TECHNICAL SESSION: Indirect Measurement of Disease Development / Spread
Unsupervised learning for efficient detection of plant disease through low altitude remote sensing
Roy Davis - Texas A&M University. Thomas Chappell- Texas A&M University, Young-Ki Jo- Texas A&M University
Low altitude remote sensing (LARS) offers promise for economical, precise, high-throughput detection of agronomically relevant phenomena including disease. A major limitation to realizing this promise has been that analytical techniques have not developed at the pace with which LARS data quality and volume have increased. In this presentation, the unsupervised learning technique of mixture modeling is discussed as an efficient and effective means to utilize LARS-sourced data for disease detection in plant pathosystems. Benefits of mixture modeling include obviating some calibration requirements, and enabling the use of image data captured during unstructured vehicle flight plans. This technique was successful in physical plant phenotyping, and is being developed for detection of diseases of turf grass. Specifically, detection will begin with the causal agent of large patch, Rhizoctonia solani, and by relating symptoms of the pathogen to observable characteristics in the data.