The words ‘Artificial Intelligence’, or AI, conjure images of sci-fi films, self-driving cars and talking robots. We tend to think of AI as a futuristic concept, somewhat out of our reach, but in reality it is already impacting the everyday goings on of lives. Google maps can predict the fastest route for your journey, you’re presented with suggestions of what to buy online, and Facebook can automatically recognise images of your friends, all using AI. Artificial intelligence is now making its way into our farming practices and helping us to improve animal welfare. It may not seem like an obvious application of advanced technology, but, when you dig a little deeper, it’s easy to see that AI and animal welfare are a well-suited pairing, especially in the world of pig farming.
Pig farming produces over one billion pigs globally each year, producing over 110 million tons of meat (USDA, 2017) and this number is due to grow as per person meat consumption rises. Keeping track of the health and welfare of such large numbers of animals is a huge task and is often carried out by farm staff and vets, which is time consuming and labour intensive. Over recent years, this problem has been worsened by a decreasing number of farmers and increasing herd sizes. One way to tackle the problem is by using modern technology to reduce the workload of the farmer. This idea has led to the rise of a new type of farming called Precision Livestock Farming. Precision Livestock Farming is a way of managing livestock by continuous, automated real-time monitoring of health and welfare, production, reproduction and environment. These technologies do not aim to replace the role of the farmer, but they can help to remove the human error aspect of monitoring and can monitor larger numbers of animals in a quicker time.
One way to monitor pigs which is affordable and efficient is to install cameras and look for changes in their behaviour. One branch of the PigSustain project does exactly that; we place cameras in the pig sheds and film the pigs 24/7. We aim to use a machine learning approach to automatically detect and track each pig in a pen without having to physically mark or tag them. This means that we teach the computer what a pig looks like, so that it learns how to detect them on its own. Imagine not knowing what a dog looked like and being shown only photographs of German Shepherds. If someone then showed you a Poodle or a Chihuahua you might not immediately recognise it as a dog, but you would learn from experience, which is exactly what our computer algorithm does. The more we teach and the more examples we feed to it, the better the computer becomes. Our initial results are now published here in the journal Sensors and our detection precision and tracking accuracy are at 97.44% and 95.2%, respectively.
The overall aim of the technology is to measure behaviour from individual pigs and to detect when that behaviour changes in some way, for example, if the pig is eating less or moving more slowly than usual. These types of subtle behavioural change can help in pointing the farmer in the direction of a pig which may need a little more attention. Our ultimate goal is to use the technology to predict when a disease or health concern might occur, which would help the farmer in the management of his herd and improve pig welfare.
The trajectories of 9 pigs tracked individually in a group pen, without the need to mark or tag.