Scholarlink

<p>Underwater imaging has gained significant attention in recent years due to its applications in marine exploration, aquatic life observation, autonomous underwater vehicle (AUV) navigation, and underwater robotics. Light is absorbed and scattered under water, making underwater photos often difficult to see, with low contrast and considerable colour distortion.These challenges limit the performance of traditional image processing and enhancement techniques, which typically rely on low-level heuristics or physics-based models and often fail to generalize across diverse underwater environments.To overcome these limitations, we propose a novel underwater image enhancement framework that integrates deep learning with nature-inspired optimization and progressive refinement. The model combines the strength of Convolutional Neural Networks (CNNs) for capturing local features with the global modelling capability of Transformers through a hybrid architecture. This design enables the network to learn both fine spatial details and long-range dependencies essential for underwater image restoration.To improve feature selection and convergence behaviour, we incorporate the Cheetah Optimization Algorithm (COA), a bio-inspired metaheuristic algorithm that mimics the agile hunting strategy of cheetahs. COA adaptively tunes key parameters during training, enhancing the network’s ability to recover details from severely degraded underwater images.Furthermore, we introduce a Multi-Residual Refinement Plus Plus (MRR++) module that applies iterative residual learning across multiple enhancement stages. This progressive approach allows the model to incrementally reduce colour distortion, noise, and artifacts, leading to superior image clarity and perceptual quality.The proposed method is trained and evaluated on benchmark underwater dataset UIEB. In wide range of challenging scenarios, the suggested model effectively restores and enhances underwater images, as shown by extensive experimental evaluations that show it consistently outperforms state-of-the-art methods in both quantitative metrics —including PSNR, SSIM, UIQM, and UIQI and qualitative visual assessments.</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

Underwater low-light image enhancement using a hybrid CNN–transformer framework with multi-scale feature fusion and cheetah optimization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月31日

查看全文
  • Divya Midhunchakkaravarthy,
  • Katakam RangaNarayana,
  • V. Lakshman Narayana

摘要

Underwater imaging has gained significant attention in recent years due to its applications in marine exploration, aquatic life observation, autonomous underwater vehicle (AUV) navigation, and underwater robotics. Light is absorbed and scattered under water, making underwater photos often difficult to see, with low contrast and considerable colour distortion.These challenges limit the performance of traditional image processing and enhancement techniques, which typically rely on low-level heuristics or physics-based models and often fail to generalize across diverse underwater environments.To overcome these limitations, we propose a novel underwater image enhancement framework that integrates deep learning with nature-inspired optimization and progressive refinement. The model combines the strength of Convolutional Neural Networks (CNNs) for capturing local features with the global modelling capability of Transformers through a hybrid architecture. This design enables the network to learn both fine spatial details and long-range dependencies essential for underwater image restoration.To improve feature selection and convergence behaviour, we incorporate the Cheetah Optimization Algorithm (COA), a bio-inspired metaheuristic algorithm that mimics the agile hunting strategy of cheetahs. COA adaptively tunes key parameters during training, enhancing the network’s ability to recover details from severely degraded underwater images.Furthermore, we introduce a Multi-Residual Refinement Plus Plus (MRR++) module that applies iterative residual learning across multiple enhancement stages. This progressive approach allows the model to incrementally reduce colour distortion, noise, and artifacts, leading to superior image clarity and perceptual quality.The proposed method is trained and evaluated on benchmark underwater dataset UIEB. In wide range of challenging scenarios, the suggested model effectively restores and enhances underwater images, as shown by extensive experimental evaluations that show it consistently outperforms state-of-the-art methods in both quantitative metrics —including PSNR, SSIM, UIQM, and UIQI and qualitative visual assessments.