Artificial intelligence for characterizing plankton traits from images

24-26 of April 2019, Villefranche-sur-Mer, France

The aim of the workshop is to gather researchers interested in applying machine learning to identify and quantify functional traits of aquatic organisms from individual images. A special focus will be given to plankton, but we also welcome participants studying other marine or freshwater organisms.

April 24th

12:00 Registration

13:00 Lunch

14:00 KEYNOTE - Cédric Pradalier - Image Processing in Interdisciplinary Projects: Experience and Lesson Learned

14:45-15:00 Amaya Alvarez Ellacuria - Image-based, unsupervised estimation of fish size from commercial landings using deep learning

15:00-15:15 Tom Lorimer - Forecasting plankton interaction dynamics using machine learning and nonparametric analysis of high-dimensional trait data

15:15-15:30 Thibault de Garidel-Thoron - Automatic recognition of calcareous microfossils

15:30-15:45 Guillaume Wacquet - Combination of machine learning methods and imaging-in-flow systems (FlowCam, CytoSense/Sub) for phytoplankton detection, classification and enumeration

15:45-16:15 Coffee & tea break

16:15-16:30 Arancha Lana - Automatic monitoring of marine fish using underwater baited cameras and machine learning methods

16:30-16:45 Laura Hoebeke - Automated hierarchical classification of animal species in camera trap images

16:45-17:00 Anya Waite - Insights from UVP images across the South Atlantic

17:00-18:45 First poster session

18:45-19:30 Walk and apéritif

20:00 Dinner

April 25th

09:00 KEYNOTE - Thomas Kiørboe - Traits-based approaches in aquatic ecology

09:45-10:00 Filippo Ferrario - Photographic sampling of benthic communities: which functional traits can we assess?

10:00-10:15 Heidi Sosik - From Billions of images to new insights in plankton ecology

10:15-10:30 Mark Ohman - Discerning zooplankton traits in situ from a new autonomous Zooglider

10:30-10:45 Fabio Benedetti - A first glimpse into the global patterns of zooplankton functional diversity from the TARA imaging datasets

10:45-11:15 Coffee & tea break

11:15-12:00 KEYNOTE - Jean-François Lalonde - Deep learning for understanding the image formation process

12:00-12:15 Simon-Martin Schröder - Step up your sorting game! Increasing efficiency with Machine Learning

12:15-12:30 Alexei Tsygvintsev - Artificial neural networks from the point of view of non-linear dynamics : local minima and convergence problems

12:30-12:45 Jeff Ellen - Improving plankton image classification using context metadata

12:45-13:00 Klas Ove Möller - Get it from the picture: extending the scope of plankton and particle imaging beyond distribution patterns

13:00-14:00 Lunch

14:00-15:00 Second poster session

15:00-16:00 Topical working groups (part 1)

16:00-16:30 Coffee & tea break

16:30-17h30 Topical working groups (part 2)

17h30-18h30 Plenary reports

20:00 Dinner

April 26th

09:00-09:30 Plenary work on "Quo Vadimus" paper

09:30-10:30 Parallel writing groups (1)

10:30-11:00 Coffee & tea break

11:00-12:00 Parallel writing groups (2)

12h00-12h30 Plenary reports and wrapping up

13:00 Lunch and goodbyes

Keynote speakers

Cédric Pradalier

Cédric Pradalier is Associate Professor in robotics and sensor processing at GeorgiaTech Lorraine, Metz.

His research interests focus on data-driven robotics for environment assessment and monitoring.

Thomas Kiørboe

Thomas Kiørboe is Professor at the Centre for Ocean Life, National Institute for Aquatic Resources, Technical University of Denmark.

His research interests encompass observing zooplankton using high speed video, studying the dynamics of particle-associated microbial communities and marine snow, and understanding how planktonic organisms swim and find their mate and their food.

Jean-François Lalonde

Jean-François Lalonde is an Associate Professor in Electrical and Computer Engineering at Laval University, Quebec City, since 2013. He is a former Post-Doctoral Associate at Disney Research, Pittsburgh.

His research interests are in computer vision and deep learning, with a particular focus on lighting estimation, 3D reconstruction, tracking, and augmented reality.

Posters

  • Sari Giering - From optics to export: imaging the biological carbon pump
  • Frederic Maps - The ARTIFACZ project: machine learning application to the identification and measurement of functional traits of zooplankton from individual images
  • Josep Alós - A novel multiple-objects high-resolution tracking system to measure human spatial behavior in coastal social-ecological systems
  • Martin Laviale - Automatic diatom identification using a deep learning approach
  • Nerea Valcárcel Pérez - Zooplankton communities structure from an upwelling coastal zone to an oligotrophic gyre in the Alboran Sea (SW Mediterranean Sea)
  • Tristan Biard - In situ imaging in an upwelling system - Past, present and future perspectives POSTER
  • Jean-Baptiste Romagnan - Prospective in the use of ML and IA applied to the Ecosystemic Approach to Fisheries at Ifremer
  • Barbara Niehoff - What can we learn from high-resolution distribution data: ecological information on Arctic copepod species
  • Érica Caroline Becker - Imaging and traditional taxonomy: an integrated approach for copepod biodiversity on the Southwestern Atlantic Ocean
  • Morten Iversen - Impact from zooplankton on the efficiency of the biological carbon pump from in situ optics and direct video observations
  • Fabien Lombard - Global structure of planktonic populations using quantitative imaging methods
  • Michiel Stock - Pairwise learning to predict species interaction networks

Context

The base of marine ecosystems is supported by planktonic organisms that have adapted to extreme environmental conditions. These adaptations, often shared by many species, represent “functional traits” that influence the fitness of individuals as well as ecosystem functioning. A better understanding of these traits appears crucial to predict the responses of marine ecosystems to the unprecedented changes affecting the Ocean. Several traits are associated to morphological features (e.g. size, shape, lipid stores, etc.), hence allowing automatic detection and measurement from images.

Imaging methods for plankton studies have multiplied and rapidly improved over the past decade. They have led to the production of massive amounts of images that are taxonomically classified and stored in easily accessible repositories and therefore particularly amenable to machine learning approaches. Such approaches provide the opportunity to revolutionize the field of plankton ecology and have immediate consequences for scientific research and ecosystem health assessment.

Thus, we wish to foster the development of new tools combining imaging methods and machine learning algorithms to automatically detect and measure important functional traits of planktonic organisms.

Researchers working on estimating the functional traits of other aquatic organisms from imaging and/or artificial intelligence tools are also welcome to participate in order to bridge the transdisciplinary gap .