Weed and crop discrimination with hyperspectral imaging and machine learning

Abstract

Herbicide-resistant weed biotypes are spreading across crop fields nationally and internationally and mapping them with traditional crop science methods - cloning plants and testing their resistance levels in a lab - are costly and time consuming. A segment of the field of precision agriculture is being developed to accurately and quickly map the location of herbicide-resistant and herbicide-susceptible weeds using advanced optics and computer algorithms. In our study, we collected hundreds of thousands of spectra of herbicide-resistant and herbicide-susceptible biotypes of the weeds kochia, mare's tail and lamb's quarter and of crops including barley, corn, dry pea, garbanzo, lentils, pinto bean, safflower, sugar beet at the Southern Agricultural Research Center in Huntley, Montana using a hyperspectral imager. Plants were imaged in a controlled greenhouse setting as well as in crop fields using ground-based and drone-based imaging platforms. The spectra were differentiated from one another using a feedforward neural network machine learning algorithm. Classification accuracies depended on what plants were imaged, the age of the plants and lighting conditions of the experiment. They ranged from 77% to 99% for spectra acquired on our ground-based imaging platform and from 25% to 79% on our drone- based platform.