FuSe: Technologies For Bioacoustic Sensing

生物多样性正面临着前所未有的下降，而地球生态系统的压力却在继续增加。认识到生物多样性的现状及其对人类福祉的好处，世界各国政府在2010年承诺采取有效和紧急行动，通过《生物多样性公约》的目标遏制生物多样性的丧失。这些目标需要监测，以评估在实现具体目标方面取得的进展。这种大规模的生物多样性评估需要能够提供对物种分布、丰度和随时间变化的大规模模式的理解的方法。这依赖于在区域到国家范围内收集具有代表性的数据的调查，以及能够提供对物种种群的知情了解的可靠分析。作为一个群体，蝙蝠(翼手目)的监测尤其具有挑战性，因为大多数蝙蝠是夜间活动的，范围广泛，很难识别。在历史上，对温带地区蝙蝠的监测侧重于密集的基于现场的视觉计数或捕获调查。这些方法有相当大的价值，但很难从这些方法中自信地推断出在更广泛的人口层面上正在发生什么。在过去的几十年里，英国一直在使用声学调查来监测蝙蝠(例如，蝙蝠保护信托基金的国家蝙蝠监测计划)，但直到最近，传感器技术和分析声学工具的进步才使识别和监测更多的易识别物种成为可能。许多第一批从录音中自动检测和识别蝙蝠物种的开源工具都是从我们之前由NERC资助的研究中开发出来的。我们开发了欧洲蝙蝠物种的声学参考库和第一个自动机器学习分类器。自那以后，我们通过EPSRC和Zooniverse资助的项目，在这项工作的基础上，利用最新的机器学习技术(卷积神经网络，CNN)，并通过这一技术开发了一条开源管道，用于检测搜索阶段的回声定位呼叫和物种识别。尽管最近在声学物种识别方面取得了令人兴奋的发展，但在可扩展的监测工具中使用它们的成本效益仍然存在巨大挑战。规模化部署的一个主要障碍是声传感器的成本，通常成本高达1000 GB。作为该联盟的一部分，牛津大学的研究重点是开发低成本传感器(http://soundtrap.io)，)，它可以将未压缩的音频记录到SD卡中。目前的传感器并不是为记录蝙蝠的高频而设计的，但成本低于GB 50的原型版本已经被改装并部署在记录蝙蝠的试验中。FUSE(生物声学传感技术)研究联盟将蝙蝠声学分析(伦敦大学学院)、计算机科学和开源传感器硬件(牛津大学)、大规模公民科学(英国鸟类学信托基金)和国家级蝙蝠种群监测(蝙蝠保护信托基金)方面的专业知识结合在一起。将蝙蝠物种鉴定的尖端分析工具与新型低成本传感器的开发相结合，并在解释通过大规模志愿者声学调查和蝙蝠监测收集的数据方面的专业知识，我们将创建一个端到端的蝙蝠种群声学监测的开源系统。这是一项具有催化作用的提议，对BAT和生物声学监测的未来有着巨大的影响。

Biodiversity is facing an unprecedented decline whilst the pressure on the earth's ecosystems continues to grow. Recognising the status of biodiversity and its benefit to human wellbeing, the world's governments committed in 2010 to take effective and urgent action to halt biodiversity loss through the Convention on Biological Diversity's targets. These targets require monitoring to assess progress towards specific goals. Such large-scale biodiversity assessment calls for methods which are able to provide an understanding of large-scale patterns in species' distributions, abundances and changes over time. This relies on surveys to collect data that are representative at a regional to national scale, and robust analysis that is able to provide an informed understanding of species' populations.As a group, bats (Chiroptera) are particularly challenging to monitor because most are nocturnal, wide-ranging and difficult to identify. Historically the monitoring of bats in temperate regions has focused on intensive site-based visual counts or capture surveys. There is considerable value in these approaches, but it is difficult to confidently infer from these what is happening at a wider population level. Acoustic surveys have been used in the UK to monitor bats for the past few decades (e.g. Bat Conservation Trust's National Bat Monitoring Programme), but it is only very recently that advances in sensor technology and analytical acoustic tools have made it possible to identify and monitor more than a handful of easy to identify species. Many of the first open-source tools for automatically detecting and identifying bat species from sound recordings were developed from our previous NERC funded research. We developed acoustic reference libraries for European bat species and the first automatic machine learning classifier. We have since built on this work through our recent EPSRC and Zooniverse funded projects, to make use of the latest machine-learning technologies (Convoluted Neural Networks, CNNs) and through this developed an open-source pipeline for the detection of search-phase echolocation calls and species identification.Despite recent and exciting developments in acoustic species identification, there remain substantial challenges for their cost-effective use within a scalable monitoring tool. A major barrier for deployment at scale is the expense of acoustic sensors, which typically cost up to £1000. As part of this consortium, research at Oxford University has focused on the development of low-cost sensors (http://soundtrap.io), which record uncompressed audio to an SD card. The current sensors are not designed to record at high frequency for bats, but prototype versions, which would cost less than £50 have already been modified and deployed in trials to record bats. The FuSe (Technologies For Bioacoustic Sensing) research consortium brings together this expertise in bat acoustic analysis (University College London), computer science and open source sensor hardware (Oxford University), with large-scale citizen science (British Trust for Ornithology), and national-level bat population monitoring (Bat Conservation Trust). Integrating cutting edge analytical tools for the species identification of bats with the development of new low cost sensors, and expertise in the interpretation of data collected through large-scale volunteer-based acoustic surveys and bat monitoring, we will create an end-to-end open-source system for the large-scale acoustic monitoring of bat populations. This is a catalytic proposal, which has huge implications for the future of bat and bioacoustic monitoring.