How well does this work?
The following images demonstrate how effective the method is. Note that these images were analyzed as part of batches of images, and that the threshold values used for a batch of images may not be optimal for each single image. We expect to see some mis-classification, but hope not to see too much.
Early-season rye cover crop (March 26, 2008) with corn residue.
Four binary mask from the above image, showing: (A) Leaves, (B) Residue, (C) Bare Soil, and (D) Borderline (pixels with color values within 2% of a threshold).
The false-color image produced from the above allows easy evaluation of accuracy of classification. Green pixels are green, residue is blue-purple, soil is red-black.
Ambiguous (unclassified) pixels are in the orange-yellow color. Note the residue in the lower left corner that has been misclassified as green - comparison with the original image shows that this item has a purplish color to it - it was classified as purple, and so got combined with the green. Altering thresholds to avoid this misclassification would likely lead to classification of purple rye leaves as residue.
An early-season hairy vetch cover crop (March 26, 2008) with residue of a winter-killed nurse crop of oats.
