor images. This is achieved with less labelled data, thereby enabling the accurate detection of piston error distribution. The experimental results demonstrate that the method exhibits high prediction accuracy, enhances the piston error detection efficiency and sensing range, and facilitates global fine phase correction (<λ/80) under closed-loop conditions. The technique demonstrates considerable potential for application in the field of simplifying the wavefront sensing and modulation process of large segmented telescopes.