Super-Resolution Reconstruction Algorithm of Underwater Image Based on Information Distillation Mechanism

Hongchun Yuan; Lingdong Kong; Shanshan Zhang; Kai Gao; Yurui Yang

doi:10.3788/LOP221324

Laser & Optoelectronics Progress, Volume. 60, Issue 12, 1210017(2023)

Super-Resolution Reconstruction Algorithm of Underwater Image Based on Information Distillation Mechanism

Hongchun Yuan, Lingdong Kong^*, Shanshan Zhang, Kai Gao, and Yurui Yang

Author Affiliations

School of Information, Shanghai Ocean University, Shanghai 201306, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(16)

Fig. 1. Schematic diagram of residual network structure based on information distillation mechanism

Download full size

Fig. 2. Residual feature distillation block

Download full size

Fig. 3. Spatial attention module

Download full size

Fig. 4. Training loss

Download full size

Fig. 5. Results of ablation experiments on GFF and SA (validation set)

Download full size

Fig. 6. Visualization of SA

Download full size

Fig. 7. Distilled feature maps from different stages (part)

Download full size

Fig. 8. Comparison of visual results (R_PSNR and M_SSIM) of im_xb_660_ from USR-248 dataset among different algorithms (×2)

Download full size

Fig. 9. Comparison of visual results (R_PSNR and M_SSIM) of im_xb_5341_ from USR-248 dataset among different algorithms (×4)

Download full size

Fig. 10. Comparison of visual results (R_PSNR and M_SSIM) of im_xb_331_ from USR-248 dataset among different algorithms (×8)

Download full size

Fig. 11. SR result generated with SRIDM in real underwater scenes

Download full size

Fig. 12. Comparison of parameters, floating-operations, and running time of different algorithms

Download full size

Table 1. Results of ablation experiments on GFF and SA (test set)
View table
Table 1. Results of ablation experiments on GFF and SA (test set)
Scale Base GFF SA R_PSNR /dB M_SSIM
×4 √ × × 27.5629 0.7142
√ √ × 27.5660 0.7143
√ × √ 27.5690 0.7146
√ √ √ 27.5851 0.7150

Table 2. [in Chinese]
View table
Table 2. [in Chinese]
$d$ R_PSNR /dB M_SSIM
0.25 27.7640 0.7640
0.5 27.7566 0.7635
0.75 27.7480 0.7631

Table 3. Comparison of average R_PSNR and M_SSIM by different super-resolution algorithms on test datasets

View table

Table 3. Comparison of average R_PSNR and M_SSIM by different super-resolution algorithms on test datasets

Algorithm	Scale	Params /10³	FLOPs /10⁹	R_PSNR /dB	M_SSIM
Bicubic	×2			32.5489	0.9083
SRCNN	×2	8	3.1	32.7391	0.9097
DSRCNN	×2	297	2896.0	33.5831	0.9186
EDSR	×2	1074	92.0	33.6245	0.9189
RCAN	×2	1192	101.6	33.6207	0.9191
CARN	×2	964	92.7	33.6721	0.9193
IMDN	×2	694	62.2	33.6810	0.9194
SRIDM	×2	668	60.1	33.6983	0.9198
Bicubic	×4			26.7702	0.7432
SRCNN	×4	8	3.1	27.1537	0.7402
DSRCNN	×4	297	2896.0	27.6512	0.7572
EDSR	×4	1222	36.0	27.7067	0.7628
RCAN	×4	1340	38.4	27.7003	0.7626
CARN	×4	1112	36.2	27.7426	0.7634
IMDN	×4	715	16.0	27.7336	0.7637
SRIDM	×4	689	15.5	27.7640	0.7640
Bicubic	×8			23.5757	0.6080
SRCNN	×8	8	3.1	24.1015	0.6043
DSRCNN	×8	297	2896.0	24.4117	0.6211
EDSR	×8	1370	22.1	24.4836	0.6304
RCAN	×8	1487	22.7	24.4814	0.6295
CARN	×8	1260	22.1	24.5004	0.6310
IMDN	×8	798	4.4	24.4907	0.6303
SRIDM	×8	772	4.3	24.5018	0.6304

Table 4. Number of params, FLOPs, run-time, and memory requirement of SRIDM on RTX 3090
View table
Table 4. Number of params, FLOPs, run-time, and memory requirement of SRIDM on RTX 3090
Model ×2 ×4 ×8
Params /10³ 668 689 772
FLOPs /10⁹ 60.15 15.48 4.31
Run-time /ms 3.85 3.73 3.59
Frames per second 259.74 268.10 278.55
Model_size /MB 2.6 2.7 3.0

Tools

Get Citation

Copy Citation Text

Hongchun Yuan, Lingdong Kong, Shanshan Zhang, Kai Gao, Yurui Yang. Super-Resolution Reconstruction Algorithm of Underwater Image Based on Information Distillation Mechanism[J]. Laser & Optoelectronics Progress, 2023, 60(12): 1210017

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Apr. 18, 2022

Accepted: Jun. 28, 2022

Published Online: Jun. 5, 2023

The Author Email: Lingdong Kong (2067718653@qq.com)

DOI:10.3788/LOP221324

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Results of ablation experiments on GFF and SA (test set)

Table 1. Results of ablation experiments on GFF and SA (test set)

Table 2. [in Chinese]

Table 2. [in Chinese]

Table 3. Comparison of average RPSNR and MSSIM by different super-resolution algorithms on test datasets

Table 3. Comparison of average RPSNR and MSSIM by different super-resolution algorithms on test datasets

Table 4. Number of params, FLOPs, run-time, and memory requirement of SRIDM on RTX 3090

Table 4. Number of params, FLOPs, run-time, and memory requirement of SRIDM on RTX 3090

Table 3. Comparison of average R_PSNR and M_SSIM by different super-resolution algorithms on test datasets

Table 3. Comparison of average R_PSNR and M_SSIM by different super-resolution algorithms on test datasets