Automated micropaleontology application

SYRACO (SYstème de Reconnaissance Automatique de COccolithes) has been under development since 1994. The first neural networks using backpropagation were written in C++ by Denis Dollfus (CEREGE thesis in 1997) in an original version (Dollfus and Beaufort, Fat neural network for recognition of position-normalised objects, Neural Networks, 1998).

A prototype microscope was automated at CEREGE in 1996 and then replaced by an automated microscope by Leica in 1997.

SYRACO allowed the first publication of scientific results from artificial intelligence (Beaufort et al., ENSO-like forcing on Oceanic Primary Production during the late Pleistocene,  Science 2001).

SYRACO then evolved following technological advances in computers (speed), neural networks (e.g.: increase in the number of layers), cameras, optics (successive switch from linear, rotational, circular polarization to bidirectional circular polarization (Beaufort et al., A universal method for measuring the thickness of microscopic calcite crystals, based on bidirectional circular polarization, Biogeosc., 2021).

The first study of automatically detected microcharcoal was published in 2003 (Beaufort et al., Continental Biomass Burning and Oceanic Primary Production Estimates in the Sulu Sea Over the Last 380 kyr and the East Asian Monsoon Dynamics, Mar. Geol. 2003)

The detection of non-fossil coccospheres in marine water samples was published in 2008: Beaufort et al. Calcite production by Coccolithophores in the South East Pacific Ocean, BioGeosc. 2008.

Thanks to the ANR FIRST grant, we have developed a system for imaging and automatically sorting foraminifera (contact Thibault de Garidel-Thoron : garidel@cerege.fr(Marchant et al., Automated analysis of foraminifera fossil records by image classification using a convolutional neural network.  J. Micropalaeontol. 2020.

Automatic classification of radiolarians started in 2020 (Tetard, et al., A new automated radiolarian image acquisition, stacking, processing, segmentation and identification workflow. Climate of the Past 2020) together with pollen (Bourel, et al., Automated recognition by multiple convolutional neural networks of modern, fossil, intact and damaged pollen grains, Computers & Geosciences 2020)

An automated culture monitoring system with daytime and "night-time" shooting that accepts up to 48 simultaneous cultures of distinct coccolithophora strains in a controlled CO2 atmosphere is now available (e.g. Suchéras-Marx et al. (2022). "Coccolith size rules - What controls the size of coccoliths during coccolithogenesis?" Marine Micropaleontology 2022).