Deep learning bimodal frequency-selective digital holography by using a confocal dual-beam setup

Abstract

This paper proposes a bimodal digital holography technique based on deep learning, marking the first application of neural networks in frequency-selective holographic reconstruction. The method achieves dual-channel mixed-mode recording and super-resolution separation reconstruction, enabling simultaneous multimodal holography and enhancing wavefront acquisition efficiency. Direct current and conjugate terms are effectively suppressed, allowing coherent and incoherent holography integration. Experiments in overlapping and non-overlapping modes with confocal dual viewpoints confirm the method's fidelity in isolating target wavefront spectra and demonstrate improved resolution after deep learning reconstruction. This technique offers broad potential in multimodal compressed holography, super-resolution, and extended field-of-view imaging.