Hierarchical Classification for Automated Image Annotation of Coral Reef Benthic Structures

TL;DR

The paper tackles automated benthic image annotation for coral reef monitoring, addressing the mismatch between detailed flat taxonomic labels and ecological analyses. It proposes a hierarchical classification (HC) framework with a top-down Local Classifier per Parent Node, leveraging a taxonomy-aligned structure and a CoralNet EfficientNet B0 backbone for feature extraction. Empirical results on a Northeast Brazilian dataset show HC yields consistent, though modest, improvements of about 1–2% in both standard F1 and hierarchical F1 over flat baselines, underscoring ecological relevance. The work highlights the potential of HC to better align automated annotations with ecological workflows and outlines future directions including broader datasets, automated tree design, and examination of computational trade-offs.

Abstract

Automated benthic image annotation is crucial to efficiently monitor and protect coral reefs against climate change. Current machine learning approaches fail to capture the hierarchical nature of benthic organisms covering reef substrata, i.e., coral taxonomic levels and health condition. To address this limitation, we propose to annotate benthic images using hierarchical classification. Experiments on a custom dataset from a Northeast Brazilian coral reef show that our approach outperforms flat classifiers, improving both F1 and hierarchical F1 scores by approximately 2\% across varying amounts of training data. In addition, this hierarchical method aligns more closely with ecological objectives.

Figures (4)

• Figure 1: Benthic image annotation process -- A predetermined number of random points are sampled within a fixed area and annotated with the corresponding organism or substrate. The relative frequencies are used as proxies for cover proportions of benthic categories within the sampled area.
• Figure 2: Hierarchical structure of benthic labels -- This conceptual diagram illustrates the multi-level categories used for classifying benthic organisms and substrates in coral reef ecosystems.
• Figure 3: Experimental results -- Comparison of flat and hierarchical classifiers for our custom benthic image annotation dataset, showing performance metrics across varying training data sizes. Error bars indicate standard deviation from random training set sampling.
• Figure 4: Number of patches corresponding to each leaf node present in our custom benthic image annotation dataset.