Deep Joint Source-Channel Coding of Underwater Image Enhancement in AUV Swarms

Underwater images commonly suffer from visual degradation due to the complexities of the underwater imaging and transmission environments. To address this issue, we propose a lightweight system for underwater image enhancement and transmission tailored for autonomous underwater vehicle swarms, improving both image quality and communication efficiency with surface equipment. The proposed model enables a target-oriented multi-space feature encoding network to process degraded pixels, followed by a reinforcement learning-based bandwidth allocation network that optimizes feature distribution under constrained bandwidth conditions. Simulation results demonstrate that the proposed system significantly enhances the speed and effectiveness of image recovery under varying channel conditions.