Identifying Lameness Movements in Sheep via Sensor Data Analysis

Lameness is one of the most significant issues for sheep well-being in the UK. Climate changes bringing mild winters and wet summers create muddy soil that is a good place for the Dichelobacter nodosus bacteria to survive and transmit easily to the sheep foot and cause footrot; which is one of the common causes for sheep lameness in the UK. According to the report from ADAS (Vickers and Wright, 2013), each lame ewe costs approximately £89.80 because of the decline in its performance alongside the extra labour and treatment costs that are needed. Although lameness is an endemic disease, it could be controlled from being spread within the whole flock. Previously, lameness can be spotted by a trained veterinarian or experienced shepherd either using a visual Locomotion Scoring (LS) or Gait Scoring (GS) system. It was a very time-consuming approach, took many efforts, and tended to be subjective. Therefore, an objective method to monitor the flock via sensor technology has been developed to collect data about the behaviour or gait of the animal to be analysed. This research aimed to utilise sensor devices to detect the early signs of lameness by collecting the movement measurements of the mounted sensor around the sheep’s neck. The collected data were analysed to classify the sheep into the lame and sound classes via machine learning approaches. However, collecting data on behaviour to study the gait changes for the sake of lameness detection in sheep is not a straightforward procedure. Firstly, the data were collected from Lodge Farm, Moulton College in Northamptonshire from several sheep. The data measurements involved movements of three axes around the neck. The y-axis was positioned to correspond with surge movements (forwards and backwards), the x-axis with sway motion (right and left), and the z-axis with heave (up and down) as shown in the figure below. Then, the raw data such as acceleration and orientation were tested by a range of machine learning classifiers. The initial results indicated that decision tree was the best machine learning classifier for the sheep sensor-based data. Moreover, the orientation of the surge axis is the best indicator of early signs of lameness; in this context orientation means the value of the angle around the axis which is the roll angle (y axis-surge). The figure shows the deployment of a mobile sensor which is used as a prototype sensor in this ongoing study.