Intellipig: An automated on-farm pig health monitoring system

我们提出的系统是一个完全非侵入式的基于面部的（人工）智能监测系统，可以自动捕获福利和健康数据，用于监测和管理猪。为什么？能够持续监测和评估农场动物的健康和福利取决于部署实用、有效和可靠的测量工具。目前的农场福利评估协议包括工作人员的日常抽查，兽医或质量保证检查员的定期抽查。所评估的福利参数通常以资源为基础，涉及基本需求的供应，或以动物为基础，主要关注诸如身体状况等易于测量的因素。大多数是在群体层面进行的，因为个人识别可能很困难。动物的行为很少被记录下来，而能告诉我们动物个体情绪状态的方法就更罕见了。这个系统将如何运作？我们的创新方法是使用一种完全非侵入性的基于面部的（人工）智能监控系统来测量动物的情绪状态和身体状况。我们已经成功开发了机器学习算法，可以使用面部生物识别技术识别个体，并能够检测面部表情的变化，从而表明猪是否有压力。我们还利用这些技术开发了身体状况评分和体重估计。在这里，我们建议将所有这些功能组合成一个基于面部的、非侵入式的动物健康和福利监测站，用于商业农场。通过采用这些最先进的机器学习技术，我们的系统将提供对单个动物的持续学习能力，从而允许早期发现健康/福利的改变，个性化的干预阈值，以及量身定制的治疗方法。这种个性化的数据记录可以与其他可测量的参数相结合，例如个人食物和水的摄入量、处理历史、生长和体重增加，这将使农场生产效率得到更好的优化。

Our proposed System is a completely non-intrusive face-based (artificial) intelligent monitoring system that can automatically capture welfare and health data for monitoring and management of pigs.Why? Being able to continuously monitor and assess farm animal health and welfare depends on the deployment of practical, valid, and reliable measurement tools. Current on-farm welfare assessment protocols involve daily spot-checks by staff, and periodic spot-checks by veterinarians or quality assurance inspectors. Assessed welfare parameters are often resource-based concerning provision for basic needs, or animal-based, looking mainly at easily-measurable factors such as physical condition. Most are performed at a group level as individual identification can be difficult. Rarely is animal behaviour recorded and even rarer still are measures that can tell us something about the emotional state of the individual animal.How will the system work? Our innovative approach is to measure an animal's emotional state as well as its body condition using a completely non-intrusive face-based, (artificial) intelligent monitoring system. We have already successfully developed machine learning algorithms that identify individuals using facial biometrics and are able to detect changes in facial expression that indicate whether a pig is stressed or not. We have also developed body condition scores and weight estimation using these techniques. Here we propose to combine all of these capabilities into one face based, non-intrusive animal health and welfare monitoring station for use on commercial farms. By employing these state-of-the-art machine learning techniques, our system will offer the capacity for on-going learning about individual animals, and consequently allow for early detection of altered health/welfare, personalised thresholds for intervention, and tailored treatment approaches. Such individualised data recording can be integrated with other measurable parameters, such as individual food and water intake, treatment history, growth and weight gain, which will allow better optimisation of farm production efficiency.