The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
a T X u a = 1
b T X 22 b = 1
Cor{U,V}= a T I 12 b = max
In fact, the problem can be described as solving the values
a and b that can make a T X )2 b be maximum under the
condition of a T X,, a -1 and b T X 22 b = 1 , which may use
Lagrang solution. After a and b are solved, two group of
random variables are transformed so as to obtain the detection
results from the difference vectors.
3.2 Classification and attribute information acquisition
After the prior information is obtained from land use database,
the attributes of the change polygon is acquired by the
supervised classification of image. Classification is done by the
supervised classification based on fuzzy subordinate level
(Wang Jian et al, 2006). Firstly, the land use information of the
detected region is gotten from the land use database. Then, the
image can be classified according to the information of land use
categories and the prior spectral information. At last, the
polygon attributes may be achieved from the database.
The fuzzy subordinate level function of the classification is
determined by Bayesian formula. That is:
„«(,)- /w**i«*> «>
£/>(<»,№ I ® y >
Where, P(co.) and P(X | co i ) respectively represents the prior
probability and conditional probability of the category (O i . In
the case of normal distribution, the conditional probability is
determined by following formula:
P(X\co i ) =
(2*) ,/2 lX,| ,/2 *
(7)
Where, V. andX / respectively represents the fuzzy mean value
vector and fuzzy variance matrix of the different category.
The fuzzy subordinate levels of the initial sample pixels are
given according to experience and statistical analysis results,
and make up of the matrix called fuzzy matrix. If the number of
samples is n, and the amount of samples pixel is m, then the
fuzzy segmentation matrix can be expressed as:
/*,(*,)
Mi (X 2 ) .,
•• mAXJ
Mi&x)
■■ Mi(X m )
mA*2) ■
mAxj,
Where, ^ (x. ) is the subordinate degree of the pixel x t0 the
category i.
All kinds of the fuzzy mean vectors and fuzzy covariance
matrixes can be obtained according to the fuzzy segmentation
matrix. Each pixel is ultimately classified by its subordinate
degree through the iterative calculation. On the one hand, the
polygons edges are vectorized to update the vector information
of the land use database, on the other hand, the attribute
information is achieved from the land use prior information.
4. PLOYGON DETECTION AND ATTRIBUTE
INFORMATION ACQUISITION
The system may implement the land use database updating
based on the orthoimages captivated at a different time, as well
as the polygon print, and the report forms of the attribute
information output. The primary operation flowline is shown in
Figure 2.
Figure 2. Operation flowline of the dynamic monitoring for land use