Multi-scale approach for coral condition assessment and Drupella sp. identification using a versatile low-cost camera system and point cloud semantic segmentation
Jiaqi Wang, Xinlei Shao, Bangzhang Ma, Fan Zhao, Kirsty Magson, Jun Sasaki, Katsunori Mizuno
Abstract
Coral reef ecosystems are increasingly threatened by climate change and human activities, and conservation organizations worldwide require efficient, cost-effective tools to document habitat conditions, assess artificial reef growth, and monitor corallivore dynamics. Traditional approaches, such as transect surveys with visual inspection, often suffer from low efficiency, limited coverage, and insufficient spatial information. To address these limitations, we developed a low-cost and versatile underwater camera system with three operating modes: surface, horizontal underwater, and curved underwater. The system records continuous video through three synchronized cameras, which are processed using Structure-from-Motion (SfM) photogrammetry to generate 3D reconstructions for habitat and colony-scale monitoring. A point cloud-based semantic segmentation model (KPConv) was then applied to segment the reconstructed point clouds into 'Drupella snails', 'Compromised coral', and 'Others', thereby indicating coral conditions affected by corallivore predation. Field surveys conducted in Koh Tao, Thailand, demonstrated the feasibility of this integrated hardware-software approach. The camera system achieved higher point densities and lower costs compared to conventional methods, while remaining lightweight and cost-effective for community-based conservation. The KPConv model yielded an overall accuracy of 0.916 and a mean Intersection over Union of 0.601, enabling the identification of compromised corals and cryptic Drupella snails. Quantitative assessments based on segmented outputs further revealed coral conditions and provided early warnings of potential Drupella snails outbreaks. In summary, this novel system integrates affordable image collection hardware with 3D reconstruction and semantic segmentation, improving survey efficiency and providing spatially explicit insights for coral reef monitoring, particularly for small-scale corallivore dynamics.