Periodic Reporting for period 1 - AutoPlayPig (Automatic detection of play behaviour in young pigs as a measure of positive affective states.)
Période du rapport: 2020-01-01 au 2021-12-31
The welfare of livestock animals is of high concern, not only to the farmer but also to the consumer and general public. It has in recent years become clear that optimal animal well-being is not merely the removal of negative experiences, but also the presence of positive experiences, meaning that even production animals should have a live worth living. Play behaviour in pigs could be an indicator of both the lack of negative and the presence of positive experiences. However, it is not possible for farmers, consultants or inspectors to observe play behaviour in young pigs as it occurs spontaneous and is very short lasting when it occurs. Thus, to use play behaviour for welfare assessment in pigs, the observations need to be performed automatically. One objective of project AutoPlayPig is to take the first steps in developing an algorithm for automatic detection of play behaviour in young pigs. More research is also needed to elucidate whether play behaviour in young pigs can be considered a measure of animal well-being. Thus, a second objective of project AutoPlayPig is to further validate play behaviour as a welfare indicator in young pigs by investigating the level of play behaviour of the individual pig, how it relates to the presence and absence of welfare threats such as weaning, ear damage and diarrhoea and how it is related to factors indicating a better thrive such as higher growth.
The project was successful in providing a proof of concept that locomotor play can be observed automatically from video, while also achieving its aim of further validating play behaviour as a welfare indicator in pigs, by providing evidence that play behaviour in young pigs is performed less in the presence of the investigated welfare threats, while being performed more in pigs that show indicators of better well-being.
The project was successful in providing a proof of concept that locomotor play can be observed automatically from video, while also achieving its aim of further validating play behaviour as a welfare indicator in pigs, by providing evidence that play behaviour in young pigs is performed less in the presence of the investigated welfare threats, while being performed more in pigs that show indicators of better well-being.
Data included in project AutoPlayPig were collected by Department of Animal Science, Aarhus University, Denmark. Data included mainly video recordings of pigs across weaning, but also clinical scorings of the pigs including saliva cortisol, ear damage, diarrhoea and the weight of the pigs on the day of weaning and two days after weaning. From the video recordings, play was observed on the day before weaning and two days after weaning during the hours without disturbance from 14:00 to 22:00. Interestingly, pigs were found to play the most during the evening hours, which are hours that have not previously been included in research on play behaviour in pigs. Besides play behaviour, pig tail posture and eating and drinking behaviour were also observed from the video on the day prior to weaning and the day after weaning.
The following is considered preliminary results of the project, as the work has not yet been peer-reviewed.
As expected and shown in previous studies, play behaviour decreased across weaning. Before weaning, higher performance of play was associated with higher growth rate, absence of liquid faeces (indicative of diarrhoea), and higher frequency of visits to the drinker. During the first two days after weaning, more play was observed in pigs with higher growth, absence of liquid faeces, absence of ear damage, and higher frequency of visits to the feeder. No relations between play behaviour and saliva cortisol, tail posture and tail motion were observed. All in all, these relations show that play behaviour decrease or completely disappear when pigs experience welfare threats and that pigs that indicate to thrive better also play more.
The video recordings were also used in a first attempt on automatic detection of locomotor play in pigs. The focus was to use methods requiring relatively low computational power to better ensure implementation in the future. However, it was also tested how well locomotor play could be detected from video by using deep learning techniques, to compare the performance and to combine the two methodologies. As a first attempt, the focus was only on locomotor play, only on one pig playing and only whether this could be classified from pigs showing low activity, pigs walking and pigs showing high activity. The best performance was achieved when combining the two methodologies with 77% of the locomotor play behaviour events detected when tested on unseen data. Thus, the algorithm still needs further improvement, but the project results provide a proof of concept that locomotor play can be detected from video.
The above results are planned to be disseminated on the EC-PLF and the EAAP conference in Summer 2022, and parts have already been presented at the WALF conference 2021. The results will also be published in open-access international peer-reviewed journals. National publication agencies will also be targeted for result dissemination to a wider audience when the results have been peer-reviewed.
Besides the above work, which represent the scientific core of project AutoPlayPig, the project also had focus on training of the principal investigator within the Precision Livestock Farming research field, including the writing of a systematic review and book chapters, applying for research funding, collaboration with industry, project management, teaching, supervision of master and PhD students, organisation of workshops, specific courses and learning a new programming language.
The following is considered preliminary results of the project, as the work has not yet been peer-reviewed.
As expected and shown in previous studies, play behaviour decreased across weaning. Before weaning, higher performance of play was associated with higher growth rate, absence of liquid faeces (indicative of diarrhoea), and higher frequency of visits to the drinker. During the first two days after weaning, more play was observed in pigs with higher growth, absence of liquid faeces, absence of ear damage, and higher frequency of visits to the feeder. No relations between play behaviour and saliva cortisol, tail posture and tail motion were observed. All in all, these relations show that play behaviour decrease or completely disappear when pigs experience welfare threats and that pigs that indicate to thrive better also play more.
The video recordings were also used in a first attempt on automatic detection of locomotor play in pigs. The focus was to use methods requiring relatively low computational power to better ensure implementation in the future. However, it was also tested how well locomotor play could be detected from video by using deep learning techniques, to compare the performance and to combine the two methodologies. As a first attempt, the focus was only on locomotor play, only on one pig playing and only whether this could be classified from pigs showing low activity, pigs walking and pigs showing high activity. The best performance was achieved when combining the two methodologies with 77% of the locomotor play behaviour events detected when tested on unseen data. Thus, the algorithm still needs further improvement, but the project results provide a proof of concept that locomotor play can be detected from video.
The above results are planned to be disseminated on the EC-PLF and the EAAP conference in Summer 2022, and parts have already been presented at the WALF conference 2021. The results will also be published in open-access international peer-reviewed journals. National publication agencies will also be targeted for result dissemination to a wider audience when the results have been peer-reviewed.
Besides the above work, which represent the scientific core of project AutoPlayPig, the project also had focus on training of the principal investigator within the Precision Livestock Farming research field, including the writing of a systematic review and book chapters, applying for research funding, collaboration with industry, project management, teaching, supervision of master and PhD students, organisation of workshops, specific courses and learning a new programming language.
Project AutoPlayPig presents the first attempt on automatic detection of play behaviour in young pigs and provides a proof of concept that can feed into future projects focusing on automatic detection of positive behaviours in pigs. The results motivate further development of such an algorithm by providing further evidence to support play behaviour as a welfare indicator in pigs. A functioning algorithm could ensure that play behaviour will be included as a welfare assessment measure in the upcoming EU Welfare Label within the pig production. This could put a positive welfare stamp on the production which could maintain or even increase consumers acceptance of certain parts of the production as it allows them to make more informed decisions about their feed intake. Automatic play behaviour monitoring could also make it
possible to evaluate new green housing systems from a positive welfare perspective to ensure that such systems do not pose welfare threats in themselves. Furthermore, a functioning algorithm could feed into future research projects on play behaviour in pigs for better understanding of the behaviour across genetics, housing and management systems, climates and across the production period.
possible to evaluate new green housing systems from a positive welfare perspective to ensure that such systems do not pose welfare threats in themselves. Furthermore, a functioning algorithm could feed into future research projects on play behaviour in pigs for better understanding of the behaviour across genetics, housing and management systems, climates and across the production period.