Scholarlink

<p>Underwater images often suffer from color distortion, contrast reduction, and detail loss due to wavelength-dependent light absorption and scattering in aquatic environments. Here we propose EUENet, an edge-aware underwater image enhancement network built upon a UNet baseline, with targeted architectural modifications to improve visual fidelity under such challenging conditions. EUENet incorporates three specialized modules to strengthen feature representation and spatial detail recovery: a Sobel Guided Fusion Block (SGFB) for edge-aware encoding, a Depthwise Separable Large Kernel Module (DSLKM) for multi-scale contextual modeling with attentions, and a Multi-scale Edge Information Selection Module (MEISM) for detail-aware decoding. A hybrid loss function, Perceptual—structural Enhancement Loss (PELoss), is further introduced to jointly optimize pixel-level accuracy, perceptual similarity, and semantic consistency. Experimental results on the UIEB and UFO-120 data sets demonstrate the superiority of EUENet over several state-of-the-art methods. Ablation studies conducted on the UIEB data set further confirm the effectiveness of each proposed component, showing consistent improvements in both visual quality and quantitative metrics. EUENet also achieves an inference speed of 63.57 FPS, satisfying real-time processing requirements. These findings highlight the effectiveness of EUENet in enhancing underwater imagery for real-world applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Edge-aware underwater image enhancement network based on improved UNet

  • 期刊论文

  • 发表日期：

    2025年10月27日

查看全文
  • Conggong Lin,
  • Yushi Zhang,
  • Guodong Chen

摘要

Underwater images often suffer from color distortion, contrast reduction, and detail loss due to wavelength-dependent light absorption and scattering in aquatic environments. Here we propose EUENet, an edge-aware underwater image enhancement network built upon a UNet baseline, with targeted architectural modifications to improve visual fidelity under such challenging conditions. EUENet incorporates three specialized modules to strengthen feature representation and spatial detail recovery: a Sobel Guided Fusion Block (SGFB) for edge-aware encoding, a Depthwise Separable Large Kernel Module (DSLKM) for multi-scale contextual modeling with attentions, and a Multi-scale Edge Information Selection Module (MEISM) for detail-aware decoding. A hybrid loss function, Perceptual—structural Enhancement Loss (PELoss), is further introduced to jointly optimize pixel-level accuracy, perceptual similarity, and semantic consistency. Experimental results on the UIEB and UFO-120 data sets demonstrate the superiority of EUENet over several state-of-the-art methods. Ablation studies conducted on the UIEB data set further confirm the effectiveness of each proposed component, showing consistent improvements in both visual quality and quantitative metrics. EUENet also achieves an inference speed of 63.57 FPS, satisfying real-time processing requirements. These findings highlight the effectiveness of EUENet in enhancing underwater imagery for real-world applications.