Visual Microfossil Identification via Deep Metric Learning

We apply deep metric learning for the first time to problem of classifying planktic foraminifer shells on microscopic images. This species recognition task is an important information source and scientific pillar reconstructing past climates. All CNN pipelines in literature produce black-box classifiers that lack visualisation options human experts cannot be applied open set problems. Here, we benchmark against these pipelines, phenotypic morphology space, demonstrate can used cluster unseen during training. show outperforms all published CNN-based state-of-the-art benchmarks this domain. evaluate our approach 34,640 expert-annotated images Endless Forams public library 35 modern foramini-fera species. Our results data leading $$92\%$$ accuracy (at 0.84 F1-score) reproducing expert labels withheld test data, $$66.5\%$$ 0.70 when clustering never encountered conclude highly effective domain serves as tool towards expert-in-the-loop automation microfossil identification. Key code, network weights, splits are with paper full reproducibility.