Camera based automation of the analysis of personalities and social structures of a herd of dairy cows using neural networks and dynamic network analysis

农场动物的个体需求和行为影响动物的健康和生产力，在动物友好型畜牧业中应予以考虑。这个精密畜牧业项目的目的是分析传统群体饲养的奶牛的不同个性，并了解个性的交互作用以何种方式影响牛群结构。将录制长期视频，并希望自动化观察日常行为(例如，喂食、使用奶牛刷子、使用的路径)以及确定奶牛之间的社会互动。作为个体动物个性常规评估的数据基础，将对被分析群体中的所有奶牛进行良好的气质测试。然后解释人格特征和日常活动之间的关系。为了实现独立于人类观察者的行为和动物相互作用的自动确定，将专门开发软件。机器学习模型(神经网络)将使用视频材料进行训练，然后将用于指定动物的活动。为了分析这一详细信息，将应用动态网络分析。羊群结构(奶牛对应于节点，交互对应于边缘)或资源的使用将被建模为网络。将计算对节点进行排名的参数和关于完整网络的参数，以揭示个性和羊群结构之间的关系。通过收集的数据(日常活动、网络分析)，可以描述单个动物的重复行为或反复出现的相互作用星座等模式。本研究项目的另一个目的是分析这些模式对外部事件的影响(例如，喂养的变化)以及人格维度之间的关系(例如，攻击性-优势或恐惧-活动)。模式识别方法将被应用于预测模式的当代发展。通过这种方式，与社会结构的实际发生状态进行比较，将能够分析外部事件的影响。此外，比较奶牛确定的参数的个体模式可以为人格维度组合的出现提供解释方法。这种对动物个性的分析是基于图像处理、网络分析和模式识别，并处理各种类型的信息。可以预期的是，关于如何在测试情况下评估个性以及如何将个性作为管理决策和改善动物福利的资产的重要发现。

The individual needs and behaviours of farm animals affect the animals’ health and productivity and should be considered in an animal-friendly husbandry. The aim of this Precision Livestock Farming project is to analyse the different personalities of conventionally group housed dairy cows and to understand, in which manner the interaction of personalities affects the herd structure. Long-term videos will be recorded and an automation of the observation of everyday behaviour (e.g. feeding, use of cow brush, used paths) as well as of the determination of social interactions between the cows is aspired. As data basis for the conventional evaluation of personalities of individual animals, well established temperament tests will be conducted concerning all cows in the analysed herd. Then relations between personality traits and everyday activities will be explained. To achieve an automated determination of behaviour and animal interactions that is independent from a human observer, software will be specifically developed. Machine Learning models (Neural Networks) will be trained using the video material and will afterwards be used to specify the animals’ activities. For the analysis of this detailed information, dynamic network analysis will be applied. Herd structures (cows correspond to nodes, interactions correspond to edges) or the use of resources will be modelled as networks. Parameters to rank the nodes and parameters regarding the complete network will be calculated to reveal relationships between personalities and herd structure. Patterns like repeating behaviours of individual animals or recurring constellations of interactions can be described by means of the gathered data (everyday activity, network analysis). An analysis of these patterns with regard to the effects of external events (e.g. changes in feeding) and to relations between personality dimensions (e.g. aggression – dominance or fear - activity) is another aim of this research project. Methods of pattern recognition will be applied to predict contemporary developments of the patterns. In this way a comparison with the actually occurred state of the social structure will enable an analysis of the effects of external events. Furthermore, to compare individual patterns of the determined parameters of the cows can supply explanatory approaches for the emergence of combinations of personality dimensions. This analysis of animal personalities is based on image processing, network analysis and pattern recognition and dealing with various types of information. It is to be expected that important findings on how the personality can be assessed outside of test situations and how personality can be an asset for management decisions and improvement of animal welfare.