Scholarlink

<p>Underwater Image Enhancement (UIE) plays a key role as a preprocessing step in improving visual quality for applications such as marine exploration, autonomous underwater vehicles, environmental monitoring, and surveillance. However, underwater images often experience problems caused by wavelength-dependent light absorption and scattering. These effects usually result in color distortion, low contrast, and poor visibility. In recent years, deep learning approaches have significantly improved the performance of UIE techniques. Despite this progress, the rapid increase in proposed models, datasets, and evaluation methods has made it difficult to perform systematic comparisons and comprehensive analyses. To address this issue, this study presents a Systematic Literature Review (SLR) of deep learning–based UIE methods. A multi-stage screening and quality assessment process was applied, where more than 2,500 initial records were examined and 196 relevant studies were selected according to predefined inclusion criteria. The selected methods are categorized into main groups: CNN-based, GAN-based, and transformer-based approaches. Their architectures, strengths, and limitations are analyzed in detail. In addition, this study reviews commonly used benchmark datasets and evaluation metrics, highlighting the differences between reference-based and no-reference quality assessment methods. Finally, several key research challenges are discussed, including dataset scarcity, model generalization, computational complexity, and real-world deployment. The paper also outlines future research directions aimed at supporting fair benchmarking and improving reproducibility in UIE research.</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

Advances in Deep Learning for Underwater Image Enhancement: A Systematic Analysis of Architectures, Benchmarks, and Open Challenges

  • 期刊论文

  • 发表日期：

    2026年7月16日

查看全文
  • Vaibhav Ingale,
  • Manish Kumar

摘要

Underwater Image Enhancement (UIE) plays a key role as a preprocessing step in improving visual quality for applications such as marine exploration, autonomous underwater vehicles, environmental monitoring, and surveillance. However, underwater images often experience problems caused by wavelength-dependent light absorption and scattering. These effects usually result in color distortion, low contrast, and poor visibility. In recent years, deep learning approaches have significantly improved the performance of UIE techniques. Despite this progress, the rapid increase in proposed models, datasets, and evaluation methods has made it difficult to perform systematic comparisons and comprehensive analyses. To address this issue, this study presents a Systematic Literature Review (SLR) of deep learning–based UIE methods. A multi-stage screening and quality assessment process was applied, where more than 2,500 initial records were examined and 196 relevant studies were selected according to predefined inclusion criteria. The selected methods are categorized into main groups: CNN-based, GAN-based, and transformer-based approaches. Their architectures, strengths, and limitations are analyzed in detail. In addition, this study reviews commonly used benchmark datasets and evaluation metrics, highlighting the differences between reference-based and no-reference quality assessment methods. Finally, several key research challenges are discussed, including dataset scarcity, model generalization, computational complexity, and real-world deployment. The paper also outlines future research directions aimed at supporting fair benchmarking and improving reproducibility in UIE research.