<b>INTEGRATING MACHINE LEARNING AND SPATIAL-SPECTRAL ANALYSIS OF HIGH-DIMENSIONAL PLANT IMAGERY FOR DETECTION OF CHEMICAL INJURY AND DISEASE</b>

<p dir="ltr">Herbicide resistance, off-target chemical injury, and crop diseases pose significant challenges to agricultural productivity and sustainability. This dissertation addresses these challenges through integration of advanced machine learning techniques with spatial-spectral analysis of high-dimensional plant imagery. Specifically, it investigates whether hyperspectral and multispectral imaging methods can effectively identify and differentiate herbicide modes-of-action (MOA), detect herbicide drift injuries, analyze interactions between herbicides, and predict disease progression earlier and more accurately than traditional approaches.</p><p dir="ltr">Employing automated, high-throughput hyperspectral systems and the novel LeafSpec touch-based hyperspectral scanner, this research achieved rapid and precise herbicide stress detection with classification accuracies exceeding 80% within one day post-treatment. Furthermore, it demonstrated the first successful integration of Colby’s interaction analysis method with hyperspectral and multispectral imaging, yielding identification accuracies above 90% for herbicide mixtures. In wheat disease studies, LeafSpec hyperspectral imaging coupled with temporal modeling allowed detection of Septoria Leaf Blotch one week earlier than visual diagnosis, significantly enhancing predictive capabilities.</p><p dir="ltr">These results confirm that combining machine learning with advanced imaging not only accelerates herbicide research but also enhances early and accurate detection of plant chemical injuries and diseases, thus supporting proactive management strategies and sustainable agricultural practices.</p>