Enhanced Animal Behavioural Analytics For Improved Cattle Welfare, Health, Productivity and Sustainability

Quant Foundry人工智能(AI)牲畜监控解决方案与布里斯托尔大学布里斯托尔兽医学院合作，旨在为识别牛的异常行为提供世界级的解决方案，以帮助快速识别不同的疾病。该解决方案在自动化农场框架内结合了人工智能驱动的动物视频分析，以提高健康和福利，降低生产成本和排放。虽然现有许多远程监控动物的解决方案，但许多解决方案需要人员的积极参与，几乎没有潜在的成本节约。其他解决方案需要使用动物必须佩戴的物理硬件，这需要每只动物巨大的设置和维护成本。在国际上，有一些正在进行的图像识别研究试验，但很少提到它们用于识别特定的疾病。许多应用于牲畜的视频系统不能很好地随着动物数量的增加而扩展，而Quant Foundry系统可以识别和跟踪多个动物，而计算开销很小。独特的创新是我们的通用计算模型，带有标准的现成硬件，将能够识别每种动物的许多不同条件。这项研究和可行性研究将在两个方面进行：(I)对应用于我们的深度学习算法的关键动物行为特征进行分类和识别；(Ii)评估硬件和识别算法的商业有效性的商业可行性研究，以及识别异常行为(如跛行和其他异常动作)的算法。验证研究将涉及在农业-计划免疫中心的西南乳业发展中心安装该系统，以在9个月的时间内记录每头奶牛挤奶后的连续视频。通过布里斯托尔大学开发的现有人工智能系统，将识别每头奶牛，并将其与布里斯托尔兽医学校专家的生产/兽医数据和行为注释联系起来，并由外部评估员核实。这些数据将用于验证和改进Quant Foundry AI解决方案，也将作为实地资源进行公众传播。最后阶段将评估初级跛行、乳房炎和Johne病解决方案的整体有效性，并确定其商业化和研究的好处。这将导致进一步的研究，以推进基本的动物福利、行为和可持续性研究。

The Quant Foundry Artificial Intelligence (AI) Livestock Surveillance Solution in collaboration with Bristol Veterinary School at the University of Bristol aims to provide a world-class solution for the identification of anomalous cattle behaviour to aid in the rapid identification of different ailments. The solution combines AI-driven video analytics of animals within an automated farm framework to increase health and welfare and lower production costs and emissions.While there are a number of existing solutions for remote monitoring of animals, many require an active involvement of people with little potential cost savings. Other solutions require the use of physical hardware that must be worn by the animal, requiring significant per-animal setup and maintenance costs. Internationally there are a number of ongoing research trials for image recognition, however there is little mention of their use identifying specific illnesses. Many video systems applied to livestock do not scale well with the number of animals, whereas the Quant Foundry system can identify and track multiple animals with little computational overhead. The unique innovation is our general computing model with standard off-the-shelf hardware that will be able to identify many different conditions for each animal. This considerably reduces the time and cost of development, deployment and upkeep.This research and feasibility study will be performed across two areas: (i) classification and identification of key animal behaviour features to be applied to our deep learning algorithm, and (ii) a commercial feasibility study to assess the commercial effectiveness of the hardware and identification algorithm for identifying anomalous behaviours such as lameness and other abnormal motions. The validation study will involve an installation of the system at Agri-EPI Centre's South West Dairy Development Centre to record continuous video of every cow after milking over a 9-month period. Through an existing AI system developed by the University of Bristol, individual cows will be identified and linked to production/veterinary data and behavioural annotations from Bristol Vet School experts, verified by external assessors. This data will be used to validate and refine the Quant Foundry AI solution, and will also be curated for public dissemination as a resource for the field.The final stage will be to assess the overall effectiveness of the primary lameness, mastitis and Johne's disease solution and determine its benefits for commercialisation and research. This would lead to further studies to advance fundamental animal welfare, behaviour and sustainability research.