UbceNet: An Underwater Image Enhancement Network for Processing Underwater Images

Due to the exponential attenuation of light propagation in water, underwater images often exhibit poor quality. In this paper, we propose the UbceNet model designed to enhance the quality of underwater images. It comprises two primary modules: image processing and color stretching. In the image enhancement module, significant features within the image are extracted through pooling operations to reduce noise in the feature map. Additionally, we employ Dwise convolution to enhance dimensionality, mitigating the risk of model overfitting. In the color stretching module, we introduce a Coordinate attention and a HardSwish residual enhancement module to enhance color performance by focusing on key image features. Our model outperforms the state-of-the-art models on the dataset, achieving a PSNR metric of 24.04dB and an SSIM of 0.83%. These results highlight the model's strong performance in optimizing underwater image quality. In the future, this model can serve as the foundation for practical applications such as underwater robot vision, water quality testing, underwater target detection and other fields.