Fig. 1. Schematic diagram of deep learning light field microscopy aberration correction principle

Fig. 2. PCANet structure. (a) Improved two-path network structure; (b) upsampling convolution module; (c) convolutional block; (d) residual module; (e) attention module

Fig. 3. PCANet decoupling submodule

Fig. 4. Diagram of experimental setup. (a) Device diagram of the light field microscopy acquisition system; (b) schematic diagram of the light field microscopy setup; (c) microlens array calibrationc (local magnification); (d) egg embryo section and onion epidermal cut specimen

Fig. 5. Comparison of resolution plates imaging results. (a) Original aberration result; (b) light field DAO reconstruction result; (c) reconstruction results of the proposed method; (d)(f)(h) enlarged image of central region I; (e)(g)(i) estimated aberration wavefront corresponding to central region I; (j)(l)(n) enlarged image of edge region Ⅱ; (k)(m)(o) estimated aberration wavefront corresponding to edge region Ⅱ; (p) strength curves of marked part in central region I; (q) strength curve of marked part in edge region Ⅱ

Fig. 6. Imaging results of 36 h egg embryo slices. (a) Raw aberration blur intensity input after optical field rendering; (b) single-channel U-Net reconstruction without phase information; (c) method in proposed; (d)(e)(f) locally enlarged images; (g)(h)(i) grayscale histograms corresponding to locally enlarged image; (j)(k)(l) edge contour extractions of three results

Fig. 7. Reconstruction results of different methods. (a) Apical margin region of onion root tip longitudinal section; (b) full field of view region of 72 h egg embryo section; (c) center region of onion root tip longitudinal section