A cross-scale composite wavefront measurement method based on deep learning is proposed to address local large gradient wavefront distortions from aero-optical effects. Since dynamic range and spatial resolution are usually a trade-off for most wavefront sensors, we propose a hybrid Shack-Hartmann-digital holographic wavefront sensing mechanism which includes Shack-Hartmann wavefront sensor (SHWFS) and Off-axis digital holography (OADH). By using the hybrid wavefront sensing mechanism and data processing method, reconstructed wavefront of SHWFS and wrapped phase of OADH are obtained separately. A multi-input efficient network called Multi System Wavefront Measurement-Net (MSWM-Net) with an attention mechanism is introduced to map the reconstructed wavefront of SHWFS and the wrapped phase of the OADH to the precise wavefront. Numerical simulations and comparisons with the Deep Learning Phase Unwrapping (DLPU) model-based phase unwrapping method and classical phase unwrapping technique demonstrate that this method resolves the challenge of mismatched data scales across the two measurement systems, enabling rapid and high-precision wavefront sensing.