Towards an automated assessment of pig behaviors on farm

Etienne Labyt (1) , G. Lagarrigues (2) , O. Sakri (1) , Christophe Godin (1) , Céline Tallet (3) , Armelle Prunier (3)

1 CEA-LETI - Commissariat à l'énergie atomique et aux énergies alternatives - Laboratoire d'Electronique et de Technologie de l'Information
2 Probayes [Montbonnot]
3 PEGASE - Physiologie, Environnement et Génétique pour l'Animal et les Systèmes d'Elevage [Rennes]

Tail biting and aggression in finishing pigs are injurious behaviors affecting health and welfare of pigs as well as productivity of the farms. In the PIGWATCH European project (ERANET Anihwa), INRA and CEA are working on development of an automated technique, based on the use of sensors and machine learning algorithms, to detect injurious behaviors or abnormal patterns of activity. A wireless ear tag was developed, including a triaxial accelerometer and an Android application for data recording, processing and alert sending to the farmer on his smartphone when injurious behaviors are detected. Pigs were housed in groups of 8 on solid floor covered daily with fresh straw. Twelve pigs, i.e. 4 per group, were fitted with these ear tags. Their activity was recorded with the sensors during a period of two months. Their behavior was analyzed using video records on selected days. They were subjected to straw deprivation followed by food restriction in order to stimulate injurious behaviors or changes in the behavioral pattern of activity. In a first step, 24 hours of video records were analyzed and synchronized with signals from sensors for each pig. Relevant mathematical features were extracted from signals to predict various pig’s behaviors and notably, discriminate injurious behaviors from normal activity. These features were used in machine learning algorithms to build a model, able to automatically predict pig’s behaviors and detect injurious ones. Regarding “marked” fights (> 3 aggressive acts within 10 s), the model has a sensitivity of 42% and a specificity of 62%. This model has been implemented in an Android App and will be assessed in farms in Germany, notably in terms of true and false alerts. We will get the feedback from farmers on the usefulness and how to improve the system ergonomic. As a third step, the whole database collected at INRA is currently processed with this model to predict other pig’s behaviors (e.g. resting, feeding) and assess individual and nycthemeral variations.

Mots clés: automatic detection; behaviour; pig; swine

Mots clés: détection automatique; modèle de prédiction; comportement animal; porc

Domaines: Sciences du Vivant [q-bio]

Etienne Labyt, G. Lagarrigues, O. Sakri, Christophe Godin, Céline Tallet, et al.. Towards an automated assessment of pig behaviors on farm. 69. Annual Meeting of the European Federation of Animal Science (EAAP), Aug 2018, Dubrovnik, Croatia. Wageningen Academic Publishers, Annual Meeting of the European Association for Animal Production, 24, 2018, Annual Meeting of the European Association for Animal Production. ⟨hal-01870571⟩