JC2 nov 2019 · 2019-10-22 · Microsoft PowerPoint - JC2 nov 2019 Author: ljeantet Created Date:...
Transcript of JC2 nov 2019 · 2019-10-22 · Microsoft PowerPoint - JC2 nov 2019 Author: ljeantet Created Date:...
Behavioural inference from signal processing using on-board multi-sensor loggers: a novel solution to improve the knowledge on ecology of sea turtleLorène Jeantet1, Víctor Planas-Bielsa2, Simon Benhamou3, Sebastien Geiger1, Jordan Martin1, Flora Siegwalt1, Pierre Lelong1, Julie Gresser4, Denis Etienne4, Gaëlle Hiélard5, Alexandre Arque5, Sidney Regis1, Nicolas Lecerf1, Cédric Frouin1, Abdelwahab Benhalilou8, Céline Murgale8, Thomas Maillet8, Lucas Andreani8, Guilhem Campistron8, Hélène Delvaux6, Christelle Guyon6, Sandrine Richard7, Fabien Lefebvre1, Nathalie Aubert1, Caroline Habold1, Yvon le Maho1,2, Damien Chevallier1.
Introduction• Sea turtle = endangered species, long-lives and migratory species• What is happening under water ?• Identification of underwater behaviours is required to predict activities and time budget over long period
adaptive conservative measures• Using remote multi-sensor recorders
Materials & Methods
1Institut Pluridisciplinaire Hubert Curien, CNRS–Unistra, 67087 Strasbourg, France.2Centre Scientifique de Monaco, Département de Biologie Polaire, 8 quai Antoine Ier, MC 98000 Monaco.3 Centre d’Écologie Fonctionnelle et Évolutive, CNRS, 1919 route de Mende, 34293 Montpellier Cedex, France.4DEAL Martinique, Pointe de Jaham, BP 7212, 97274 Schoelcher Cedex, France. 5Office de l’Eau Martinique, 7 Avenue Condorcet, BP 32, 97201 Fort-de-France, Martinique, France. 6 DEAL Guyane, Rue Carlos Finley, CS 76003, 97306 Cayenne Cedex, France. 7Centre National d'Etudes Spatiales, 2 place Maurice Quentin, 75039 Paris Cedex 01, France.8Association POEMM, 73 lot papayers, Anse a l'âne, 97229 Les Trois Ilets, Martinique.
* Corresponding author: Lorène Jeantet [email protected]
Aims of this study : Validation of behavioural identification from multi-sensor signalsAutomatic identification of behaviours from multi-sensor signals
Nishizawa et al. 2013
Free-ranging immature green turtles (n=13)equipped with a device combining :
• video-recorder• tri-axial accelerometer• tri-axial gyroscope• depth recorder
Raw acceleration, gyroscope and depth profiles for several behaviours expressed by one green turtle
Workflow of the automatic behavioural identification from acceleration, angular speed and depth data adapted to the green turtle.
Results
The overallprocedureidentifiedunderwaterbehaviours with an accuracy of 95%
Pie chart of the actual (determined from the video) vs. predicted mean durations of the various behaviours displayed by 3 immature free-ranging green turtles.
Comparison of the 9 inferred main behavioural categories (in red) and of the observed ones (in blue) for several hours for immature green turtle
DiscussionValidation of the behavioural signals
promote the use of multi-sensor recorders for a better understanding of sea turtle ecology Automatic behavioural identification mixed model : « craft » approach combined with machine learning approach
able to predict fine-scale behaviours as “Feeding” ,“Scratching” and “Resting”easily replicable and adaptable to other marine species
ConclusionMulti-sensor miniaturised logger + automatic behavioural identification procedure long term time budget estimation of sea turtle Identification of sea turtle’s feeding behaviours improve our understanding on their energy strategy
Support the establishment of appropriate conservative actions
Heave(z-axis)Sway
(x-axis)
Surge(y-axis)
Data Collection Identification of behavioural signals Automatic behavioural identification modelling
© Fabien Lefebvre
Rawvalues
of depth
Rawvalues
of depth
Segmentation according to
the depthand dive duration
Depth>0.3m and duration >
5s
Computation of descriptive statistics
trained learning algorithms
Calculation of time budget
Prediction of surface behaviour
Raw values of depth
Raw values of depth
Stay at surface
Breathing
duration> 6s
< 6s
DBA (Window size=2s ) and RA computation
Raw values of acceleration, depth and angular speed
Prediction of diving behaviour
Correction of the acceleration and angular speed tilt
PELT Segmentation according to : Depth (pen.value=50) DBA (pen.value=50) Pitch speed gy (pen.value=20)