Lorène Jeantet

Accelerometers are becoming ever more important sensors in animalattached technology, providing data that allow determination of body posture and movement and thereby helping to elucidate behaviour in animals that are difficult to observe. We sought to validate the identification of sea turtle behaviours from accelerometer signals by deploying tags on the carapace of a juvenile loggerhead (Caretta caretta), an adult hawksbill (Eretmochelys imbricata) and an adult green turtle (Chelonia mydas) at Aquarium La Rochelle, France. We recorded tri-axial acceleration at 50 Hz for each species for a full day while two fixed cameras recorded their behaviours. We identified behaviours from the acceleration data using two different supervised learning algorithms, Random Forest and Classification And Regression Tree (CART), treating the data from the adult animals as separate from the juvenile data. We achieved a global accuracy of 81.30% for the adult hawksbill and green turtle CART model and 71.63% for the juvenile loggerhead, identifying 10 and 12 different behaviours, respectively. Equivalent figures were 86.96% for the adult hawksbill and green turtle Random Forest model and 79.49% for the juvenile loggerhead, for the same behaviours. The use of Random Forest combined with CART algorithms allowed us to understand the decision rules implicated in behaviour discrimination, and thus remove or group together some ‘confused’ or under-represented behaviours in order to get the most accurate models. This study is the first to validate accelerometer data to identify turtle behaviours and the approach can now be tested on other captive sea turtle species.

Artificial neural networks are powerful supervised learning algorithms that are based on deep learning and have been poorly exploited in movement ecology.In this study, we adapt a fully convolutional neural network that was originally developed for biomedical 3D image segmentation: the V-net. We test it on a labelled dataset collected from animal-borne video recorders combined with multi-sensors (accelerometers, gyroscopes and depth recorders) deployed on free-ranging immature green turtles (Chelonia mydas). The proposed model, fitted for 1D data, is able to predict six behavioural categories for green turtles with an AUC score of 88%. Thus, diverted from its initial purpose and tested on sea turtle, the V-net is a very efficient method of behavioural identification that should be easily generalized to a wide range of species.

Bioacoustics, the exploration of animal vocalizations and natural soundscapes, has emerged as a valuable tool for studying species within their habitats, particularly those that are challenging to observe. This approach has broadened the horizons of biodiversity assessment and ecological research. However, monitoring wildlife with acoustic recorders produces large volumes of data that can be labor-intensive to analyze. Deep learning has recently transformed many computational disciplines by enabling the automated processing of large and complex datasets and has gained attention within the bioacoustics community. Despite the revolutionary impact of deep learning on acoustic detection and classification, attaining both high detection accuracy and low false positive rates in bioacoustics remains a significant challenge. An intriguing yet unexplored avenue for enhancing deep learning in bioacoustics involves the utilization of contextual information, such as time and location, to discern animal vocalizations within acoustic recordings. As a first case study, a multi-branch Convolutional Neural Network (CNN) was developed to classify 22 different bird songs using spectrograms as a first input, and spatial metadata as a secondary input. A comparison was made to a baseline model with only spectrogram input. A geographical prior neural network was trained, separately, to estimate the probability of a species occurring at a given location. The output of this network was combined with the baseline CNN. As a second case study, temporal data and spectrograms were used as input to a multi-branch CNN for the detection of Hainan gibbon (Nomascus hainanus) calls, the world’s rarest primate. Our findings demonstrate that adding metadata to the bird song classifier significantly improves classification performance, with the highest improvement achieved using the geographical prior model (F1-score of 87.78% compared to 61.02% for the baseline model). The multi-branch CNNs also proved efficient (F1-scores of 76.87% and 78.77%) and simpler to use than the geographical prior. In the second case study, our findings revealed a decrease in false positives by 63% (94% of the calls were detected) when the metadata was used by the multi-branch CNN, and an increase of 19% in gibbon detection. This study has uncovered an exciting new avenue for improving classifier performance in bioacoustics. The methodology described in this study can assist ecologists, wildlife management teams, and researchers in reducing the amount of time spent analyzing large acoustic datasets obtained from passive acoustic monitoring studies. Our approach can be adapted and applied to other calling species, and thus tailored to other use cases.

The accelerometer, an onboard sensor, enables remote monitoring of animal posture and movement, allowing researchers to deduce behaviors. Despite the automated analysis capabilities provided by deep learning, data scarcity remains a challenge in ecology. We explored transfer learning to classify behaviors from acceleration data of critically endangered hawksbill sea turtles (Eretmochelys imbricata). Transfer learning reuses a model trained on one task from a large dataset to solve a related task. We applied this method using a model trained on green turtles (Chelonia mydas) and adapted it to identify hawksbill behaviors such as swimming, resting and feeding. We also compared this with a model trained on human activity data. The results showed an 8% and 4% F1-score improvement with transfer learning from green turtle and human datasets, respectively. Transfer learning allows researchers to adapt existing models to their study species, leveraging deep learning and expanding the use of accelerometers for wildlife monitoring.