An intelligent, plant identification scheme will require multiple types of information to be extracted from digitized scenes. Statistical measures were calculated for reflectance of in situ leaf surfaces in the near-infrared, red, and blue wavebands. Reflectance was quantified by image intensity within a leaf region near the leaf periphery. Mean, variance, and skewness were selected as significant statistical measures. The intensity statistics depended on NIR reflectance, spatial density of veins, and visibility of specular reflections. Three out of forty-eight observations were misclassified when leaf orientation was not a factor. The error rate increased to 16 out of 66 when leaf orientation was a factor. The results indicate that in order to discriminate among individual leaves, the training set must account for leaf orientation with respect to the illumination source.
No takes yet. Share an insight, caveat, or question.
Franz et al. (1991) studied this question.