Fig. 1. Convolution operation

Fig. 2. Maximum pooling and average pooling operations

Fig. 3. Structure diagram of residual module

Fig. 4. Structure diagram of multi-scale residual attention network

Fig. 5. Multi-scale dense feature extraction module

Fig. 6. UNet3+-Avg module

Fig. 7. Construction process of layer-3 decoder feature map $X_{De}^3$

Fig. 8. Attention residual recovery module

Fig. 9. CBAM module

Fig. 10. Underwater image example

Fig. 11. Comparison of results of different enhancement methods

Fig. 12. Comparison of results of ablation experiments

Fig. 13. Edge detection results