Coherent diffraction imaging under broadband illumination with a classified monochromatization deep learning strategy

Coherent diffraction imaging (CDI) is a promising lensless imaging technique that can reconstruct the structure of objects by processing the recorded diffraction pattern. High-quality CDI reconstruction typically requires illumination from a coherent light source with high temporal coherence. However, in CDI under broadband illumination, a broader spectral bandwidth will lead to ambiguity in the reconstruction results of conventional phase retrieval methods. Here, we propose a broadband CDI method based on a classified monochromatization deep learning strategy. First, we introduce a classification neural network that predicts the spectral range category of unknown broadband diffraction patterns. Subsequently, several monochromatization neural networks aimed at specific spectral ranges are employed to retrieve the corresponding monochromatic pattern more efficiently. With appropriate training, the classified monochromatization deep learning strategy is applicable to broadband CDI with unknown ultra-wide spectral bandwidth. It is expected that this method can provide a new approach for broadband imaging applications.