VETCARE+: A Deep Learning Application for Early Detection of Lameness in Dairy Cattle

Lameness in dairy cattle indeed appears to be a prevalent and pertinent issue that impacts the welfare of the animals themselves and the economic viability of farms in general. This paper presents a new deep learning method tailored to the task of accomplishing on-time lameness detection in dairy cattle. It does so effectively by combining multiple sensor data with high level behavioural analysis, which forms the crux of nature. We aim to create a state-ofthe-art system that can detect using an integration of a wide range of sources of data, including but not limited to accelerometer, advanced computer vision, automated behavioural pattern recognition and milking robots’ data. We believe the new system should identify it with even more accuracy and at a significantly higher speed than current methods could attain so far. For the ambitious objective followed by this approach, we are using a multi-modal deep neural network as the central component of this proposed system. It leverages a set of several advanced features, including the latest notions of transfer learning and attention mechanisms, which enable it to utilize and make the most of a variety of data types. Initial trials conducted so far have shown that our advanced system is able to predict lameness with an impressive accuracy rate of 92%. At such a level of prediction, it can be done up to ten days ahead of what might have been determined by the traditional, visual methods used so far, without a substantially high false positive rate at merely 3%. These therefore hold much promise and with these expectations, great improvements will be witnessed both in the welfare of the dairy cattle and that of the farms in terms of productivity. This will be through early interventions and effective treatments of cases of lameness that may arise.