The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
454 
interactive visual interpretation method which forms along with 
the computer technology and the remote sensing image 
processing technology's development. The image can be 
zoomed in, zoomed out, moved and enhanced on the software 
platform. According to the screen interpretation symbol of land 
objects, the interpretation person outlines the border line 
directly with the mouse along the image characteristic edge 
after the image achieves the best visual interpretation effect. 
The supervised classification is one of the computer automatic 
classification methods. Its most basic characteristic is before the 
classification people have investigated the sample area. With 
the artificial visual interpretation, people have got the prior 
knowledge about land objects’ category attribute. Then the 
computer trains the discrimination function to complete the 
classification according to the knowledge. The classical 
supervised classification methods are the maximum likelihood 
method, the minimum distance method, parallelepiped method 
and so on (Lillesand, 1994; Cshowengerdt, 1983). 
The interactive visual interpretation can make full use of the 
interpretation person's experience and knowledge. It is flexible 
and good at extracting spatial information, but spends more 
time and has difference between individuals. Computer 
automatic classification consumes shorter time, and its 
repeatability is good. But because it is difficult to use the 
person’s knowledge and extract spatial information, it causes 
the wrong classification and omitting some land objects easily 
(Zhang Yinhui, 2002). 
3. REMOTE SENSING IMAGE INTERPRETATION 
METHODS SERVING URBAN PLANNING 
3.1 Supervised Classification 
As a result of the urban objects' complexity as well as the urban 
planning’s requirements of the function districts partition, only 
the computer automatic classification cannot satisfy the needs 
of the urban present situation investigation. Now take 
supervised classification as an example to discuss it. 
The supervised classification usually carries out the 
classification of land objects in the image by establishing study 
samples of typical land objects, such as the building, the water 
body, the vegetation, the road and so on. Although supervised 
classification consumes shorter time, but is not suitable to the 
land objects classification in the urban planning. Firstly, in the 
way of the remote sensing classification's requirements of the 
image, because there are too many kinds of the urban land 
objects, the matter of same kind of objects having different 
spectrums and the different kinds of objects having same 
spectrum is very common. It is also impossible to establish the 
samples of all types. So it is easy to make the wrong 
classification or there are not enough types. And the buildings 
are very crowded. The shadow’s influence is serious, which 
reduces the accuracy of classification results. Next, in the way 
of requirements of urban planning, the key point of urban 
planning and the management is studying the rationality of the 
urban function unit and the structure, the scientific nature of 
urban configuration as well as the appraisal and forecast of 
urban environment ecosystem and the urban environment 
pollution (Shen Qi, 2006). The land objects having similar or 
the same spectral characteristic may belong to the identical 
category, but their functions in the urban are not always same. 
For example, the main road and the inferior road belong to the 
same road type, but they are actually different in municipal 
transportation's supporting capacity. A certain urban function 
unit contains many kinds of land objects mostly. In other words, 
an urban function unit contains different spectral characteristics. 
It is impossible to realize the discrimination of urban function 
units by supervised classification or any other computer 
automatic classification methods. 
3.2 Man-Machine Interactive Visual Interpretation Based 
On GIS 
Compared with the supervised interpretation, the man-machine 
interactive visual interpretation can use the operating person's 
experience and the knowledge fully, which is advantageous in 
associating with other non-remote sensing information, and can 
distinguish the goals and their nature and function in the image 
using the contrastive analysis method. Therefore, although 
visual interpretation consumes more time, but in the way of 
classification accuracy and the scientific nature, the visual 
interpretation has more superiorities, and is more suitable for 
the urban planning, compared with supervised classification and 
other computer automatic classification methods. 
First of all, the remote sensing image used in the urban present 
situation investigation should be able to display the land 
objects’ geometry characteristics and geographical positions as 
well as the detail information precisely. The high spatial 
resolution's aerial image containing the rich spatial information 
and the clear textural property can satisfy urban information 
demands well. It has very big superiority in the fine land object 
extraction and the quantitative investigation, and is helpful in 
realizing the more environmentally-friendly, more economical, 
and more accurate planning, which has already became the 
essential data to establish modernistic city and realize urban 
sustainable development (Shen Qi, 2006). 
In the interpretation process, the urban digital line graphics 
about road, building, land objects points and so on should be 
referred to for a clearer understanding about the urban land 
object’s function. 
In the GIS platform, the existing urban land object’s vector data 
layer can be added. And the high resolution aerial image can be 
zoomed in, zoomed out and moved for the man-machine 
interactive visual interpretation easily. Meanwhile the 
classification result can be written into the attribute database 
directly, which is advantageous to the data’s management and 
sharing. 
3.3 Case Study 
Take 0.2 meter resolution's aerial image and 1:2000 urban 
digital line graphics (contains road, building and other layers) 
of some district in Beijing, as well as the ArcGIS platform as an 
example to explain the man-machine interactive visual 
interpretation method’s superiority in present situation 
investigation serving the urban planning. 
For the first place, register the image based on the existing road 
layer. Then do the edit on the basis of the road layer according 
to land objects points distribution and land objects’ 
characteristics reflected from the image, as well as <Urban 
Land Classification And Planning Construction Classification 
Standard CBJ137-90>, to get the different urban function units 
(e.g. Figure 1).