Acoustic Channel-aware Autoencoder-based Compression for Underwater Image Transmission

Image transmission in Underwater Internet of Things (UW IoT) is a challenging problem due to the characteristic low bandwidth and variable path loss of the underwater acoustic channel. However, to enable intelligent and collaborative exploration of the underwater environment, such a communication is of paramount importance. To address such challenges, a reliable and energy-efficient Machine Learning (ML)-based underwater image transmission system is proposed where images are compressed using a data-based approach and robust compression codes are learned. The system uses an Autoencoder (AE) to enable intelligent, data-driven selection of coding parameters. The AE is evaluated in the presence of underwater acoustic fading channel information to achieve efficient and robust image transmission, and is compared against model-based approaches.