PFI-TT: Using machine listening for non-invasive monitoring of the status and wellbeing of commercial poultry flocks

这项创新技术转化伙伴关系（PFI-TT）项目的更广泛影响/商业潜力是提高我们食品供应链中商业养鸡的健康和生产力。家禽业竞争激烈，家禽已成为生产效率最高的动物蛋白之一。美国每年养殖数十亿只禽鸟，其中大多数是由独立农场主与大型生产商合作饲养的。该项目旨在对鸡群进行无创、持续的监测，及时提醒养殖户注意问题，并帮助养殖户评估其鸡群的福利和需求。预计最终结果将使鸟类更健康，得到更好的照顾，并提高农场生产力。拟议的项目使使用机器学习来监控音频环境成为可能，在这种情况下是养鸡场，而不需要大量标记的训练数据。通常用于音频机器学习的方法需要大量标记训练样本集和调优。然而，人类可以在特定情况下熟悉这些声音，并且在没有经过大量训练的情况下注意到不属于自己的事情。通过使用学习环境中声音结构和学习这些声音的典型变化的算法，该项目中使用的方法可以像人类听众一样注意到农场音景中的重大变化（异常）。这些异常的特征是严重的，可以引起农民或研究人员的注意，他们可以选择标记或忽略新声音。标记的声音可以用来自动标记未来的异常。该项目将在多个商业养鸡场部署监听设备，收集连续的音频记录和元数据，并利用收集到的数据提高系统的性能和实用性，以监测鸡的福利和它们的环境。这种倾听方法是无创的，不会给动物造成压力。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact / commercial potential of this Partnerships for Innovation-Technology Translation (PFI-TT) project is to enhance the well-being and productivity of commercially raised chickens in our food supply chain. The poultry industry is highly competitive, and poultry has become one of the most efficient animal proteins to produce. Billions of birds are grown in the United States each year and most are raised by independent farmers in cooperation with larger producers. This PFI project is aimed at providing non-invasive, continuous monitoring of chicken flocks to alert the farmer to problems in a timely manner and aid the farmer in assessing the welfare and needs of his flocks. The end result is expected to be healthier and better-cared for birds as well as improved farm productivity.The proposed project makes it possible to use machine learning to monitor audio environments, in this case chicken farms, without needing extensive labeled training data. The approaches typically used for audio machine learning require extensive labeled training sample sets and tuning. However, humans can become familiar with the sounds in a particular situation, and notice when something occurs that does not belong, without significant training. By using algorithms that learn the structure of sounds in an environment and learn the typical variation of those sounds, the methods used in the proposed project can act similarly to a human listener to notice significant changes (anomalies) in a soundscape on a farm. The anomalies are characterized by severity and can be brought to the attention of a farmer or researchers who can choose to label or ignore the new sound. Labeled sounds can then be used to automatically label future anomalies. This project will deploy listening devices into multiple commercial chicken farms, collect continuous audio recordings and metadata, and use the collected data to improve the performance and usefulness of the systems for monitoring the chickens' welfare and their environment. This listening method is non-invasive and causes no stress to the animals.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.