TECHNICAL SESSION: Indirect Measurement of Disease Development / Spread
Machine learning-based early rice disease detection using spectral profiles
Anna Conrad - The Ohio State University. Guo-Liang Wang- The Ohio State University, Wei Li- The Ohio State University, Da-Young Lee- Department of Plant Pathology, The Ohio State University, Pierluigi (Enrico) Bonello- The Ohio State University
Local and landscape-level detection of plant diseases can be a laborious and time-consuming process. Once disease symptoms are widespread, options for management may be limited, particularly in developing countries that lack resources. Therefore, methods capable of detecting diseases before the onset of symptoms, and in a relatively inexpensive manner, would be useful for more proactive and targeted disease management. The objective of this study was to evaluate the applicability of near-infrared (NIR) and Raman spectroscopy, combined with machine learning, for early detection of rice sheath blight (ShB), a devastating disease affecting rice production. To test this approach, we collected NIR and Raman spectra from leaves of the ShB-susceptible rice cultivar, Lemont, and inoculated the base of the stems with agar blocks containing the fungus Rhizoctonia solani, the causal agent of ShB. Spectra were collected from asymptomatic rice as early as one day post-inoculation. Machine learning, including support vector machine analysis, was then used to build and evaluate the accuracy of disease predictive models. Our results suggest that machine learning can be used to diagnose infected but asymptomatic rice plants based on spectral profiles collected from plants in early stages of disease development. This technique holds promise for application in the field, although field testing is needed to validate and refine disease predictive models.