Computer Vision and Machine Learning Applications for Dairy Farming
Author | : Rafael Ehrich Pontes Ferreira |
Publisher | : |
Total Pages | : 0 |
Release | : 2024 |
ISBN-10 | : OCLC:1450507953 |
ISBN-13 | : |
Rating | : 4/5 (53 Downloads) |
Book excerpt: With recent advancements in precision livestock farming (PLF) and machine learning (ML) techniques, computer vision systems (CVS) have gained popularity as powerful tools for individual animal monitoring. These systems can capture phenotypes from multiple animals simultaneously in an automated and non-intrusive manner. Individual animal identification is crucial for matching animals with their predicted phenotypes, which can be achieved through external identification systems or computer vision-based animal identification algorithms. While previous studies have focused on using computer vision techniques for identifying dairy cows based on unique coat color patterns, these methods are limited to specific breeds that present such patterns. Furthermore, there is a lack of research on the long-term applicability of these methods, considering visual changes due to growth or physiological states. Chapter 1 discusses current applications of computer vision for animal identification, while Chapter 2 explores methods using 3-dimensional representations of the dorsal surface of dairy calves for identification without relying on coat color patterns. These methods are evaluated on calves during their growth stage, accounting for changes in body shape and size. In Chapter 3, the potential of pseudo-labeling is assessed for improving the performance of neural networks for animal identification. The results show promising performance with a fraction of annotated data compared to traditional methods. Chapters 4 and 5 focus on developing machine learning pipelines for phenotype prediction, specifically early detection of postpartum subclinical ketosis (SCK) using prepartum data exclusively. Various techniques are explored for extracting features from image, text, genotype, and cow behavior and historical data. Data fusion techniques are explored to integrate those features into the machine learning pipelines, and a cloud computing-based framework is proposed to automate data processing, feature extraction, and phenotype prediction. Overall, this dissertation highlights the potential of machine learning and computer vision in guiding data-driven management decisions in dairy farming. By automating processes and integrating data from multiple sources and modalities, these techniques offer opportunities for improving farm profitability, productivity, and animal welfare, particularly through individual animal monitoring and early detection of health issues.