WG 1 – Automation – Physical handling of foraminifera:
No system of handling and manipulation of objects in the size range of foraminifera have so far been finalized. Several prototypes and patents have been developed and filed and prove the feasibility of handling these microparticles.
WG 2 – Image acquisition
The strategy for imaging foraminifera will favor the acquisition of 3D or pseudo-3D images (as 3D would imply an omnidirectional scan) rather than 2D images. High resolution images have been shown to improve the classification rates of foraminifera [MacLeod, 2007]. Though a 3D imaging system would be optimal for recognition rates as some diagnostic characters may be hidden using a single view of a foraminifera, processing times by neural networks and morphometric methods might prevent the implementation of 3D imaging in the automaton. Our choice to best resolve this tradeoff between image resolution and processing time will be explored by comparing a set of different approaches.
WG3 – Recognition and classification of foraminifera
Many algorithms are now widely employed for pattern recognition (e.g. neural networks, Support Vector Machine, Random Forest, bag of features) and are used in many fields including biomedicine and recognition of living plankton. At CEREGE, neural networks are used for the recognition of coccolithophorids (SYRACO software) and more recently were improved by combined morphometric analyses. However this approach is effective for coccoliths that are analyzed on microscopic slides, and cannot be directly implemented for foraminifera because of their random 3D positioning on a sorting tray.
WG4 – Integration and validation for paleoceanographic studies
The prototype will be assembled at the end of the project on ATG’s premises, by integrating the physical handling part with the image acquisition line. Two time-series will be analyzed to prove that physical sorting reaches an acceptable threshold for further geochemical and morphological measurements.