ENVI Classification of Multispectral Images to Track Alfalfa Bloom; Comparing Two Vegetation Indices
Additional Funding Sources
This project is supported by a 2018-2019 STEM Undergraduate Research Grant from the Higher Education Research Council.
Abstract
Alfalfa seed crop managers must coordinate the release of the crop’s pollinators with peak alfalfa bloom to maximize seed yield. High resolution blue, green, and near infrared (NIR) imagery may provide useful information to track alfalfa blooms and better predict the optimal time to release the pollinators.
This study used ENVI image analysis software classification methods to quantify blooms in NIR imagery. Two vegetation indices were applied to the images to increase the spectral separability of the flowers and to compare their accuracy. The vegetation indices are GNVI=(Green-NIR)/(Green+NIR) and FVI=(Blue/Green)*(NIR/Green). After applying the vegetation indices, regions of interest (ROI’s) were selected and used as training data for ENVI’s classification algorithms.
Three ENVI classifications performed well at identifying and quantifying the blooms from the imagery. The classification methods are Constrained Energy Minimization, Maximum Likelihood, and Support Vector Machine. Constrained Energy Minimization with the FVI vegetation index had the highest accuracy of 98.35%.
ENVI Classification of Multispectral Images to Track Alfalfa Bloom; Comparing Two Vegetation Indices
Alfalfa seed crop managers must coordinate the release of the crop’s pollinators with peak alfalfa bloom to maximize seed yield. High resolution blue, green, and near infrared (NIR) imagery may provide useful information to track alfalfa blooms and better predict the optimal time to release the pollinators.
This study used ENVI image analysis software classification methods to quantify blooms in NIR imagery. Two vegetation indices were applied to the images to increase the spectral separability of the flowers and to compare their accuracy. The vegetation indices are GNVI=(Green-NIR)/(Green+NIR) and FVI=(Blue/Green)*(NIR/Green). After applying the vegetation indices, regions of interest (ROI’s) were selected and used as training data for ENVI’s classification algorithms.
Three ENVI classifications performed well at identifying and quantifying the blooms from the imagery. The classification methods are Constrained Energy Minimization, Maximum Likelihood, and Support Vector Machine. Constrained Energy Minimization with the FVI vegetation index had the highest accuracy of 98.35%.
Comments
T97