WEKO3
アイテム
マルチビュー動物行動分析におけるディープラーニングの応用
https://doi.org/10.18997/0002000296
https://doi.org/10.18997/00020002960c8e0500-5cda-4b53-8054-8546093160b9
| 名前 / ファイル | ライセンス | アクション |
|---|---|---|
sei_k_470.pdf (5.3 MB)
|
|
| アイテムタイプ | 学位論文 = Thesis or Dissertation(1) | |||||||
|---|---|---|---|---|---|---|---|---|
| 公開日 | 2023-11-29 | |||||||
| 資源タイプ | ||||||||
| 資源タイプ識別子 | http://purl.org/coar/resource_type/c_db06 | |||||||
| 資源タイプ | doctoral thesis | |||||||
| タイトル | ||||||||
| タイトル | Multi-View Animal Behaviour Analysis with Deep Learning | |||||||
| 言語 | en | |||||||
| タイトル | ||||||||
| タイトル | マルチビュー動物行動分析におけるディープラーニングの応用 | |||||||
| 言語 | ja | |||||||
| 言語 | ||||||||
| 言語 | eng | |||||||
| 著者 |
Blanco Negrete, Salvador
× Blanco Negrete, Salvador
|
|||||||
| 抄録 | ||||||||
| 内容記述タイプ | Abstract | |||||||
| 内容記述 | Understanding and analyzing animal behaviour is essential in various scientific disciplines, including neuroscience, psychology, ecology, genetics, and pharmacology. Automated detection and analysis of animal behaviours can significantly enhance research efficiency and accuracy in these areas. However, existing systems often rely on engineered features and are restricted to single-view analysis. This dissertation presents a novel approach to multi-view animal behaviour detection using deep learning that does not require pose estimation or other engineered features, works with small amounts of data, and is flexible. This dissertation introduces two primary contributions. The first involves adapting state-of-the-art human pose estimation systems for animal pose estimation. A multiple-monkey pose estimation system and a multi-view 3D pose estimation for marmosets are introduced. The objective is to demonstrate the challenges and limitations of using pose estimation and feature engineering in general for behaviour analysis. The "In the Wild" dataset, known as MacaquePose, was collected containing monkey images in various environments to train a deep neural network for 2D multimonkey pose estimation. At the same time, the multi-view 3D marmoset dataset was collected in a laboratory setting. The second core contribution presents a novel multi-view behaviour detection system. The system captures behaviours using various perspectives, allowing a comprehensive understanding of the animal’s movements. The system uses three neural networks; the first neural network (NN1) extracts Regions of Interest (ROIs) for each view, and NN2 is a classification network that creates a heat map that encodes confidence for the desired behaviour across views and time within a time window. NN3 is trained to create a final prediction from the heat map. This approach allows the use of small amounts of data; it avoids using pose estimation or other engineered features, and the three networks can be trained separately or reused, making it easy to adapt to new behaviours or animals. The developed system was trained to detect rats’Wet Dog Shake (WDS) behaviour, making it the first to detect WDS behaviour. The WDS behaviour is relevant in studying various animal disease models, including acute seizures, morphine abstinence, and nicotine withdrawal. This behaviour has a short duration; it occurs spontaneously and infrequently, making it challenging to detect and analyze accurately. Still, using three views, the developed system can detect it with a precision of 0.91 and recall of 0.86. Notably, while this dissertation talks mainly about detecting WDS behaviour in rats, the developed multi-view deep learning system holds potential for broader applications in analyzing various animal behaviours. Multiple camera views enhance the system’s ability to generalize across perspectives, add redundancy, and reduce occlusion, resulting in more accurate and robust behaviour detection. Therefore, it can be adapted and extended to detect and analyze other animal behaviours in diverse species. In conclusion, this dissertation presents a novel approach to multi-view animal behaviour detection using deep learning. The developed system opens new avenues for research in animal behaviour and welfare, and its potential applicability across diverse species makes it a valuable tool for studying complex animal behaviours. | |||||||
| 言語 | en | |||||||
| 目次 | ||||||||
| 内容記述タイプ | TableOfContents | |||||||
| 内容記述 | 1 Introduction|2 Related Work|3 Multi-Camera Setups and Animal Preparation|4 Methodology|5 Results and Discussion|6 Conclusion | |||||||
| 言語 | en | |||||||
| 備考 | ||||||||
| 内容記述タイプ | Other | |||||||
| 内容記述 | 九州工業大学博士学位論文 学位記番号：生工博甲第470号 学位授与年月日: 令和5年9月25日 | |||||||
| キーワード | ||||||||
| 主題Scheme | Other | |||||||
| 主題 | multi-view | |||||||
| キーワード | ||||||||
| 主題Scheme | Other | |||||||
| 主題 | behaviour detection | |||||||
| キーワード | ||||||||
| 主題Scheme | Other | |||||||
| 主題 | rat | |||||||
| キーワード | ||||||||
| 主題Scheme | Other | |||||||
| 主題 | wet-dog shake (WDS) | |||||||
| キーワード | ||||||||
| 主題Scheme | Other | |||||||
| 主題 | nonhuman primates | |||||||
| キーワード | ||||||||
| 主題Scheme | Other | |||||||
| 主題 | deep learning | |||||||
| 学位授与番号 | ||||||||
| 学位授与番号 | 甲第470号 | |||||||
| 学位名 | ||||||||
| 学位名 | 博士（工学） | |||||||
| 学位授与年月日 | ||||||||
| 学位授与年月日 | 2023-09-25 | |||||||
| 学位授与機関 | ||||||||
| 学位授与機関識別子Scheme | kakenhi | |||||||
| 学位授与機関識別子 | 17104 | |||||||
| 学位授与機関名 | 九州工業大学 | |||||||
| 学位授与年度 | ||||||||
| 内容記述タイプ | Other | |||||||
| 内容記述 | 令和5年度 | |||||||
| 出版タイプ | ||||||||
| 出版タイプ | VoR | |||||||
| 出版タイプResource | http://purl.org/coar/version/c_970fb48d4fbd8a85 | |||||||
| アクセス権 | ||||||||
| アクセス権 | open access | |||||||
| アクセス権URI | http://purl.org/coar/access_right/c_abf2 | |||||||
| ID登録 | ||||||||
| ID登録 | 10.18997/0002000296 | |||||||
| ID登録タイプ | JaLC | |||||||
