A system applied to recognize specialized objects automatically in remote sensing images is
presented in this paper. The architecture of system and the flow of process are introduced. An approach to
set up object model based on straight-line features and an algorithm of matching between image and model
is proposed. Object model is composed with both a classifying model and a describing model. The former
describes the class features of object and is used to detect the existence of object and its location, while the
latter is a quantitative description of object and can be used to identify different object belong to the same
class. As to matching algorithm, both geometric and physical properties are used hierarchically, and the
possible errors made in line extraction are taken into account, making it more robust and efficient.