In close relation to production, physiology and health, behaviour is an important indicator to evaluate animal welfare. It can also provide feedback on the quality of the animal's environment. The aim of this paper is to propose an automatic system for continuously measuring activity levels of pigs by using camera images. It was also tested whether dynamic responses (i.e. steps) in the activity level can be linked to different behaviour types. Reference data were collected by human experts who labelled the behaviour of the pigs in the video recordings, in parallel to the automatic activity monitoring system. For the experiments two adjacent pig pens in a commercial pig house were used (5.9×2.6 m), each populated by about 15 pigs. An infrared-sensitive CCD camera was mounted 5 meters above the floor of the pen. Images were captured with a resolution of 768×586 pixels at 1 Hz frame rate. Software was developed to measure the activity level of animals from the camera image. Four zones were defined in the image, each covering half a pen. Every second, the algorithm logged the camera image and the activity index for each zone. This activity index was calculated as the fraction of the floor space in the pen zone that contained motion in between two subsequent camera images. The behaviour of the animals in each zone of every recorded video image was visually labelled as one out of five possible behaviour scores: 'no activity', 'fighting', 'biting', 'nuzzling' and 'eating'. For each of the behaviour types except the 'no activity' behaviour, the response of the activity index to the occurrence of the behaviour was modelled as the average and the step of the activity index within a time window between 30 seconds before and 30 seconds after the start of the behaviour. A Naïve Bayesian Classifier was trained for the behaviour types 'fighting', 'biting', nuzzling' and 'eating' using the mean activity value and the size of the activity step as features, calculated from the activity level and labels for day 1. Classification was tested by automatically classifying the features calculated for day 2 and comparing with the corresponding manual labelling. Results showed a correct classification rate of 73.3% for 'fighting', 60% for 'eating', 46.2% for 'biting' and 24% for nuzzling. However, the behaviours 'fighting', 'biting' and 'nuzzling' were often misclassified as each other. The general conclusion is that the global activity level of the group can only be used to recognise behaviour types that are performed simultaneously by the majority of the animals, such as 'eating'. It proves to be too crude to be an indicator of individual animal behaviour.

Real-time monitoring of pig activity and behaviour recognition / T. Leroy, F. Borgonovo, A. Costa, J.M. Aerts, M. Guarino, D. Berckmans - In: Precision Livestock Farming '09 / [a cura di] C. Lokhorst, P.W.G. Groot Koerkamp. - Wageningen : Wageningen Academic publisher, 2009. - ISBN 978-90-8686-112-5. - pp. 275-282 (( convegno Joint International Agricultural Conference tenutosi a Wageningen nel 2009.

Real-time monitoring of pig activity and behaviour recognition

F. Borgonovo;A. Costa;M. Guarino;
2009

Abstract

In close relation to production, physiology and health, behaviour is an important indicator to evaluate animal welfare. It can also provide feedback on the quality of the animal's environment. The aim of this paper is to propose an automatic system for continuously measuring activity levels of pigs by using camera images. It was also tested whether dynamic responses (i.e. steps) in the activity level can be linked to different behaviour types. Reference data were collected by human experts who labelled the behaviour of the pigs in the video recordings, in parallel to the automatic activity monitoring system. For the experiments two adjacent pig pens in a commercial pig house were used (5.9×2.6 m), each populated by about 15 pigs. An infrared-sensitive CCD camera was mounted 5 meters above the floor of the pen. Images were captured with a resolution of 768×586 pixels at 1 Hz frame rate. Software was developed to measure the activity level of animals from the camera image. Four zones were defined in the image, each covering half a pen. Every second, the algorithm logged the camera image and the activity index for each zone. This activity index was calculated as the fraction of the floor space in the pen zone that contained motion in between two subsequent camera images. The behaviour of the animals in each zone of every recorded video image was visually labelled as one out of five possible behaviour scores: 'no activity', 'fighting', 'biting', 'nuzzling' and 'eating'. For each of the behaviour types except the 'no activity' behaviour, the response of the activity index to the occurrence of the behaviour was modelled as the average and the step of the activity index within a time window between 30 seconds before and 30 seconds after the start of the behaviour. A Naïve Bayesian Classifier was trained for the behaviour types 'fighting', 'biting', nuzzling' and 'eating' using the mean activity value and the size of the activity step as features, calculated from the activity level and labels for day 1. Classification was tested by automatically classifying the features calculated for day 2 and comparing with the corresponding manual labelling. Results showed a correct classification rate of 73.3% for 'fighting', 60% for 'eating', 46.2% for 'biting' and 24% for nuzzling. However, the behaviours 'fighting', 'biting' and 'nuzzling' were often misclassified as each other. The general conclusion is that the global activity level of the group can only be used to recognise behaviour types that are performed simultaneously by the majority of the animals, such as 'eating'. It proves to be too crude to be an indicator of individual animal behaviour.
Animal environment; Animal welfare; Automatic activity measuring; Behaviour
Settore AGR/10 - Costruzioni Rurali e Territorio Agroforestale
European conference on precision agriculture)
ECPA
European Conference on Precision Livestock Farming
ECPLF
European Federation for Information Technology in Agriculture, Food and the Environment
EFITA
Book Part (author)
File in questo prodotto:
Non ci sono file associati a questo prodotto.
Pubblicazioni consigliate

Caricamento pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/2434/68622
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
social impact