Research on high-resolution reconstruction methods for microlens array snapshot hyperspectral microscopy system combining transfer learning and data decomposition-reconstruction.

The microlens array snapshot hyperspectral microscopy system has significant potential in dynamic microscopic monitoring due to its ability to rapidly acquire three-dimensional data cubes without scanning. However, its low spatial resolution and the incompatibility of existing image fusion methods limit its broader application. To address these challenges, this study proposes an image fusion method that combines transfer learning with data decomposition and reconstruction. The method reduces spectral dimensionality, enhances sample diversity through data decomposition, and reconstructs predictions to align with the characteristics of snapshot systems, achieving high spatial resolution hyperspectral image reconstruction. Experimental results demonstrate that the proposed method significantly outperforms direct training in terms of metrics such as PSNR and SAM, highlighting its strong potential for high-resolution reconstruction in snapshot systems and offering wha we believe to be a new perspective for their practical applications.