计算机技术与发展2025，Vol.35Issue(11)：28-37,10.DOI:10.20165/j.cnki.ISSN1673-629X.2025.0171
基于动态卷积与注意力机制的水下图像增强
Underwater Image Enhancement Based on Dynamic Convolution and Attention Mechanism
摘要
Abstract
Underwater images often suffer from blurring,low contrast,and color distortion due to scattering and absorption,which seriously affect the visual quality of underwater images.To address these issues,we propose D-SCAT(Dynamic Spatial-Channel Attention Transformer),an underwater image enhancement model based on Swin Transformer.Firstly,the network adopts an encoder-bottleneck-decoder structure,integrating spatial and channel attention for joint processing of spatial structure and color features.Secondly,the bottleneck layer captures global features and long-range dependencies using Swin Transformer's local window self-attention.Finally,shallow features are fused at the decoder to enhance details and textures,generating natural and realistic underwater images.Experimental results on UIEB and EUVP datasets show that D-SCAT outperforms classical and state-of-the-art methods.On the UIEB test set,it achieves PSNR of 24.105 1 dB,SSIM of 0.914 0,and MSE of 0.006 1,surpassing comparison methods by 0.649 6 dB,0.017 6,and0.000 9,respectively.The model effectively corrects color distortion,restores details,and enhances contrast and sharpness.关键词
水下图像增强/空间-通道注意力/Swin Transformer/自适应卷积/深度学习Key words
underwater image enhancement/spatial-channel attention/Swin Transformer/adaptive convolution/deep learning分类
信息技术与安全科学引用本文复制引用
田乾坤,宋伟..基于动态卷积与注意力机制的水下图像增强[J].计算机技术与发展,2025,35(11):28-37,10.基金项目
国家重点研发计划项目(2024YFC2815004,2024YFC2815205) （2024YFC2815004,2024YFC2815205）
