Single-frame structured illumination microscopy based on ensemble learning

Structured Illumination Microscopy (SIM) has become one of the most important fluorescence super-resolution technologies in living cell research due to its advantages of full-field imaging and low light damage. However, SIM is a multiframe imaging technique that requires complex parameter estimation, which severely limits imaging quality and speed. To address these issues, this paper proposes an ensemble learning based single-frame composite structured illumination microscopy (E-SIM). This method obtains the full-field modulation information through only one frame composite structured illumination and combines the advantages of Convolutional Neural Networks (CNNs) and Transformer to reconstruct high-quality super-resolution images from a single-frame input image, which significantly reduces phototoxicity and photobleaching. Experimental results show that E-SIM can successfully reconstruct high-resolution images in various biological samples.