• Reduce text

    Reduce text
  • Restore text size

    Restore text size
  • Increase the text

    Increase the text
  • Print


Artificial intelligence for the welfare of pigs

Researchers from the Joint Research Unit for the Physiology, Environment and Genetics of Animals and Livestock Systems (PEGASE), in collaboration with the CEA-LETI, have developed an automated detection device for pig behaviour. It is a system based on individual monitoring to help farmers intervene quickly in the event of behaviour which affects the health and welfare of the animals. This work is carried out as part of the European programme PigWatch.

Finishing pigs on straw litter. © INRA, MAITRE Christophe
By Communication Department INRA Bretagne-Normandie
Updated on 03/27/2019
Published on 02/28/2019

For many years now, one of the major issues in pig farming has been reducing injurious behaviours, such as fighting or tail-biting. By detecting these kinds of behaviour or, if possible, early warning signals of them, farmers can intervene very early on to reduce welfare problems. “As part of the PigWatch European project, we have developed an automated method to monitor pig behaviour”, explains Armelle Prunier, researcher at PEGASE. Monitoring the behaviour of finishing pigs will have multiple applications. At the scientific level, it will help analyse the factors that determine behaviour. This will open possibilities for breeding calmer animal breeds. On farms, it will help detect problems rapidly before they get out of hand. “Abnormal agitation may indicate imminent injurious behaviours. On the contrary, a low level of activity may indicate that an animal is sick” points out Armelle Prunier.

An algorithm which analyses behaviour

In order to obtain automated techniques to monitor behaviour, INRA and the CEA developed sensors, inserted in ear tags, which contain a wireless accelerometer and observed pigs by video cameras. The data registered are analysed by artificial intelligence algorithms. To write these algorithms, the activities of 12 pigs were registered and observed at regular intervals over 2 months. Their behaviour was identified and the signals from the sensors were correlated with these behavioural observations. "This step is crucial", stresses Armelle Prunier. "It makes it possible to link a signal to a given behaviour and the algorithm can learn to recognise and categorise the signals".

On a farm where animals are monitored, data are transmitted to a program which analyses them in real time. If two accelerometers go crazy at the same time, this no doubts indicates aggressive behaviour or a fight. The algorithm is still being developed but the results are encouraging since the software correctly identifies about 50% of fighting behaviours. Once finely tuned, this tool will enable the farmer, warned by SMS, to act rapidly. “We are going to work on the events preceding cannibalism episodes, in order to determine early warning signals", plans Armelle Prunier, "which will be a precious help so that farmers can focus on animals at risk by adapting their practices”.

Find out more

The European project PigWatch

Project website: https://pigwatch.net/