243 
THE STUDY OF LAND USE CHANGE DETECTION BASED ON SOLE PERIOD RS 
IMAGE 
Song Weidong 3 , Wang Jingxue 3 *, Qin Yong b 
J School of Geomatics, Liaoning Technical University, Fuxin, Liaoning, 123000, China—xiaoxuel861@163.com, 
song_wd@ 163 .net 
b Surveying Engineering Department, Shandong University of Technology, Zibo, Shandong, 255049, China 
Commission VI, WG VIM 
KEY WORDS: Sole period RS Image, Spot, Maximum Likelihood Classification, Spot Class Statistic and Judge Function, Change 
Detection in Land Use 
ABSTRACT: 
Depending on the land use vector data, this paper discusses one method of land use changing detection based on the sole period RS 
image. Firstly, this paper overlays the present land use map and RS image in space, then carries out the category attribute judgment 
in the land use spot units according to Maximum Likelihood Classification results. In this step, this paper proposes the Spot Class 
Statistic and Judge Function. And then this paper matches the land use spot units with the classification results, marks the changed 
spots in high-light, and calculates those areas. Finally, this paper carries the precision analysis and evaluation on detecting results 
according to the changed spots areas. The experiment results indicate that this method can obtain a preferable detecting effect. 
1. INTRODUCTION 
Land use is rational or not, not only directly relates to the 
quality of human's life, the social environment and the 
ecological environment, but also decides whether the land use 
sustainable developmental strategy is feasible. Understanding 
the reasons, the process and the future development tendency 
of land use changing, will be the center for various countries in 
the research of land use/land cover. 
Through the investigating to land use changing situation of 
some regions, and gaining the land use changing information in 
certain time, are advantageous to prompt accurately grasp land 
use changing, understand and analyze the present land use 
situation and the change rules of these regions, then serve for 
constituting reasonable land use and regulatory policies; It 
has the extremely important significance to promptly 
understand the region economy development condition by 
studying the space-time distribution, the quantity, the 
characteristic, the dynamic change process and the 
development tendency of the land use. 
The methods of land use change detection mainly can be 
divided into two kinds: one is to use the different period RS 
images to detect the change information, and the other is to use 
the sole period RS image and the present land use map to detect 
the change information. 
At present, most of areas do not have multi-period RS image, 
only have the sole period RS image. According to this general 
situation, the paper discusses a remote sensing change detection 
method, which is based on the land use vector graph. This 
paper utilizes the Maximum Likelihood Classification into the 
sole period SPOT5 multi-spectral image, and compares the 
classification results with the land use vector graph, we could 
detect change information. 
2. MAXIMUM LIKELIHOOD CLASSIFICATION 
2.1 Maximum Likelihood Judgment Rule 
The Maximum Likelihood Classification utilizes the statistic 
features of RS data, and regards the distribution of RS 
multi-band data as multi-dimension normal distribution to set 
up classification judgment function, then obtains the 
classification results. In the viewpoint of probability and 
statistic, if want to judge the vector X in certain position will 
belong to which kind of categories, it should be decided by 
judgment function: 
g t № = % /X) = P(X/w k )P(w k ) 
k = (1,2,3,•••,/«) (l) 
In this formula, W k represents the k category, P(w t /X) 
represents the probability of X belongs to W k , which is called 
post-examination probability, P(\V k ) represents the 
pre-examination probability, which is the probability of X 
existing in W k . In this research, this paper doesn’t consider 
the instance of different categories mixed in one pixel, and 
regards pixel X only belonging to one given category. Because 
this given category is unknown, this research should calculate 
the all probabilities of X belonging to each category, and 
compare those magnitudes, then divide this pixel to the 
category having the biggest probability. The judgment rule is: 
P(w k IX)>P(w ¡ IX) (2) 
when 
k * j, j = (1,2,3, — ,m) 
formula is existing, 
then 
X ew k 
2.2 Multi-variable Probability Density Function 
The known pixels belonging to each category constitute some 
certain point clusters in plane or space. Each unidimensional 
data of each category forms one normal distribution in its own 
number axis, and the multi-dimension data of this category 
forms one multi-dimension normal distribution. As the RS data 
usually represent multi-variable normal distribution, it makes