Detecting Urban Vegetation Using a Object-oriented Method with QuickBird Imagery

Monitoring urban vegetation is one of the major environmental applications in remote sensing. This paper presents multi-scale segmentation and object-oriented image analysis approaches to extract urban vegetation information from high spatial resolution images. Multi-scale segmentation is used to segment an image into highly homogeneous image objects in any chosen resolution and generate a hierarchical image object network. Advantages of object-oriented analysis are meaningful statistic and texture calculation, an increased uncorrelated feature space using shape and topological features (neighbor, super-object, etc.), and the close relation between real-world objects and image objects. This paper applies these new techniques to extract urban vegetation information in Lianyungang City with QUICKBIRD images. Four object layers which represent different area of vegetation are built to extract different scales of vegetation. To different applications, the segmentation scale differs in thousands ways, and here four segmentation scales 800, 300, 90 and 20 are selected corresponding to four object layers. Based on multi-scale segmentation, Classification is conducted by fuzzy logic, and membership functions are used to produce class description, which consists of a set of fuzzy expressions from appropriate sample objects. Image objects are classified to proper classes by evaluation of membership function classifiers. Not only spectral information but also spatial, texture, and contextual characteristics of image are considered for classification. The result of vegetation information extraction is promising and the accuracy of classification is higher than other conventional approaches. It is obvious that these new image analysis approaches offer satisfying solutions to extract information quickly and efficiently for high resolution images and can be applied to many other application fields as well.