Title: Machine learning based ICQP paradigm for plant stress phenotyping
Dr. Arti Singh is an Assistant Professor in the Department of Agronomy at Iowa State University with 11 years of plant breeding experience. After obtaining her PhD degree from G.B. Pant University in India, she worked as a Post-doctoral fellow at the University of Saskatchewan and then at Agriculture and Agri-Food Canada prior to joining Iowa State University. She has authored a textbook ‘Disease and Insect Resistance in Plants’ and published peer reviewed research articles in reputed and high impact journals including Proceeding of National Academy of Sciences and Trends in Plant Science. She has been funded competitive grants by the USDA-NIFA, NSF, Iowa Soybean Association, IA Soybean Research Center, and United Soybean Board. She leads a green (Vigna radiata) and black gram(Vigna mungo) breeding program, and her research projects are geared towards harnessing genetic diversity for genetic gain, utilization of advanced data analytics particularly machine and deep learning for early disease signatures, and genetic/genomic studies on abiotic and biotic stress resistance.
Machine Learning (ML) approaches are rapidly emerging and are being deployed at an unprecedented scale in agriculture to generate automated solutions with higher accuracy. ML methods are versatile tools that can assimilate large amounts of heterogeneous data and provide reliable solutions to complex problems, such as in plant stress phenotyping. Our team is deploying ML tools to analyze image based plant stress data from complex and integrated phenotyping platforms, including unmanned aerial vehicles, unmanned ground vehicles and smartphones for identification (type of foliar stress), classification (low, medium, or high stress), quantification (% stress severity) and prediction (early stress detection), the ICQP paradigm, of stresses to generate insights that were previously not possible. We are also utilizing Deep learning (DL) for automated feature extraction from images to circumvent hand crafting (labelling) of images and provide “explainability” of key features utilized by DL methods to improve user confidence on a hitherto ‘black box’ approach. Finally, the applicability of the learned model was projected on non-soybean stress images to test and validate transfer learning, which opens up exciting opportunities. ML and DL driven data analytics provide new tools to researchers and producers for plant stress phenotyping.
