PlanktonID – Combining deep learning, in situ imaging and citizen science to resolve the distribution of zooplanktonin major upwelling regions

Kiko, Rainer GND; Christiansen, Svenja; Schröder, Simon-Martin; Koch, Reinhard; Stemmann, Lars

Recent publications revealed the global importance of single-celled zooplankton, belonging to the super group Rhizaria and highlighted the need of in-situ imaging to study these fragile organisms. The advance of in situ plankton imaging techniques leads to increasing amounts of image data sets that require identification to different taxonomic levels. Automatic classification by computer algorithms provides the means for fast data availability, however the accuracy of those algorithms still requires manual identification by humans. We combined state of the art automatic image classification by convolutional neural networks (deep learning) with a citizen science project to classify a large dataset of ~ 3 million images from an Underwater Vision Profiler 5 (UVP5). On our website, citizen scientists can confirm or reject the automatic assignment of UVP5 images to different plankton categories in a memory-like game. Inbuilt quality controls and multiple validations per image enable scientific analysis of the citizen science data. In total more than 500 users have validated more than 300.000 images until now. We will present further data on citizen scientist engagement, data quality assessment and the distribution analysis of large protists (Rhizaria) in the Mauretanian, Benguela and Humboldt Current upwelling systems.


Citation style:
Could not load citation form.


Use and reproduction: