Zhao, Kai2024-04-092024-04-092024https://hdl.handle.net/2097/44189The last decade has witnessed a rapid expansion of computer vision and scientific computing, particularly within industrial robotics, automotive technologies, and healthcare. Concurrently, solid material modeling and computer visualization in agriculture have emerged as useful tools to solve complex problems. In the domain of modern seed breeding programs, the efficient and accurate analysis of seeds is paramount for successful outcomes. Phenotypes, encompassing morphological parameters of agricultural entities like seeds, roots, and leaves, are crucial to predicting their behavior under varying environmental conditions. Seed volume is particularly significant among these phenotypes, but even this fundamental metric, seed volume, proves challenging to calculate efficiently using conventional equipment. While volume estimation is a thoroughly studied problem for regular objects, the same principles do not seamlessly extend to irregularly shaped objects like seeds. The irregular shapes of seeds further complicate the accurate estimation of seed volume while ensuring its preservation.    This dissertation delves mainly into applying computer vision and scientific computing within the agricultural domain, explicitly focusing on low-throughput phenotyping (LTP). The primary emphasis lies in simulating, visualizing, and accurately calculating the volume of individual seeds. In this context, the dissertation proposes a comprehensive system comprising hardware and software components to address this challenge.    This dissertation presents several noteworthy contributions. The primary contribution involves proposing an innovative modeling framework designed to capture seed images effectively. Compared to previous frameworks, this new framework supports various backgrounds and enhances the ability to efficiently capture accurate images by rotating the seed at a constant rate and parsing the video to extract images at fixed angles.    The second contribution is extending traditional digital image processing (DIP) methods. A comprehensive analysis and comparison of diverse standard contour detection approaches is conducted using various background color contexts. The traditional space carving algorithm is extended by using a novel, multi-threaded approach to reduce the time required to carve seeds significantly.    Furthermore, this research extends the application of convolutional neural network (CNN)-based methodologies to detect edges in DIP methods. Following a thorough analysis of two prominent CNN methods, a novel, straightforward CNN model is proposed to facilitate contour detection. A detailed analysis of both DIP and CNN-based approaches measures the accuracy and efficiency of each approach and extension.    The dissertation contributes to the field by developing a user-friendly interface tailored for efficient seed analysis by agronomists.    Another direction for computer vision algorithms is at the field level in estimating fractional vegetation cover using Hough lines and linear iterative clustering.    Finally, model checking is used to verify the correctness of computer vision algorithms and other distributed algorithms, such as those to ensure mutual inclusion and mutual exclusion.en-US© the author. This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).http://rightsstatements.org/vocab/InC/1.0/3DContour detectionPhenotypingSeed analysisDissertation