• Optics and Precision Engineering
  • Vol. 32, Issue 10, 1582 (2024)
Xiaohua XIA1,*, Yuquan ZHONG1, Peng HU1, Yunshi YAO1,2..., Jiguang GENG2 and Liangqi ZHANG2|Show fewer author(s)
Author Affiliations
  • 1Key Laboratory of Road Construction Technology and Equipment, Ministry of Education, Chang'an University, Xi'an70064,China
  • 2Henan Wanli Transportation Technology Group Co. Ltd., Xuchang461000, China
  • show less
    DOI: 10.37188/OPE.20243210.1582 Cite this Article
    Xiaohua XIA, Yuquan ZHONG, Peng HU, Yunshi YAO, Jiguang GENG, Liangqi ZHANG. Underwater image enhancement synthesizing multi-scale information and attention mechanisms[J]. Optics and Precision Engineering, 2024, 32(10): 1582 Copy Citation Text show less
    Generator network structure
    Fig. 1. Generator network structure
    Multi-scale hybrid convolution structure
    Fig. 2. Multi-scale hybrid convolution structure
    Attention mechanism structure
    Fig. 3. Attention mechanism structure
    Discriminator network structure
    Fig. 4. Discriminator network structure
    Loss function variation chart with iterations
    Fig. 5. Loss function variation chart with iterations
    Comparison of the processing results of the seven methods
    Fig. 6. Comparison of the processing results of the seven methods
    PSNR of the result obtained by the model proposed in this paper and comparison methods
    Fig. 7. PSNR of the result obtained by the model proposed in this paper and comparison methods
    SSIM of the result obtained by the model proposed in this paper and comparison methods
    Fig. 8. SSIM of the result obtained by the model proposed in this paper and comparison methods
    UIQM of the result obtained by the model proposed in this paper and comparison methods
    Fig. 9. UIQM of the result obtained by the model proposed in this paper and comparison methods
    NIQE of the result obtained by the model proposed in this paper and comparison methods
    Fig. 10. NIQE of the result obtained by the model proposed in this paper and comparison methods
    Comparison of standard color card tests
    Fig. 11. Comparison of standard color card tests
    MethodsPSNRSSIMUIQMNIQE
    UDCP14.4960.5591.8434.748
    RGHS18.0160.7462.3504.574
    IBLA18.8700.7002.1584.728
    UGAN21.0690.7862.8864.496
    FUnIE-GAN22.1460.7872.8194.482
    UIE-WD21.6050.7872.8454.615
    Ours22.4990.7892.9114.175
    Table 1. Average score of each method on the entire test set
    MethodsUDCPRGHSIBLAUGANFUnIE-GANUIE-WDOurs
    Time/s2.6520.7295.6780.0140.0130.0100.014
    Table 2. Comparison results of operating time
    Attention mechanismHybrid convolutionGlobal encoderPSNRSSIMUIQMNIQE
    22.4990.7892.9114.175
    ×20.7580.6962.8035.026
    ×21.6380.7432.8044.061
    ×21.9730.7582.7704.289
    Table 3. Comparison results of ablation experiments
    Xiaohua XIA, Yuquan ZHONG, Peng HU, Yunshi YAO, Jiguang GENG, Liangqi ZHANG. Underwater image enhancement synthesizing multi-scale information and attention mechanisms[J]. Optics and Precision Engineering, 2024, 32(10): 1582
    Download Citation