004 
N., 
ion 
ote 
APPLICATION OF A GIS AS A MODELING TOOL FOR REMOTE SENSING IMAGE 
ANALYSIS OF AGRICULTURAL FIELDS 
A.A. Abkar*, S.B. Fatemi ^ 
" SCWMRI, Soil Conservation and Watershed Management Research Institute, P.O. Box 13445-11236, Tehran, Iran 
abkar@alumni.itc.nl 
" KN Toosi University of Technology, P.O. Box 15875-4416, Tehran, Iran - sbfatemi 9 yahoo.com 
KEY WORDS: Agriculture, Analysis, Classification, GIS, Remote Sensing 
ABSTRACT: 
In the context of the analysis of remotely sensed data the question arises of how to analyse large volumes of data. In the specific case 
of agricultural fields in flat areas these fields can often be modelled in terms of geometric primitives such as triangles and rectangles. 
In this case the options are classical i.e. bottom-up, starting at the pixel level and resulting in a segmented, labelled image or top- 
down, starting with a model for image partitioning and resulting in a minimum cost estimation of shape hypotheses with 
corresponding parameters. Standard bottom-up classification met 
the pixel individually. But various errors are involved in the image 
assumptions in the classification algorithms, sensor effects, atmosp 
hods usually concern the pixel as a main element and try to label 
analysis with these methods. Mixed pixels, simplicity of the basic 
heric effects, and radiometric overlap of land cover objects lead to 
the wrong detection in image analysis. In this paper we propose a Model-Based Image Analysis (MBIA) approach to analyze the 
remotely sensed data. In this manner using the available knowledge about the remote sensing system we generate some hypothesis 
maps and then test them using the radiometric measurements (images). In order to test the method we used the boundaries of the 
agricultural fields stored in a GIS to model the objects in the scene. The results of the method have been compared with the result of 
a traditional Maximum-Likelihood classification and a stand 
approach we could reach to the 9496 overall accuracy. 
I. INTRODUCTION 
Today remote sensing is a major source of data and information. 
which is used in various fields. Using the remotely sensed 
images we can obtain up-to-date, cheaper, and variety of data 
for different applications. Classification is a common and 
powerful information extraction method, which is used in 
remote sensing field. There are many classification methods that 
have their own advantages and drawbacks. 
Standard classification methods usually concern the pixel as a 
main element and try to label the pixel individually. But various 
errors are involved in the image analysis with these methods. 
Mixed pixels, simplicity of the basic assumptions in the 
classification algorithms, sensor effects, atmospheric effects, 
and radiometric overlap of land cover objects lead to the wrong 
detection in image analysis. 
To overcome theses drawbacks and errors some new methods 
have been developed using the external knowledge about the 
objects. Object-based, knowledge-based classifications are 
some examples of these efforts. But these approaches usually 
are much complex to reach the more accurate results for a 
specific purpose. Then often they cannot be employed in the 
commercial systems simply. 
This research concerns a powerful classification method, which 
is called Model-Based Image Analysis. A remote sensing 
system has two major components including RS data 
acquisition system and data analysis system [Abkar, 1999]. Data 
acquisition system part has four main parts including: 
atmosphere, the scene, sensor, and energy source. 
1281 
ard Object-Based Classification using the boundaries. Using this 
Model-based classification is based on the modelling the data 
acquisition part of remote sensing system. With generating the 
geometric hypothesis about the existing objects on the basis of 
models, we can find the best values of the geometric hypothesis 
parameters. This is done by parameter estimation viz evaluating 
the cost of the set of parameters using the generated likelihood 
vectors (probability for radiometry given radiometric class) 
from multispectral remote sensing data. Finally we choose the 
best parameters based on the minimum cost estimation. For the 
modelling the remote sensing processes we can use any 
knowledge, which we have about them contained in a GIS such 
as air photos, existing maps, etc. 
2. USING GIS AS A MODELING TOOL 
As it was mentioned in the previous section, GIS is a powerful 
data and knowledge source for MBIA. GIS data, as ancillary 
data in image processing and analysis have been used in last 
decade [For example see Hutchinson (1982)]. Firstly, the use of 
GIS in image processing have been limited in providing prior 
knowledge for image processing and analysis, like control 
points for geocoding and prior probabilities for image 
classification. In fact, in this manner we have incorporated GIS 
data to aid the RS techniques but we don't perform any explicit 
integration of GIS and RS [Abkar. 1994]. 
In more advanced methods, GIS can be used to improve the 
image analysis results. The simplest usage of GIS data and/or 
knowledge is in the evaluation stage of the image analysis 
particularly image classification. The results of the image 
analysis are compared with the GIS information and the results 
are assessed. Then, we can decide to perform some 
improvements on the used algorithm. Rule based systems