Coral Reef Benthos Classification Using Data from a Short-Range Multispectral Sensor

A five-band short-range multispectral sensor (MicaSense RedEdge-M) was adapted to an underwater housing and used obtain data from coral reef benthos. Artificial illumination required most of the spectral range sensor; optimal distance for obtaining these 0.5 m, bottom. Multispectral orthomosaics were developed using structure-from-motion software; have advantage producing ultra-high spatial resolution (down 0.4 × mm/pixel) images over larger areas. Pixel-based supervised classification a (R, G, B, RE bands) orthomosaic accurately discriminated among different benthic components; schemes defined 9 14 components such as brown algae, green sponges, crustose coralline species with high accuracy (up 84% overall accuracy, 0.83 Kappa Tau coefficients). The least useful band acquired by camera this application near-infrared (820–860 nm) associated its rapid absorption in water column. Further testing is explore possible applications orthomosaics, including assessment health colonies, well automation their processing.