The use of computer vision techniques as noninvasive tools to monitor parameters related to the well-being and productive performance of cattle and pigs

Abstract

Among a wide range of factors that can affect animals’ wellbeing, stress levels and health status have been identified as relevant factors in production animals. The increasing awareness about animal welfare and the impact that it has on farm productivity has been promoting scientific research and the development of novel and less invasive methods to monitor animals and obtain measurements that can be used as indicators to assess animal wellbeing, including stress and health status. Research presented in this thesis deals with red-green-blue (RGB) and thermal infrared (TIR) imagery, and computer vision techniques, as non-invasive tools to measure physiological changes in cattle and pigs to assist in the assessment of their stress and health.

One study was conducted to evaluate the performance of the proposed methods, which were used to analyse RGB and TIR imagery to measure eye-temperature, heart rate (HR) and respiration rate (RR) in cattle. The study was performed in a robotic dairy farm, where TIR and RGB cameras recorded ten dairy cows during six handling procedures, across two consecutive days. Simultaneously, core body temperature, HR and RR were measured using standard methods for comparison with the data obtained from the recorded images using the developed algorithms. A feature tracking algorithm was developed to facilitate the processing of RGB videos, which showed an accuracy between 92% and 95% depending on the area analysed. From the physiological parameters analysed, the highest correlations were observed between eye-temperature and intravaginal temperature (r = 0.8; P<0.01), and between remote RR and the RR obtained from visual observations (r = 0.87; P<0.01).

A further two studies were carried out to implement the proposed computer-based methods to remotely measure eye-temperature, heart rate and respiration rate of cattle, and to investigate whether these measures could be used to evaluate the physiological response of cattle to stressful situations and whether they could be used as predictors of beef quality. For these two studies, 215 beef cattle were recorded with RGB and TIR cameras on the farm and at the abattoir to obtain eye-temperature, HR, and RR measurements. Cuts of the respective beef were evaluated by consumers, and the ultimate pH (pHu) and meat colour were obtained from the respective carcasses. It was observed that the physiological variables of cattle were higher at the abattoir compared to the farm. Moreover, eye-temperature obtained on farm and at the abattoir were highly correlated. However, the results of these studies indicated that these measurements had low contribution when predicting beef quality.

Finally, two studies were performed to investigate the use of the proposed computer vision methods with RGB and TIR imagery to monitor pigs and detect changes in eye-temperature, ear-base temperature, HR, and RR. The objective was to identify whether these physiological changes could assist in the early detection of pigs affected by respiratory disease. The first of these studies was performed under experimental conditions, where pigs were challenged with Actinobacillus pleuropneumoniae (APP). Images (TIR and RGB) were recorded after this challenge, and the data obtained was then compared between the group of pigs that presented clinical signs of respiratory disease and the group of pigs that were considered healthy. Clear changes in temperature and HR were observed six or more hours before the clinical observations identified sick animals. Conversely, changes in RR were detected in the last period of observations, around the time when clinical signs started to be present.

The second of these experiments used different cameras and included improvements to the proposed methods to monitor pigs constantly and in a commercial setting. A total of 48 pigs were monitored between 9 and 20 weeks of age. Eye-temperature, HR and RR measurements were compared between the pigs that were identified as sick and those that were considered healthy. Similarly to what was observed in the previous study, changes in these parameters were identified before the clinical observations indicated signs of illness (up to 2 days before), where the earliest changes were observed in eye-temperature and HR, and the latest changes were observed in RR.

This thesis provides evidence that computer vision techniques may be suitable as a non-invasive method for monitoring farm animals. Therefore, it prompts further investigation via controlled studies to continue the development and automatization of these techniques, leading to the improvement of science-based industry-relevant monitoring systems.