The paper introduces AQUA-Net, a deep-learning architecture for underwater image enhancement. The abstract explains that underwater images are commonly degraded by wavelength-dependent light absorption and scattering, which leads to severe color distortion, low contrast, and a hazy appearance.
AQUA-Net stands for Adaptive Frequency Fusion and Illumination Aware Network. The name points to two core mechanisms: adaptive fusion of frequency information and processing that accounts for illumination conditions. The abstract does not provide further architectural details.
The abstract also starts to position AQUA-Net against existing deep learning models, but the excerpt cuts off mid-sentence. As a result, the specific advantages over prior work are not available in the source.