Adaptive Semantic Communication for Underwater Images Based on Residual Vector Quantization

Reliable image transmission over underwater acoustic channels is challenged by severe rate constraints, channel time variation, Doppler-induced distortion, and the high latency of conventional bit-level communication. In this paper, we propose an adaptive semantic communication framework based on Residual Vector Quantization (Res-VQ) for underwater image transmission. First, semantic features are progressively quantized through the joint adaptation of codebook size and the number of quantization layers, enabling fine-grained rate-distortion control under dynamic channel conditions. Furthermore, a semantic-region-aware differentiated quantization allocation strategy is developed to assign more quantization resources to semantically important regions, while also supporting future unequal error protection through subchannel-aware resource allocation, thereby reducing transmission overhead while preserving the reconstruction quality of key targets. Notably, the proposed framework avoids reliance on conventional channel coding, which helps reduce transmission overhead and end-to-end latency. Experimental results validate the effectiveness and robustness of the proposed method in underwater semantic communication scenarios.