Please join me in congratulating GRG collaborators Ratneel Deo, Kate Whitton, Tristan Salles, and Jody Webster on publishing their latest article in the journal (Nature) Scientific Data.

Deo, R., John, C. M., Zhang, C., Whitton, K., Salles, T., Webster, J. M., & Chandra, R. (2024). Deepdive: Leveraging Pre-trained Deep Learning for Deep-Sea ROV Biota Identification in the Great Barrier Reef. Scientific Data, 11(1), 1-12. https://doi.org/10.1038/s41597-024-03766-3

This paper is Open Access so feel free to download it directly here.

This paper focuses on developing a quantity-controlled dataset for deep-sea biota classification using images collected by ROV SuBastian on RV Falkor Cruises. The dataset includes 3,994 images categorized into 33 classes, meticulously labelled through a human-in-the-loop process to ensure high-quality annotations.

The study benchmarks the performance of several deep learning models—ResNet, DenseNet, Inception, and Inception-ResNet—on this dataset. Despite the challenges posed by class imbalance, the Inception-ResNet model achieved AUC scores over 0.8 for each class. The paper contributes to marine conservation efforts by providing a valuable resource for automated deep-sea biota classification, which can enhance biodiversity assessments and ecosystem health evaluations.

#MarineScienceSydneyUni