Tropical Indian ocean annotated planktonic foraminifera image dataset from surface sediments
|Author(s)||Adebayo Michael1, Bolton Clara1, Marchant Ross2, Bassinot Franck3, Conrod Sandrine1, De-Garidel Thoron Thibault1|
|Affiliation(s)||1 : Centre Européen de Recherche et d’Enseignement des Géosciences de l’Environnement (CEREGE), Aix- Marseille Université, CNRS, IRD, Coll. De France, INRAE, Aix-en-Provence, France
2 : School of Electrical Engineering & Robotics, Queensland University of Technology, Brisbane, Australia
3 : Laboratoire des Sciences du Climat et de l’Environnement (IPSL), CEA-CNRS-UVSQ, Université Paris-Saclay, Paris, France
|Note||While the images in this dataset have been reviewed by at least one human expert classifier, we acknowledge that there could be some cases of misidentification.|
|Keyword(s)||Planktonic foraminifera, tropical Indian Ocean, MiSo, Images, convolutional neural network, CNN, automated classification|
The planktonic foraminifera images contained in this dataset come from coretops sampled in the tropical Indian Ocean using the RV Marion Dufresne and the BARAT 94 cruise onboard the RV Baruna Jaya I. The samples are archived at Centre Européen de Recherche et d'Enseignement de Géosciences de l'Environnement (CEREGE, France) and Laboratoire des Sciences du Climat et de l'Environnement (LSCE, France). The planktonic foraminifera specimens in this database are from the >150 μm sieve fraction and images were captured using MiSo, a state-of-the-art microfossil sorting machine developed at CEREGE. Since the telecentric lens used for image capturing had a field depth of approximately 90 μm and therefore cannot fully capture most foraminifera under view, the captured images were fused to produce Z-stack images. A 70 μm separation between images was adopted for the stack. All images were outputted at a resolution of 1159.4 pixels per millimetre. A Convolutional Neural Network model (Base Cyclic 16), also developed at CEREGE (Adebayo et al., submitted to G3), was used to automatically identify and classify the images. Model accuracy was confirmed by comparing machine model classification with results from classification by human classifiers on coretop samples that were neither part of the training nor testing sets across 21 classes. Result shows 98% accuracy in the machine labels. For more details on the systems and procedures for image acquisition, transfer learning, input dimensions, training, model evaluation, and CNN selection, please see Marchant et al. (2020). This directory contains 185, 222 images belonging to 36 taxa classes of planktonic foraminifera, including foraminifera fragments.
Marchant, R., Tetard, M., Pratiwi, A., Adebayo, M. and de-Garidel Thoron, T. (2020). Automated analysis of foraminifera fossil records by image classification using a convolutional neural network. Micropalaeontology, 39, 183–202. https://doi.org/10.5194/jm-39-183-2020
Adebayo, M. B., Bolton, C., Marchant, R., Bassinot, F., Conrod, S., and de-Garidel Thoron, T. Environmental controls on size distribution of recent planktonic foraminifera in the tropical Indian Ocean: A calibration study. Geochemistry, Geophysics, Geosystems (Submitted).
|Acknowledgements||The authors thank Jean-Charles Mazur for their laboratory assistance. This project was funded by the CNRS-INSU LEFE IndSO grant, and ANR grants FIRST (ANR-15-CE4-0006-01) and iMonsoon (ANR-16-CE01-0004-01).|