Detection of Weed Species in Soybean Using Multispectral Digital Images

Abstract

The objective of this research was to assess the accuracy of remote sensing for detecting weed species in soybean based on two primary criteria: the presence or absence of weeds and the identification of individual weed species. Treatments included weeds (giant foxtail and velvetleaf) grown in monoculture or interseeded with soybean, bare ground, and weed-free soybean. Aerial multispectral digital images were collected at or near soybean canopy closure from two field sites in 2001. Weedy pixels (1.3 m2) were separated from weed-free soybean and bare ground with no more than 11% error, depending on the site. However, the classification of weed species varied between sites. At one site, velvetleaf and giant foxtail were classified with no more than 17% error, when monoculture and interseeded plots were combined. However, classification errors were as high as 39% for velvetleaf and 17% for giant foxtail at the other site. Our results support the idea that remote sensing has potential for weed detection in soybean, particularly when weed management systems do not require differentiation among weed species. Additional research is needed to characterize the effect of weed density or cover and crop–weed phenology on classification accuracies.