Abstract: With the improvement of remote sensing image spatial resolution, the spatial
information of ground objects is more abundant, the ground objects’ size, shape and the
relationships between adjacent ground objects are reflected better. As a result, high resolution
image classification methods are currently more emphasized on using spatial information and
many artificial subjective factors are participated in the classification process. In the abandoned
area it is difficult to interpret. Too much fine segmentation causing plenty of operations is another
difficult problem of high resolution images classification. The object-oriented classification of
high resolution images with spectrum analysis is put forward as follows: SAM is firstly classified
roughly with mask operation, and then fine object-oriented classification is conducted. That solves
better the problems of type uncertainty of ground objects and too much fine segmentation in
object-oriented classification. The river and road of the Tibet naqu area are extracted with eight
bands WorldView2 data of 0.5 meters spatial resolution and accuracy reaches 96.6%.

Key words: WorldView2 remote sensing image, spectrum analysis, object-oriented, SAM