the high-resolution remote sensing images in time T2 and land 
use map in time Tl, Development of image processing and 
interpretation techniques, automatic extraction the land use 
change, automatic updating land use maps, and so on, Has been 
a main research topic of remote sensing technology. 
The new methodology of LUCC change detection based on the 
land use segments other then the pixels. In detail, firstly, by 
means of the sample segments according to land use map in 
time Tl, the features of each type of the land use classes are 
obtained in time Tl. Secondly, each samples are analysed, if 
the index of regional similarity between the image segments of 
Tl and T2 is accepted, the samples in time T2 could be 
remained, otherwise the new samples around that sample are 
selected and are judged by the similarity between the samples 
of Tl. Thirdly, each segments of T2 will be classified 
according to the minimum Euclid distance to the T2 sample 
segments accepted, and the corresponding land use type will be 
assigned to the current segment. 
2.2 Methodology and work procedures 
In general, the method has three components: 
Based on the features of the area (A) and perimeter (P) of the 
i k = ^ 
land use segment, the shape index K is defined as : 
2.2.2 The construction of the change index 
When the RS image has been conflated with the land use map 
in time Tl, the preceding characters of each type of the land use 
classes are extracted based on the area of land use segment. 
After that, the characters of each segment are calculated on the 
image in time T2, we can evaluated the similarity of the region 
inside each land use segment at different time. 
In this study, we put forward two types of indicators of 
evaluation of the regional similarity, they are : Index of regional 
similarity (T'V) as well as ratio of characters’ difference( a tj ). 
T t j is defined as: 
1 ( X ik X jk ) 
mh S7 
(i) 
2.2.1 Feature extraction and expression based on the 
LUCC map 
1) Grey character 
There are two kinds of statistical analysis for multi-spectral 
remote sensing data : modular statistics and multi- statistics. 
Traditional statistics includes: Mode, Median, Mean, Range, 
Variance, Standard Deviation, etc. 
Here, Mean , Maximum, Minimum and Standard Deviation was 
chosen for characters for spectra character. 
2) Texture character 
Texture features is one of the most important features of images. 
Texture classification is the basic problem of image processing. 
There are lots of methods for texture analysis, the following 
types can be summed up: statistical approaches, Structure 
approaches, random approaches and so on. At present, the 
traditional method is the statistical method in remote sensing 
image texture analysis . 
where V- : regional similarity of each characters in i 
segment(Tl) andj segment(T2) 
X ik : character K of segment i 
Xj k : character K of segment j 
171 : numbers of characters 
S k : standard deviation of the character k 
where 
s k = 
l 
\n-\ M 
n : numbers of segment each land use classes 
k = (1,2,...m) 
171 : numbers of characters 
(2) 
Due to diversity of dimension among three kind of characters, 
we replace X jk and X jk with x' ik and x'- k . 
In statistical approaches, the gray level cooccurrence matrix 
(GLCM) is used abroad and thought of as most effective texture 
features. In this study, we calculated four 8-gray levels 
cooccurrence matrix with the neighbor pixels displacement 
belonging to {(1, 0), (0, 1), (1, 1), (1, -1)}. We haven’t 
considered the rotation-invariant characteristic of texture 
features in our experiments. We extract four statistics from four 
GLCM: entropy (ENT), contrasts (CON), angular second 
moment (ASM) and inverse difference moment (IDM), with 
their definitions refer to [15]. The major drawback to this type 
of method is long computation time. 
3) Shape character 
The size and shape of the object is one of the important features 
for the object recognition, although spectral feature can 
describe the majority information of the object. But for some 
classes (such as : irrigated fields and orchards, rural settlements 
and independent mining sites) spectral values are very similar, 
but differ in shape. Therefore, for some special classes we need 
to extract the shape features. 
where 
1 « 
X, =-£*a 
» ¡-1 
fly is defined as : 
X 
a H = 
jk 
X ik + X jk 
where x' ik and x' jk are same to equation 3 
(3) 
(4) 
2.2.3 Change detection based on the land use segment 
Each sample are analysed, if the index of regional similarity 
between the image segments of Tl and T2 is accepted, the 
samples in time T2 could be remained, otherwise the new