Files
Abstract
Palmer amaranth (Amaranthus palmeri) remains one of the most economically detrimental weeds in the United States' upland cotton (Gossypium hirsutum) production. Timely and effective management of this aggressive plant is critical to safeguarding yield, preserving fiber quality, and supporting sustainable production. This dissertation explores Palmer amaranth’s agronomic impact, integrates machine learning with unmanned aerial systems (UASs), and develops route optimization tools to enhance precision management.Chapter Two examines the effects of varied Palmer amaranth densities on cotton yield and fiber quality over a three-year field study. Four weed control rates were imposed using a glufosinate-based herbicide program. Fiber quality was assessed using High Volume Instrument (HVI) analysis and Advanced Fiber Information System (AFIS) testing. Results indicated that decreased weed control levels led to increased trash and leaf content in harvested fiber and up to a 54% yield reduction. These findings highlight the agronomic and economic consequences of inadequate weed control.
Chapter Three investigates the use of UAS imagery and machine learning for in-season detection of Palmer amaranth. Annotated UAS images were used to train Faster-Region-based Convolutional Neural Network (RCNN) and You Only Look Once (YOLO) version 11 object detection models (YOLOv11n). YOLOv11n achieved an AP@0.5 = 0.925, while Faster-RCNN required longer training, resulting in an AP@0.5 = 0.77. All models exhibited limitations in differentiating individual plants within dense clusters, highlighting the need for refined annotation strategies and expanded early-season image collection.
Chapter Four presents a route optimization framework for UAS-based economic poison applications. Four heuristic and one exact route optimization algorithm were identified for their use on UAS routing. Monte Carlo simulations were used to evaluate the identified algorithm’s performance across varied polygon counts and spatial clustering. Simulated Annealing and Nearest Neighbor consistently produced shorter routes than competing options under different conditions, with statistical analysis confirming the influence of routing method, polygon count, and clustering on travel distance. A local web-based beta testing interface was developed for visualization of the five routing algorithms and a more efficient and interactive platform.
Together, these chapters address the threat posed by Palmer amaranth and offer integrated solutions to improve identification and site-specific control. By combining applied agronomic research, machine learning, emerging technologies, and optimization tools, this work supports growers by protecting crop productivity and improving their operational efficiency.