A computer vision-based method for spatial-temporal action recognition of tail-biting behaviour in group-housed pigs. (July 2020)

• Record Type: Journal Article
• Authors: Liu, Dong; Oczak, Maciej; Maschat, Kristina; Baumgartner, Johannes; Pletzer, Bernadette; He, Dongjian; Norton, Tomas
• Is Part Of: Biosystems engineering. Volume 195(2020)
• Page Start: 27
• Page End: 41
• Publication Date: 2020-07
• Subjects: Action recognition -- Computer vision -- Pig behaviour -- Precision livestock farming -- Tail biting; Bioengineering -- Periodicals; Agricultural engineering -- Periodicals; Biological systems -- Periodicals; Génie rural -- Périodiques; Systèmes biologiques -- Périodiques; 631
• DOI: 10.1016/j.biosystemseng.2020.04.007
• Languages: English
• ISSNs: 1537-5110

Abstract

Abstract : As a typical harmful social behaviour, tail biting is considered to be a welfare-reducing problem with economic consequences for pig production. Taking a computer-vision based approach, in this study, we have developed a novel method to automatically identify and locate tail-biting interactions in group-housed pigs. The method employs a tracking-by-detection algorithm to simplify the group-level behaviour to pairwise interactions. Then, a convolution neural network (CNN) and a recurrent neural network (RNN) are combined to extract the spatial-temporal features and classify behaviour categories. The performance of the proposed method was evaluated by quantifying the localisation accuracy and behaviour classification accuracy. The results demonstrate that the tracking-by-detection approach is capable of obtaining the trajectories of biters and victims with a localisation accuracy of 92.71%. The spatial-temporal features trained by CNN and RNN are robust and effective with a category accuracy of 96.25%. In total, our proposed method is capable to identify and locate 89.23% of tail-biting behaviour in group-housed pigs. Highlights: A novel method was proposed to recognise and locate pig tail-biting behaviour. A tracking algorithm was proposed to extract pairwise interactions from the group. The CNN + LSTM model was used to recognise interactive actions. The method can locate and identify 89.23% of tail-biting interactions in the group.