Fig. 1. Network structure

Fig. 2. Images of test board 1 under different illuminations. (a) Target 1 in the clear water; (b) 21.614l lx; (c) 13.826 lx; (d) 6.947 lx; (e) 0.925 lx

Fig. 3. Comparison of before and after preprocessing in fresh water. (a) Before degradation; (b) after degradation; (c) preprocessing

Fig. 4. Relation between initial learning rates and loss function values

Fig. 5. Influence of the skip connection on the restoration effect. (a) Without one-dimensional convolution; (b) with one-dimensional convolution

Fig. 6. Effect of the one-dimensional convolution on the loss function values

Fig. 7. Effect of the sub-pixel convolution on the enhancement of underwater photoelectric image

Fig. 8. Effect of the sub-pixel convolution on the loss function values

Fig. 9. Convergence curve of the proposed network structure

Fig. 10. Results of different scenes. (a) Scene 1, effect of the test target board 1(13.826 lx); (b) Scene 2, effect of the test target board 2(13.826 lx); (c) Scene 3, effect of the test target board 3 (13.826 lx); (d) Scene 4, real underwater photoelectronic test results; (e) Scene 5, effect of the target board 1 (6.947 lx)

Table 1. Structure of the dataset

Table 2. Average time for the different methods

Table 3. [in Chinese]

Table 4. RMSC values of different methods