DCA-Net: underwater image enhancement network based on dual attention and dynamic convolution

Underwater Image Enhancement (UIE) aims to address the effects of absorption and scattering of light in underwater environments, which often lead to degraded image quality. The challenges of UIE development stem from common issues in underwater scenes, such as low contrast, high turbidity, visual blurriness, and color distortion. Fortunately, advancements in deep learning have introduced innovative solutions to this field. Despite the significant potential of deep learning techniques in UIE, current approaches still face limitations in robustness and adaptability, making it difficult to effectively handle the dynamic nature of underwater environments. To address these limitations, we introduce an innovative UIE network, named DCA-Net, which integrates dual attention mechanisms and convolutional operations to achieve enhanced adaptability and robustness. DCA-Net leverages enhanced channel attention while incorporating comprehensive and localized strip attention, allowing it to better capture contextual information during image restoration. The empirical findings across various benchmark datasets reveal that DCA-Net consistently surpasses current approaches, delivering enhanced performance in terms of both perceptual quality and quantitative assessment measures.