Energy-efficient underwater wireless optical semantic communication based on channel-adaptive conditional computation

Existing underwater wireless optical semantic communication (UWOSC) systems lack the ability to adapt the computational load to varying channel conditions, which leads to energy inefficiency and thus limits their application on energy-constrained underwater devices, such as autonomous underwater vehicles and underwater sensor nodes. In this Letter, we introduce conditional computation into UWOSC for the first time, to the best of our knowledge, and develop an energy-efficient UWOSC system based on channel-adaptive conditional computation for image transmission. Specifically, we design a semantic codec that supports early computation termination and employ a two-stage knowledge distillation strategy, enabling the system to perform semantic transmission over multiple computational paths. A semantic outage probability-guided path selector is further designed to adaptively determine the computational path according to channel conditions. Experimental results demonstrate that the proposed UWOSC system achieves superior transmission performance over traditional UWOC systems with lower energy consumption, while maintaining competitive performance relative to existing UWOSC schemes.