DOI： 10.2326/osj.24.129

researchmap

Publishing type：Research paper (scientific journal)   Publisher：Japan Society of Medical Entomology and Zoology  

DOI： 10.7601/mez.75.177

researchmap

Publishing type：Research paper (scientific journal)   Publisher：Wiley  

Abstract

Machine learning‐based behaviour classification using acceleration data is a powerful tool in bio‐logging research. Deep learning architectures such as convolutional neural networks (CNN), long short‐term memory (LSTM) and self‐attention mechanism as well as related training techniques have been extensively studied in human activity recognition. However, they have rarely been used in wild animal studies. The main challenges of acceleration‐based wild animal behaviour classification include data shortages, class imbalance problems, various types of noise in data due to differences in individual behaviour and where the loggers were attached and complexity in data due to complex animal‐specific behaviours, which may have limited the application of deep learning techniques in this area.

To overcome these challenges, we explored the effectiveness of techniques for efficient model training: data augmentation, manifold mixup and pre‐training of deep learning models with unlabelled data, using datasets from two species of wild seabirds and state‐of‐the‐art deep learning model architectures.

Data augmentation improved the overall model performance when one of the various techniques (none, scaling, jittering, permutation, time‐warping and rotation) was randomly applied to each data during mini‐batch training. Manifold mixup also improved model performance, but not as much as random data augmentation. Pre‐training with unlabelled data did not improve model performance. The state‐of‐the‐art deep learning models, including a model consisting of four CNN layers, an LSTM layer and a multi‐head attention layer, as well as its modified version with shortcut connection, showed better performance among other comparative models. Using only raw acceleration data as inputs, these models outperformed classic machine learning approaches that used 119 handcrafted features.

Our experiments showed that deep learning techniques are promising for acceleration‐based behaviour classification of wild animals and highlighted some challenges (e.g. effective use of unlabelled data). There is scope for greater exploration of deep learning techniques in wild animal studies (e.g. advanced data augmentation, multimodal sensor data use, transfer learning and self‐supervised learning). We hope that this study will stimulate the development of deep learning techniques for wild animal behaviour classification using time‐series sensor data.

DOI： 10.1111/2041-210x.14294

researchmap

Publishing type：Research paper (scientific journal)   Publisher：Oxford University Press (OUP)  

Abstract

Rare behaviors displayed by wild animals can generate new hypotheses; however, observing such behaviors may be challenging. While recent technological advancements, such as bio-loggers, may assist in documenting rare behaviors, the limited running time of battery-powered bio-loggers is insufficient to record rare behaviors when employing high-cost sensors (e.g. video cameras). In this study, we propose an artificial intelligence (AI)-enabled bio-logger that automatically detects outlier readings from always-on low-cost sensors, e.g. accelerometers, indicative of rare behaviors in target animals, without supervision by researchers, subsequently activating high-cost sensors to record only these behaviors. We implemented an on-board outlier detector via knowledge distillation by building a lightweight outlier classifier supervised by a high-cost outlier behavior detector trained in an unsupervised manner. The efficacy of AI bio-loggers has been demonstrated on seabirds, where videos and sensor data captured by the bio-loggers have enabled the identification of some rare behaviors, facilitating analyses of their frequency, and potential factors underlying these behaviors. This approach offers a means of documenting previously overlooked rare behaviors, augmenting our understanding of animal behavior.

DOI： 10.1093/pnasnexus/pgad447

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1016/j.biocon.2023.110025

Scopus

researchmap

Publishing type：Research paper (scientific journal)   Publisher：Elsevier BV  

DOI： 10.1016/j.biocon.2023.109994

researchmap

Publishing type：Research paper (scientific journal)   Publisher：Proceedings of the National Academy of Sciences  

Cyclones can cause mass mortality of seabirds, sometimes wrecking thousands of individuals. The few studies to track pelagic seabirds during cyclones show they tend to circumnavigate the strongest winds. We tracked adult shearwaters in the Sea of Japan over 11 y and found that the response to cyclones varied according to the wind speed and direction. In strong winds, birds that were sandwiched between the storm and mainland Japan flew away from land and toward the eye of the storm, flying within ≤30 km of the eye and tracking it for up to 8 h. This exposed shearwaters to some of the highest wind speeds near the eye wall (≤21 m s ^–1 ) but enabled them to avoid strong onshore winds in the storm’s wake. Extreme winds may therefore become a threat when an inability to compensate for drift could lead to forced landings and collisions. Birds may need to know where land is in order to avoid it. This provides additional selective pressure for a map sense and could explain why juvenile shearwaters, which lack a map sense, instead navigating using a compass heading, are susceptible to being wrecked. We suggest that the ability to respond to storms is influenced by both flight and navigational capacities. This may become increasingly pertinent due to changes in extreme weather patterns.

DOI： 10.1073/pnas.2212925119

researchmap

Authorship：Lead author   Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.32313/jahes.38.1_109

researchmap

Publishing type：Research paper (scientific journal)   Publisher：Wiley  

DOI： 10.1002/ecy.3297

researchmap

Other Link： https://onlinelibrary.wiley.com/doi/full-xml/10.1002/ecy.3297

Authorship：Lead author   Language：English   Publishing type：Research paper (international conference proceedings)  

Web of Science

researchmap

Event date： 2020.3

Presentation type：Poster presentation  

researchmap

Event date： 2019.11

Presentation type：Poster presentation  

researchmap

Event date： 2019.11

Presentation type：Poster presentation  

researchmap

Event date： 2019.3

Presentation type：Poster presentation  

researchmap

Event date： 2019.3

Presentation type：Poster presentation  

researchmap

Presentation type：Oral presentation (general)  

researchmap