Drought stress detection using low-cost computer vision systems and machine learning techniques

Authors: RAMOS-GIRALDO, PAULA - North Carolina State University; REBERG-HORTON, CHRIS - North Carolina State University; Locke, Anna; Mirsky, Steven; LOBATON, EDGAR - North Carolina State University

Submitted to: IEEE IT Professional
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/27/2020
Publication Date: 5/21/2020
Citation: Ramos-Giraldo, P., Reberg-Horton, C., Locke, A.M., Mirsky, S.B., Lobaton, E. 2020. Drought stress detection using low-cost computer vision systems and machine learning techniques. IEEE IT Professional. https://doi.org/10.1109/MITP.2020.2986103.
DOI: https://doi.org/10.1109/MITP.2020.2986103

Interpretive Summary: Visual assessments of wilting are still critical indicators of crop drought stress for on-farm and research applications. The automation of these assessments could improve precision irrigation technologies and increase the throughput of crop phenotyping for breeding applications. Here, cameras were built using low-cost, commercially available parts. Images from these cameras were used to train machine learning (ML) models to evaluate leaf wilting. Using ML, the computer was able to match expert human ratings with over 80% accuracy.

Technical Abstract: The real-time detection of drought stress has major implications for preventing cash crop yield loss due to variable weather conditions and ongoing climate change. The most widely used indicator of drought sensitivity/tolerance in corn and soybean is the presence or absence of leaf wilting during periods of water stress. We develop a low-cost automated drought detection system using computer vision coupled with machine learning (ML) algorithms that document the drought response in corn and soybeans field crops. Using ML, we predict the drought status of crop plants with more than 80% accuracy relative to expert-derived visual drought ratings.