Key points are not available for this paper at this time.
Urban vegetation plays an important role in quality of life. However, accurate urban vegetation maps cannot be easily acquired from multispectral remotely sensed data alone because the spectral bands are indistinct among different vegetation classes. This study aimed to detect urban vegetation categories from IKONOS imagery based on an object-oriented method that can integrate both spectral and spatial information of objects in the classification procedure and thus can improve classification capability. Considering the characteristics of urban vegetation in IKONOS imagery, a two-scale segmentation procedure was designed to obtain ‘objects’, and the feature set for vegetation objects was constructed. Redundant information among the features was then removed by using correlation analysis, the Jeffries–Matusita (J–M) distance and principal component transformation (PCT). Finally, the vegetation objects were identified by the classification and regression tree (CART) model. The results show that IKONOS imagery can be used to map vegetation types with a total accuracy of 87.71%. Segmentations involving both micro and macro scales could acquire better vegetation objects than using a single scale. The correlation analysis combined with the J–M distance and PCT was efficient in optimizing the feature set. The rule-based classification method is suitable for identifying urban vegetation types using the feature set with a complex structure.
Zhang et al. (Tue,) studied this question.