Fig. 1. Schematic of optical system image degradation and its recovery process

Fig. 2. Grid network architecture

Fig. 3. Schematic of grid network chunking and discarding

Fig. 4. Schematic of normal convolution and dilated convolution

Fig. 5. Diagram of spatial attention model

Fig. 6. Architecture of DAT-GridNet

Fig. 7. Diagram of the dotted block in Fig.6

Fig. 8. Schematic of the detail of the spatial attention mechanism block

Fig. 9. Example images. (a) Degraded image of atmospheric turbulence; (b) ideal image

Fig. 10. Flowchart of B-spline interpolation alignment

Fig. 11. Experimental results of atmospheric turbulence degradation image restoration. (a) Input images; (b) TSR-WGAN; (c) MPRNet; (d) DeblurGANv2; (e) GridDehazeNet; (f) proposed algorithm; (g) ground-truth

Fig. 12. Comparison experimental results of atmospheric turbulence degradation image restoration at real scene. (a) Input image; (b) DeblurGANv2; (c) GridDehazeNet; (d) proposed algorithm

Fig. 13. Comparison of running time of different methods

Fig. 14. Experimental results using different modules. (a) Input image; (b) baseline; (c) baseline+spatial attention block; (d) baseline+dilated convolutions; (e) baseline+dilated convolutions+spatial attention block; (f) ground-truth