Field identification of weed species and glyphosate-resistant weeds using high resolution imagery in early growing season

Abstract

Accurate weed mapping in early growing season is an essential step in the site-specific weed management system. This study focuses on validating the potential application of high resolution multispectral and thermal UAS images in classification of weed species and glyphosate-resistant weeds at early phenological stages. A field experiment was conducted to evaluate supervised classification methods to identify three-weed species including waterhemp (Amaranthus rudis), kochia (Kochia scoparia), and ragweed (Ambrosia artemisiifolia L.). The accuracy of six classification algorithms namely Parallelepiped, Mahalanobis Distance, Maximum Likelihood, Spectral Angle Mapper, Support Vector Machine and Decision Tree implemented at pixel and object-based levels in weed species classification were evaluated. Thermal infrared imagery was also used to assess the canopy temperature variance within the weed species to identify the glyphosate-resistance status in detected weeds. The object-based algorithms developed with mosaicked imagery effectively classified weed species with the overall accuracy and Kappa coefficient values greater than 86% and 0.77, respectively. The lowest accuracy and Kappa coefficient (67% and 0.58) were observed for pixel-based Mahalanobis Distance algorithm. The canopy temperature-based classification of susceptible and resistant weeds resulted in the discrimination accuracies of 88%, 93% and 92% in glyphosate-resistant kochia, waterhemp and ragweed, respectively.