ABI Innovation: Computational Methods for Bioacoustic Avian Species Monitoring

该项目的重点是开发一个计算框架，用于从现场音频记录中智能监测鸟类物种。该系统将从麦克风集合中收集音频，以自动和自适应地推断不同鸟类的时空存在/不存在和丰度。该系统将自适应地寻求、请求和纳入专家反馈，以提高适应新环境的能力。通过在相对较大的空间尺度上以极精细的时间分辨率自动收集鸟类数据，我们将解决两个基本的生态和保护重要性问题（1）动物行为能否缓冲物种对环境变化的影响？具体而言，鸟类是否有能力在快速或极端环境变化的情况下转移繁殖区域（例如，天气）？(2)热带森林的丧失和破碎如何改变鸟类的分布？这项研究将建立在最近开发的机器学习多实例多标签框架的基础上，并寻求自适应和主动地训练物种存在/不存在和丰度估计模型，目标是最大限度地提高学习效率，同时最大限度地减少人类标记工作。该系统主动识别和适应新物种（和环境）的独特能力将大大减少部署和更新生物声学监测系统所需的开销。该项目将为研究生物多样性作为全球生境丧失和气候变化的函数提供计算创新，为生态系统及其对人类活动的反应提供重要的科学知识。该项目还将为收集鸟类种群数据提供一个新的技术标准，可在全球范围内部署，以监测一系列多样化和不断发展的物种。该项目提供的以研究为基础的教育和培训机会将有助于俄勒冈州州立大学在生态系统信息学的新兴领域培养新一代的研究人员。外联活动将1）让高中教育工作者参与俄勒冈州自然资源教育（ONRE）计划组织的研讨会，2）与国际会议和研讨会一起组织年度数据挑战/竞赛，3）通过REU计划从代表性不足的群体中招募女本科生和K-12学生从事计算机科学和工程职业。有关该项目的更多信息，请访问http://eecs.oregonstate.edu/research/bioacoustics/。

This project focuses on the development of a computational framework for intelligent monitoring of avian species from in situ audio recordings. The system will gather audio from a collection of microphones to automatically and adaptively infer spatio-temporal presence/absence and abundance for different bird species. The system will adaptively seek, request, and incorporate expert feedback to improve the ability to adjust to a new environment. By enabling the automatic collection of bird data at extremely fine temporal resolutions over relatively large spatial scales, we will address two questions of fundamental ecological and conservation importance (1) Can animal behavior buffer species against the effects of environmental changes? Specifically, do birds have the capacity to shift breeding territories in instances of rapid or extreme environmental changes (e.g., weather)? (2) How does loss and fragmentation of tropical forest alter the distributions of bird species?This research will build upon the recently developed Multi-Instance Multi-Label framework from machine learning, and seek to adaptively and actively train species presence/absence and abundance estimation models with the goal of maximizing the learning efficiency while minimizing the human labeling efforts. The system's unique ability to actively identify and adapt to new species (and environments) will greatly reduce the overhead required to deploy and update the bioacoustic monitoring system. This project will provide computational innovations for studying biodiversity as a function of global habitat loss and climate change, contributing significant scientific knowledge about ecosystems and their responses to human activities. The project will also provide a new technology standard for collecting bird population data that could be deployed worldwide for monitoring a diverse and evolving set of species. Research-based education and training opportunities offered by this project will help prepare a new generation of researchers in the emerging area of Ecosystem Informatics at Oregon State University. Outreach activities will 1) involve high school educators in workshops organized by the Oregon Natural Resources Education (ONRE) program, 2) organize annual data challenges/competitions in association with international conferences and workshops, 3)recruit female undergraduates and K-12 students from under-represented groups to careers in computer science and engineering through REU programs. Further information on this project can be found at http://eecs.oregonstate.edu/research/bioacoustics/.