Semantic Segmentation of High-resolution Side-scan Sonar Imagery to Assess Changes in Restored Oyster Reef Habitats

Restored oyster reefs contribute to enhancing ecosystem services in coastal environments. Spatial and temporal changes in habitat availability can affect reef persistence and restoration benefits, so understanding these changes is crucial for optimizing restoration efforts. A set of six subtidal restored oyster reefs in Pamlico Sound, NC were studied using an unmanned surface vehicle equipped with a high-resolution bathymetric side-scanning sonar (550kHz/1600kHz). To examine how construction and location of a restored habitat affects its persistence over time, mapping was conducted in August 2016, immediately after restoration was completed, and again 21 months later in May 2018. These data were processed to create a 5-cm resolution bathymetric grid and a 0.5-cm resolution sonar mosaic. A U-Net fully convolution network was used to classify reef pixels using a combination of image-based and image-texture features calculated from the side-scan gray level co-occurrence matrices. Habitat characteristics such as area, perimeter, reef relief, rugosity, and patchiness were calculated from the segmented images and compared between the two mapping events. Significant habitat loss occurred via sedimentation at all reefs but varied spatially, which emphasizes the importance of considering location and landscape context when designing underwater restoration projects.