Underwater acoustic direction recognition based on asymmetric metasurfaces and lightweight deep learning

We propose a single-hydrophone direction recognition scheme, which integrates asymmetric metasurface physical coding and lightweight deep learning algorithmic decoding to address the limitations of complex structures and slow responses in conventional underwater acoustic direction recognition technologies. By breaking structural symmetry and mapping acoustic signals from different incident angles to acoustic fingerprints, physical encoding of the incident direction is achieved. Accordingly, two modes of underwater acoustic direction recognition are defined: mechanical scanning and “physical encoding + intelligent decoding” (P+I). The P+I mode collects the hydrophone’s acoustic response at only two fixed positions (initial and 180° rotated about its axis), forming a 2D feature vector. A lightweight multilayer perceptron (MLP) establishes an end-to-end mapping between features and incident angles, enabling direction recognition. This scheme holds broad promise for applications in marine engineering such as underwater detection, navigation, and communication.