Object-based urban vegetation mapping with high-resolution aerial photography as a single data source

Abstract

In this article, we demonstrate an object-oriented method for detailed urban vegetation delineation by using 1 m resolution, four-band digital aerial photography as the only input data. A hierarchical classification scheme was developed to discriminate vegetation types at both coarse and fine scales. The processes of vegetation extraction include the examination of spectral and spatial relationships, object geometry, and the hierarchical relationship of image objects. The advantages of four different segmentation methods were combined to identify feature similarities, both among image objects and with their neighbours. Image growth took place if those neighbours satisfied a series of criteria given a set of features of class-defined objects. Object-based classification results demonstrated higher accuracy than those using pixel-based classification methods. The object-oriented method achieved overall classification accuracies of 87.5%, 90.5%, and 90.5% at three different levels of class hierarchy, and very high producer's accuracies were demonstrated in the classes of tree, crop, and different types of grass. The object-oriented classification method described here proved effective for separating vegetation types defined by life form, area, or shape without using additional remote-sensing data sources with different resolutions or any ancillary data such as digital elevation models.

Acknowledgement

This research was supported by the project, Partnering for Land Use Sustainability, funded by the Department of Forestry and Natural Resources, Purdue University.