Underwater images typically suffer from low contrast, color distortion, and blurred details caused by light absorption and scattering, which severely limit the performance of visual perception tasks such as marine ecosystem monitoring, shipwreck inspection, and autonomous underwater vehicle (AUV) navigation. Conventional physics-based restoration methods are highly sensitive to water types and illumination conditions, thus lacking robustness in practical scenarios. To overcome these limitations, we propose a deep learning-based underwater image enhancement framework termed SimpleEnhanceNet, which adopts an encoderdecoder convolutional neural network (CNN) with skip connections. The network is trained on the public UIEB and EUVP datasets in a supervised manner, where the optimization objective combines pixel-wise mean squared error with perceptual loss to jointly preserve structural fidelity and perceptual quality. Extensive experiments demonstrate that our method achieves superior performance over traditional approaches on the UIEB benchmark, yielding improvements of 5.82 dB in peak signal-to-noise ratio (PSNR) and 0.21 in structural similarity index (SSIM). Moreover, the no-reference metrics UIQM and UCIQE also exhibit substantial gains. Qualitative comparisons confirm that SimpleEnhanceNet effectively restores natural colors and enhances scene clarity across diverse underwater conditions. These results highlight its potential for real-time deployment in AUV navigation, environmental monitoring, and marine exploration.
Chenyu Yu (2025) studied this question.