Sensor-based electronic monitoring of feeding and drinking activity of nursery pigs in swine farms
Date
2024-12
Authors
Park, Yunsoo
Major Professor
Advisor
Pandey, Santosh
Kim, Jaeyoun
Shearer, Jan
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Daily activity monitoring of farm animals is important to get a general assessment of the health and wellness of the animals. Such activity may include eating and drinking behavior which is often reduced in the occurrence of any illness. Sick animals are more lethargic with prolonged lying-down behavior and reduced eating and drinking activity. In swine farms, farm personnel visually look for pigs with any abnormal behavior (e.g., reduced weight, lethargy, drooping face) who are then separated for closer monitoring or additional treatment. It is often challenging to visually assess a large pig population in a barn, and any missed events of identifying sick pigs could prolong the illness and create worsening health conditions for the animal. As such, there has been a need for precision livestock management technologies that can provide low-cost and high-resolution observations of pig populations for early disease detection.
Under the field of precision livestock management, automated monitoring technologies have been introduced to provide easy, automated, and hassle-free round-the-clock observations of pigs in swine farms. The driving motivation of these technologies is to provide 24/7 monitoring of pig populations and alert the producer for any abnormal behavior or prolonged lethargy that may be a sign of sickness. The most prevalent automated technology in swine farms is computer vision systems that incorporate security cameras to record videos and deep learning models trained to identify specific behavioral signatures suggestive of sickness. Camera systems are low-cost, easily installable, and trainable on several deep-learning datasets available today. However, they require annotation and labeling of large videos and further model validation, which is exceptionally time-consuming and laborious. This has been a limiting factor in the deployment of computer vision systems in swine farms.
In this thesis, we propose an alternative technology solution for swine health monitoring based on electronic sensors and accompanying data pipelines. The sensor board comprises activity sensors connected to a central microprocessor chip. The microprocessor chip collects the sensors’ data at regular intervals and transfers the data to the receiver after prescribed times. The sensor platform was designed, fabricated, tested, and customized for swine farms, particularly to identify the feeding activities of nursery pigs in individual pens in the swine farm. The sensor boards were installed at strategic locations on the feeders in individual pens so as to record any mechanical displacement of the feeder unit components during eating. The sensor boards were tested during a nutritional study conducted at the swine farm where the intent was to assess whether pigs prefer to feed with added sugar or added lactose compared to their normal feed. Both sensors’ data and video recordings were collected for 28 hours of pen monitoring.
The data preprocessing involved cleaning the data, selecting the optimal sliding window size, extracting the key features, and training the different machine learning models to predict feeding events. Labeling the data from video recordings was a significant task that required considerable time and effort. A number of combinations of window size and step size were tested with the models to determine the best accuracy in event prediction. It was found that the Random Forest Classifier gave the best prediction results with reasonable accuracy.
In summary, this thesis provides an alternative sensor-based technology solution to computer vision systems that has been initially tested in swine farms for pen-level feed event monitoring. The sensor platform may appeal to swine producers because of its small form factor, low cost, and off-grid operations, which is beneficial to hog producers in rural and remote locations. The future scope could involve rigorous and prolonged testing over many days on the swine farm and continued discussions with the ground personnel to improve the technology.
Series Number
Journal Issue
Is Version Of
Versions
Series
Academic or Administrative Unit
Type
thesis