研究目的

To improve the performance of extracting shaded VS members at individual tree level for urban vegetation mapping thus to avoid some almost inevitable mistakes involved in typical classi?cation.

研究成果

NFSRG, a series of classi?ed results are organized by a seeded-region-growing process. The iteratively captured VS members are always serve as newly added training samples for the next classi?cation or as newly added seeds for the next growing. By imposing the expansion rate on the phase of capturing the shaded VS, some almost inevitable mistakes involved in typical single classi?cation can be avoided. The accuracy assessments have revealed that more than 96% of VS objects can be automatically and accurately extracted by NFSRG.

研究不足

NFSRG works only in a low-dimensional feature space. More other features, for extracting other members instead of vegetation, can be allowed, but a low dimensionality is still required, otherwise too complex relationships may damage its universality. In addition, more buffers can be derived by sectional ?tting but the geographic connection between members of relevant classes also needs to be promised.