Spectral characteristics of stems and leaves of various crop and weed species were studied using a diode–array
spectrometer. Five feature wavelengths were selected to form color indices as input variables to a classification model for
weed detection. The feature wavelengths also served as the basis for design of an optical weed sensor. Based on experimental
data, color indices insensitive to illumination variations were designed and tested on the sensor. Laboratory tests showed that
the sensor identified wheat, bare soil, and weeds (several species combined) with classification rates of 100%, 100%, and
71.6%, respectively, for the training data set when the weed density was above 0.02 plants/cm 2 . The classification rates for
the validation data set were 73.8%, 100%, and 69.9%, respectively. When the density of weeds was low, as in the case of a
single weed plant, more than 50% of the weeds were misclassified as soil. Misclassifications between wheat and weeds were
not observed at any weed and wheat densities tested.