An interactive, machine-assisted workflows for analyzing coral morphology, combining computer vision and expert oversight to improve efficiency and reproducibility in coral research.

Coral reefs are among the most diverse and ecologically significant marine ecosystems, yet assessing their structural dynamics remains a labor-intensive and time-consuming task. Conventional approaches rely on manual interpretation of photographic and CT-scan data, limiting both scale and temporal resolution in monitoring efforts. To address these constraints, this project develops cloud-based workflows that integrate advanced computer vision techniques with expert input. The human-in-the-loop design ensures that automation supports, rather than substitutes, scientific judgment—allowing researchers to refine model outputs, correct segmentation errors, and iteratively improve system performance on specialized datasets.

At the core of the initiative lies a modular analysis platform capable of processing both two- and three-dimensional coral imagery. Pre-trained object detection and segmentation models provide an initial framework for identifying morphological features, while interactive annotation tools enable targeted refinement and adaptive learning. The platform supports automated post-processing and quantitative trait extraction, producing standardized and reproducible measurements of coral morphology relevant to ecological and conservation research.

Beyond its immediate application to coral ecology, the project illustrates how interactive machine learning can augment domain expertise across environmental sciences. By coupling algorithmic scalability with human oversight, it contributes to the development of AI-driven research infrastructures that advance both efficiency and interpretability in data-rich, expert-dependent fields.

Contact

Robert Leist, robert.leist@dfki.de