Underwater Image Enhancement-Enabled Semantic Transmission System with Super-Resolution

To address persistent challenges of color distortion and optical scattering noise in underwater image transmission, this paper proposes a semantic transmission system with underwater image enhancement (SUITS), integrating Deep Joint Source-Channel Coding (DeepJSCC) with unsupervised color correction and super-resolution (SR). The key innovation lies in embedding a self-supervised color transfer module into the DeepJSCC framework, enabling real-time chromatic correction and degradation suppression without paired training data or reference images. By incorporating channel state feedback, SUITS dynamically recalibrates feature representations to adapt to slow-fading channel conditions, while end-to-end training ensures robustness against environmental uncertainties. Experimental results on benchmark datasets show that SUITS significantly outperforms basic DeepJSCC approaches. Specifically, it achieves approximately $10 \%$ and $5 \%$ improvements in UCIQE and UIQM, respectively. The results indicate that SUITS achieves superior performance under low signal-to-noise ratio (SNR) conditions, thereby providing a robust solution for reliable image transmission in complex underwater environments.