Underwater Image Enhancement: Comparison Between UDnet and GAN Models

As the largest ecosystem on Earth, the ocean holds abundant biological resources, mineral deposits, energy sources, and a wealth of scientific mysteries. With intensifying international competition in marine exploration and ongoing scientific and technological advances, obtaining clear, high-quality underwater images has become critical for deep-sea topographical surveys and seafloor resource exploration. However, due to the absorption and scattering of light in water, as well as the impact of suspended particles, underwater images often suffer from blurriness, low contrast, and color distortion. To address these issues, this paper proposes an innovative unsupervised image enhancement framework, employing UDnet and GAN models under the same training configurations—500 epochs, a batch size of 10, and a learning rate of 1×10<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">–4</sup>. By comparing the loss function iteration curves and the visual quality of enhanced images, we observe that UDnet demonstrates superior performance in correcting color shifts, increasing brightness and contrast, and reducing image blurriness, with faster convergence and more stable generalization capabilities than GAN. Consequently, UDnet is selected as the primary enhancement method, and its output undergoes objective evaluation via PSNR, UCIQE, and UIQM metrics. This research not only offers a robust, efficient solution for underwater image enhancement in practical applications—such as deep-sea expeditions and seafloor resource assessments—but also provides insights for other complex scenarios requiring advanced image processing and enhancement. Looking ahead, further integration of underwater physical models with deep learning theory is expected to drive notable improvements in ocean exploration and sustainable development.