<scp>RapidBenthos</scp> : Automated segmentation and multi‐view classification of coral reef communities from photogrammetric reconstruction

Abstract Underwater photogrammetry is routinely used to monitor large areas of complex and heterogeneous ecosystems, such as coral reefs. However, deriving data on benthic components (i.e. sand, rubble, coral and algae) from photogrammetry products has remained challenging due to the highly time‐consuming process of manual data extraction. We developed a machine learning approach to quantify benthic community composition in coral reefs from orthomosaics, which requires no manual delineation of benthic components for training or implementation. The current study presents RapidBenthos, an automated workflow that segments and classifies large‐area images. Our pipeline (1) uses a pre‐trained segmentation model, eliminating the need for manually generated fine‐scale segmented training data, and (2) classifies the resulting segments from multiple views using the underlying survey images, allowing for classification to fine taxonomic levels. Within a test photomosaic built from a coral reef area of 40 m −2 , the model automatically detected 43 different benthic classes. Validation resulted in an overall classification accuracy of 0.96 and a segmentation accuracy of 0.87, when compared to a manually digitised replica. The RapidBenthos workflow was 195 times faster than manual segmentation and classification. Additional validation of 524 Acropora coral colonies from 11 additional test plots resulted in a segmentation accuracy of 0.92 and classification accuracy of 0.88 to the coarser ‘ Acropora’ group. RapidBenthos has the capability to extract an unprecedented level of data from photomosaics of coral reefs or other complex environments, allowing to sustainably scale photogrammetric monitoring technique both in replicate and survey extent, which consequently can lead to new research questions and more informed ecosystem management.