<p>Underwater image enhancement is a challenging problem due to severe visual degradations caused by wavelength-dependent absorption, scattering, and non-uniform illumination. This paper proposes WaterNet (WNet) based convolutional block attention module (CBAM) using a patch generative adversarial network (PGAN). In this paper, an attention-guided adversarial framework has been proposed that embeds spatial and channel attention modules into a WaterNet-inspired generator with hierarchical skip connections to strengthen feature propagation. A PatchGAN discriminator enforces local structural fidelity, while a hybrid loss function combining adversarial and pixel-level reconstruction constraints ensures a trade-off between perceptual quality and content preservation. Extensive experiments across multiple benchmark datasets demonstrate that the proposed approach consistently outperforms conventional model-based and state-of-the-art deep learning methods in terms of visual realism, perceptual consistency, and structural accuracy. Ablation studies confirm the complementary roles of spatial and channel attention, and cross-dataset evaluations reveal strong generalization to diverse underwater environments without dataset-specific fine-tuning. These results establish WNetCBAM-PGAN as a robust and scalable solution for real-world applications in underwater exploration, autonomous robotics, and marine scientific imaging.</p>