<p>We develop an underwater acoustic semantic communication approach for underwater image transmission in this study. The developed approach has two main merits. First, it enables efficient underwater image transmission, in terms of microvolume textual semantics, via the underwater acoustic channel, which typically has limited bandwidth. Second, it enhances robust underwater image transmission by leveraging semantic communication in the underwater environment, where the signal-to-noise ratio (SNR) is typically low. To achieve these goals, semantic representations for underwater images are developed. Specifically, at the transmitter’s end, semantic extraction is conducted by converting an original underwater image into descriptive text, from which an entity list is subsequently identified. The entity list represents the textual semantics of the underwater image, providing a microvolume representation for efficient underwater image transmission. The entity list is then transmitted through the underwater acoustic channel to the receiver. At the receiver’s end, semantic reconstruction is conducted by converting the received entity list back into descriptive text, which is subsequently converted into a generated underwater image semantically consistent with the original. The semantic extraction and semantic reconstruction of the underwater image are empowered by deep learning technologies with the support of a reliable semantic knowledge base. The experimental results prove that our approach offers advantages, in terms of microvolume and robust representations, over standard approaches to underwater image transmission. Therefore, this approach establishes a new baseline for underwater image transmission, which could potentially overcome the challenges posed by the narrow bandwidth and low SNR of the underwater acoustic channel.</p>