Qiming Qin
"ree of
A NEW APPROACH TO OBJECT RECOGNITION ON HIGH RESOLUTION
SATELLITE IMAGE
Qiming QIN ,Yinhuan YUAN, Rongjian LU
Peking University, PR China,100871
Institute of Remote Sensing and Geographic Information System
gin ming € yahoo.com
KEY WORDS: Pattern Recognition, Satellite Image, Feature Extraction
ABSTRACT
The data of satellite remote sensing can provide the object information of the earth's surface accurately. Now the
commercial satellites can provide the satellite images with the resolution between 1 meter and 10 meter. With those
high spatial resolution images, shape features and texture features of the ground objects are very clear. With the
characters of the high resolution satellite image, we will use the recognition of the type of water bodies as example,
x by the begin with the extraction of the spectrum features of the ground objects from the satellite digital images , separate
water bodies from background and recognize it through the classification of the image. After the tracing of
of the boundary we extract and describe the shape features of water bodies and implement the recognition of various water
38, N) bodies on the partition of areas. The experiments of the recognition in the satellite image prove this recognizing
approach is feasible for high spatial resolution satellite images.
1 INTRODUCTION
The traditional method acquiring the object's information from the satellite image is through visual
interpretation. The method needs the interpreter has abundant geo-science knowledge and visual interpreting
experience and the interpreter takes a lot of time to interpret it. Its labor intensity is high. The quality of
interpretation is affected by the interpreter' s experience and the familiarity of the region etc. In order to solve this
problem, a few exports begin to try to classify with computer using the spectrum features of the ground objects
from 1970s. Now the commercial satellites can provide the satellite images with the resolution between 1 meter
and 10 meter. With these high spatial resolution images, shape features and texture features of the ground objects
including large buildings in city ,roads, rivers, lakes and other man-made objects is very clear. With the characters
of high resolution satellite image, we describe and recognize the water bodies in satellite image with two level
during the recognition of water bodies with computer in this research. The low level describing object is pixel, and
the value every pixel corresponding to is the spectrum feature value of this ground object. We can recognize the
OO
This project is supported by Chinese National Natural Science Funds (No. 49671064)
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 753