parameter to be empirically greater than the petiole size, at the
cost of misclassifying twigs and small branches as leaves.
Our method succeeded in detecting first-, second-, and
third-order branches, but small twigs and fine branches were
falsely classified as leaves. Theoretically, the geometric meth-
od can be applied to various tree species, provided there is a
clear branch structure in the point cloud. Multi-scanning to
record the branch structure of each individual tree is therefore
Figure 8. Leaf points (first row, intensity lower than t2) and wood points (second row, intensity greater than t1) separated using the intensi-
ty approach, with different thresholds, for the camphor tree. The intensity values were standardized by setting the maximum value at 1.0.
Figure 9. Leaf points (first row, intensity lower than t2) and wood points (second row, intensity greater than t1) separated using intensity
approach, with different thresholds, for the magnolia tree.
774
October 2015
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
