Preliminary Large-scale computer vision-based analysis of ant foraging dynamics enabled by novel mac

新型 mac 实现的基于大规模计算机视觉的蚂蚁觅食动力学初步分析

• 批准号: 2900538
• 金额: --
• 依托单位: Imperial College London
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2019
• 资助国家: 英国
• 起止时间: 2019 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2900538
• 关键词: Preliminary; Large; scale; computer; vision

In my PhD, I want to build on these first results and to study in detail the factors which underlie theontogeny of task choices in individual ants, and how these choices result in behavioural patternsobserved at the level of the colony, in particular how the colony adjusts the size-distribution of theirforaging parties when feeding on food sources of different properties. To this end, I will continue thedevelopment of versatile computer vision and machine learning tools. In particular, I will createclassifiers which allow the repeated identification of individuals across non-continuous recordings. Thetraining data required to power the deep learning algorithms behind this recognition will be providedthrough a cooperation with the Zoological Garden in Wuppertal, Germany, which has already been setup. We will conduct live experiments in the zoo, and encourage visitors to partake in data analysis.Through this "citizen science" approach, we will be able to train and test our algorithms on anunprecedented scale. The results of this project will provide novel insights into a fundamental biologicalquestion - the adaptive advantage and hence evolution of division of labour in eusocial insects - usingmodern engineering methods. Indeed, the models developed in this project will be versatile andapplicable in the analysis of various interactions between a large number of individuals.

在我的博士论文中，我想建立在这些初步结果的基础上，详细研究个体蚂蚁任务选择的个体发生的基础因素，以及这些选择如何导致在殖民地水平上观察到的行为模式，特别是当以不同性质的食物来源为食时，殖民地如何调整觅食方的大小分布。为此，我将继续开发多功能的计算机视觉和机器学习工具。特别是，我将创建分类器，允许在非连续记录中重复识别个体。为这种识别背后的深度学习算法提供动力所需的训练数据将通过与德国伍珀塔尔动物园的合作提供，该合作已经建立。我们将在动物园进行现场实验，并鼓励游客参与数据分析。通过这种“公民科学”方法，我们将能够以前所未有的规模训练和测试我们的算法。该项目的结果将为一个基本的生物学问题提供新的见解-适应性优势，从而在真社会性昆虫的分工进化-使用现代工程方法。事实上，在这个项目中开发的模型将是通用的，适用于分析大量个体之间的各种相互作用。