Computer Science and Engineering, Department of


First Advisor

Ashok Samal

Second Advisor

Sruti Das Choudhury

Date of this Version



Gampa, S. (2019). A Data-driven Approach for Detecting Stress in Plants Using Hyperspectral Imagery (Master's thesis). University of Nebraska - Lincoln.


A THESIS presented to the faculty of the Graduate College at the University of Nebraska in partial fulfillment of requirements for the degree of Master of Science, (Computer Science), under the supervision of Ashok Samal and Sruti Das Choudhury, Lincoln, Nebraska: May 2019

Copyright (c) 2019 - Suraj Gampa


A phenotype is an observable characteristic of an individual and is a function of its genotype and its growth environment. Individuals with different genotypes are impacted differently by exposure to the same environment. Therefore, phenotypes are often used to understand morphological and physiological changes in plants as a function of genotype and biotic and abiotic stress conditions. Phenotypes that measure the level of stress can help mitigate the adverse impacts on the growth cycle of the plant. Image-based plant phenotyping has the potential for early stress detection by means of computing responsive phenotypes in a non-intrusive manner. A large number of plants grown and imaged under a controlled environment in a high-throughput plant phenotyping (HTPP) system are increasingly becoming accessible to research communities. They can be useful to compute novel phenotypes for early stress detection.

In early stages of stress induction, plants manifest responses in terms of physiological changes rather than morphological, making it difficult to detect using visible spectrum cameras which use only three wide spectral bands in the 380nm - 740 nm range. In contrast, hyperspectral imaging can capture a broad range of wavelengths (350nm - 2500nm) with narrow spectral bands (5nm). Hyperspectral imagery (HSI), therefore, provides rich spectral information which can help identify and track even small changes in plant physiology in response to stress.

In this research, a data-driven approach has been developed to identify regions in plants that manifest abnormal reflectance patterns after stress induction. Reflectance patterns of age-matched unstressed plants are first characterized. The normal and stressed reflectance patterns are used to train a classifier that can predict if a point in the plant is stressed or not. Stress maps of a plant can be generated from its hyperspectral image and can be used to track the temporal propagation of stress. These stress maps are used to compute novel phenotypes that represent the level of stress in a plant and the stress trajectory over time. The data-driven approach is validated using a dataset of sorghum plants exposed to drought stress in a LemnaTec Scanalyzer 3D HTPP system.

Advisers: Ashok Samal and Sruti Das Choudhury