Fast structured illumination microscopy via transfer learning with correcting

In microscopic imaging, structured light illumination technology has emerged as a significant super-resolution imaging technique due to its inherent advantages, including rapid imaging speed and low phototoxicity. However, the conventional super-resolution image reconstruction algorithm is still widely used today, despite its slow calculation speed, difficulty in setting parameters, and unpredictable artifact generation. With the advancement of computer technology, image reconstruction and artifact removal can be accomplished rapidly using deep learning. Nowadays, deep learning training requires a large number of data samples, but obtaining a large number of biological samples under structured light illumination is difficult. This paper proposes a new method based on transfer learning that generates a pre-training network using its own simulation data and trains a correction using a small amount of biological sample data to achieve super-resolution imaging of biological samples. Additionally, it can reconstruct sampled wide-field images from 9 to 3 frames, as well as the trained network, which can quickly reconstruct structure illumination microscopy(SIM).