摘要：We developed a novel integrated technology for diver-operated surveying of shallow marine ecosystems. The HyperDiver system captures rich multifaceted data in each transect: hyperspectral and color imagery, topographic profiles, incident irradiance and water chemistry at a rate of 15-30 m(2) per minute. From surveys in a coral reef following standard diver protocols, we show how the rich optical detail can be leveraged to generate photopigment abundance and benthic composition maps. We applied machine learning techniques, with a minor annotation effort (<2% of pixels), to automatically generate cm-scale benthic habitat maps of high taxonomic resolution and accuracy (93-97%). The ability to efficiently map benthic composition, photopigment densities and rugosity at reef scales is a compelling contribution to modernize reef monitoring. Seafloor-level hyperspectral images can be used for automated mapping, avoiding operator bias in the analysis and deliver the degree of detail necessary for standardized environmental monitoring. The technique can deliver fast, objective and economic reef survey results, making it a valuable tool for coastal managers and reef ecologists. Underwater hyperspectral surveying shares the vantage point of the high spatial and taxonomic resolution restricted to field surveys, with analytical techniques of remote sensing and provides targeted validation for aerial monitoring.
