Machine-learning and image recognition to monitor spatio-temporal changes in the behaviour and dynamics of species interactions

机器学习和图像识别，用于监测物种相互作用的行为和动态的时空变化

• 批准号: 2131794
• 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: 2018
• 资助国家: 英国
• 起止时间: 2018 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2131794
• 关键词: Machine; learning; image; recognition; monitor

Changes in land use and climate affect the distribution, abundance and behaviour of species globally. Management and mitigation of these processes requires a good understanding of how they affect wildlife, which requires effective monitoring over large areas and long time periods. Mammals are often scarce, shy, elusive or nocturnal, so are difficult to monitor. The increasing affordability of camera traps - devices that can be deployed over long periods to take photographs of passing wildlife, 24 hours a day - has contributed to their popularity as a survey method. Camera trapping has revolutionised prospects for monitoring mammals but generates huge numbers of images to be analysed. Some projects have successfully turned to "citizen scientists" to help with image processing but demand is likely to outstrip supply in the near future.Over the past decade, analytical techniques have been developed to enable the estimation of key community, population and behavioural parameters from camera trap imagery, yielding data on not only distribution and abundance, but also parameters related to activity, movement and inter-specific interactions, applicable across a wide range of both individually recognisable and unmarked species. However, these techniques can increase the burden of image processing, begging the questions:- Can automated image analysis make these promising techniques more accessible and widely used? - Can the widespread, long term use of these techniques catalyse a step change in our understanding of the status of mammal populations and the processes driving change?In this project, working with an interdisciplinary team of ecologists and computer scientists, the student will mobilise machine learning techniques to push forward our ability to automatically process large quantities of camera trap imagery, with a particular focus on using the data generated to understand the dynamics of interactions between species and their environment. The ecological insights generated will be aimed at informing the management of threatened species.

土地使用和气候的变化影响到全球物种的分布、丰度和行为。管理和缓解这些过程需要很好地了解它们如何影响野生动物，这需要在大面积和长时间内进行有效监测。哺乳动物通常稀少、害羞、难以捉摸或夜间活动，因此很难监测。相机陷阱-可以长时间部署的设备，每天24小时拍摄经过的野生动物的照片-越来越便宜，这有助于它们作为一种调查方法的普及。相机诱捕技术彻底改变了监测哺乳动物的前景，但也会产生大量需要分析的图像。一些项目已成功地求助于“公民科学家”来帮助进行图像处理，但在不久的将来很可能会供不应求，过去十年来，开发了分析技术，以便能够从相机陷阱图像中估计关键的社区、人口和行为参数，不仅产生关于分布和丰度的数据，而且还产生与活动、运动和种间相互作用有关的参数，适用于广泛的个体可识别和未标记的物种。然而，这些技术可能会增加图像处理的负担，引发以下问题：自动化图像分析能否使这些有前途的技术更容易获得和广泛使用？- 这些技术的广泛、长期使用能否催化我们对哺乳动物种群状况和驱动变化的过程的理解发生一步变化？在这个项目中，与生态学家和计算机科学家组成的跨学科团队合作，学生将利用机器学习技术来推动我们自动处理大量相机陷阱图像的能力，特别关注使用生成的数据来了解物种与环境之间相互作用的动态。所产生的生态见解将旨在为受威胁物种的管理提供信息。