Validation of non-invasive sensor technologies to measure interaction with enrichment material in weaned fattening pigs

• Published in: Applied animal behaviour science Vol. 263; p. 105923
• Main Authors: Veldkamp, Fleur, Garcia-Faria, Tomas Izquierdo, Witjes, Vivian L., Rebel, Johanna M.J., Jong, Ingrid C. de
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.06.2023
• Subjects: Accelerometer; Algorithm; animal behavior; Behaviour; body temperature; body weight; Enrichment; farms; neural networks; nose; Passive infra-red; Pigs; swine; video cameras; wood; Pigs; Enrichment; Accelerometer; Passive infra-red; Behaviour; Algorithm
• ISSN: 0168-1591; 1872-9045; 1872-9045
• DOI: 10.1016/j.applanim.2023.105923

Measuring animal behaviour is important in the assessment of animal welfare. When interaction with the enrichment material (EM) can be measured, it can be used for detecting an increasing/decreasing interest in a certain EM. In this study, non-invasive sensor technologies were validated for measuring interaction with EM in pens with weaned fattening pigs. The experiment was carried out in four pens with six weaned pigs per pen (until a body weight of ± 25 kg) at a semi-commercial farm. Pens were provided with EM (ball and piece of wood (and a rope in two of the four pens) connected to a chain). Different sensor technologies were tested: passive infra-red detectors (PIDs), tri-axial accelerometers (TAA) and neural network model algorithms (NNMA 1 and NNMA 2) based on video images. Per pen, a PID was placed above the EM which measured the movement of body heat around the chain (⌀20 cm) in volts per second. A TAA was attached to the EM (at the top of the chain) and measured acceleration based on X-, Y- and Z-axis co-ordinates every second. A video camera was placed above each pen to record video images that were used to feed the NNMAs and for behavioural observations. Interaction with EM (shake, carry, nose, bite, chew or root) was manually scored per second per pig (pooled per pen afterwards) for 30 min of video footage per pen per week and was compared with data from PIDs, TAAs and NNMAs. F1 score (F1) and Matthews Correlation Coefficient (MCC) were calculated to measure the performance of the sensor technologies. PIDs (F1 = 0.380, MCC = 0.192), as well as TAAs (X-axis: F1 = 0.482, MCC = 0.345; Y-axis: F1 = 0.524, MCC = 0.401; Z-axis: F1 = 0.465, MCC = 0.320; XYZ-axis: F1 = 0.474, MCC = 0.333), overestimated interaction with the EM which might be due to the relatively small pen size, resulting in piglets touching the EM without intentional interaction with the EM. NNMAs achieved the highest performance parameters (NNMA 1: F1 = 0.554, MCC = 0.466; NNMA 2: F1 = 0.540, MCC = 0.445). Overall, only moderate F1s and MCCs were reached. The results indicated that the individual sensor technologies are not yet appropriate to measure interaction with the EM. However, there is potential to measure interaction with EM by applying a multi-sensor approach (combination of PID, TAA and NNMA), but this merits further study. •Monitoring animal behaviour is important in the assessment of animal welfare.•Sensor technologies were validated to measure interaction with enrichment material.•Algorithms, Passive Infra-red Detectors and accelerometers were used.•Individual sensor technologies are not yet appropriate to measure interaction.•Potential to combine sensor technologies for measuring interaction.