Files
Abstract
As the global population approaches nearly 10 billion by mid-century, agricultural production faces unprecedented challenges. Despite advancements in crop breeding, the rate of improvement has proven insufficient to meet the growing demand for food and fiber, necessitating a significant increase in agricultural productivity. Traditional phenotyping methods are labor-intensive and inefficient, creating an urgent need for high-throughput approaches. In-field digital crop phenotyping, empowered by advanced sensing technologies and innovative data discovery methods, has emerged as a promising solution to enhance throughput and accelerate breeding advancements for more productive crops. This work aimed to automate unmanned aerial vehicle (UAV)-based phenotyping pipelines and optimize terrestrial laser scanning (TLS) methodologies for in-field phenotyping through the use of advanced data analytics and the integration of unmanned ground vehicles (UGVs). UAV-based imagery, combined with machine learning algorithms and the Segment Anything foundation model for image segmentation (SAM), enabled precise estimation of yield-related traits in cotton and peanut breeding trials, including cotton boll number, canopy height, growth habit, and main stem prominence, facilitating accurate pre-harvest yield predictions and in-field phenotypic trait assessments.
Additionally, novel proximal sensing methods based on TLS with a customized autonomous Husky ground mobile robot were developed to automate field phenotyping and improve in-field crop measurements. A new methodology for spatiotemporal registration of 3D point cloud data obtained through multi-scan TLS enabled continuous monitoring of cotton crop growth dynamics throughout the growing season. Furthermore, an autonomous ground robotic system equipped with a 3D laser scanner demonstrated efficient field navigation and data collection capabilities across different field layouts.
By introducing these innovative and optimized methodologies, this research addresses critical challenges in agricultural research and crop breeding. Leveraging advanced remote and proximal sensing technologies with data analytics, this work significantly contributes to the development of digital agriculture and plant phenomics, offering valuable insights and practical solutions for researchers and practitioners interested in advancing sustainable agricultural practices and enhancing crop phenotypes.