Graduate Studies
First Advisor
Tami Brown-Brandl
Date of this Version
2-2024
Document Type
Article
Citation
A thesis presented to the faculty of the Graduate College at the University of Nebraska in partial fulfillment of requirements for the degree of Master of Science
Major: Major: Agricultural and Biological Systems Engineering
Under the supervision of Professor Tami Brown Brandl
Lincoln, Nebraska, February 2024
Abstract
With increasing urbanization, more people are switching from eating mainly grains and root vegetables to consuming more meat and dairy. This transition has transformed animal farming from small-scale operations to massive commercial production industries that confine animals in artificial settings. The pork industry is no exception to this trend. Sows, which are pregnant pigs, are typically relocated to farrowing crates a few days before delivery and remain confined for four weeks after giving birth. This practice enables producers to closely monitor the health of both the mother and her piglets to provide individualized care that is challenged in group-housed pen settings. While this confinement reduces the preweaning mortality rate among newborn piglets by protecting them from accidental crushing under the mother, it restricts the natural movements of the sow. This restriction can lead to injuries such as shoulder lesions, which are classified as tissue damage on the top of the shoulder. The timely application of zinc oxide, chlortetracycline spray, or a change in bedding to rubber mats can help treat these lesions and prevent them from progressing to severe bone-deep ulcers that may negatively impact the carcass value at the time of sale.
The goals of this study were:
- To evaluate the use of RGB cameras to detect lesions and estimate their size from a top-down perspective using computer vision algorithms.
- To analyze the sow’s lying behavior, various physical and environmental factors, and their impact on lesion size and occurrence.
The findings showed that YOLOv5 performed better at lesion localization with an AP@0.5 score of 0.92 compared to other models. Both traditional binarization techniques and deep learning-based methods were explored for pixel segmentation and found OTSU's binarization was the most effective method with a dice coefficient of 0.82. Furthermore, various physical and environmental factors and their impact on lesion size and occurrence. The study evaluated three distinct farrowing stall designs. Each design was equipped with either one or two heat lamps, using a 2 x 3 block design. Each stall underwent six treatment scenarios across seven farrowing cycles. The results showed a significant interaction between crate size and the number of heat lamps indicating that the smallest crate with two heat lamps resulted in the highest incidence of shoulder lesions Also, sows with higher parity and low weights showed high lesion occurrence. But no significant impact of these parameters was found on lesion severity and size.
Overall, this work shed light on the potential of computer vision to improve the welfare of sows in commercial pork production. It shows the importance of optimizing confinement conditions to minimize shoulder injuries and improve overall animal well-being.
Advisor: Tami Brown-Brandl
Comments
Copyright 2024, Shubham Bery. Used by permission