27
sments by birds and
adjoining farmland
Logy. 16:349-358.
Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986
A GIS-based image processing system for agricultural purposes
(GIPS/ALP) - A discussion on its concept
J.Jin King Liu
Remote Sensing Group, Energy and Mining Research/Service Organization, Industrial Technology Research Institute, Shin
Chu, Taiwan
AB^TRAC.. Taree major bottlenecks wnich hinder the development of a complete operational application
of remote sensing tecnniques to the monitoring and survey in low to medium economic return agricultural
areas are the data volume, classification accuracy and the geometric accuracy.
A GIS-based image processing system for agricultural land use and land cover purposes (GIPS/ALP) is
thus devised to resolve these difficulties. Instead of handling the whole lot of image data, this algorithm
samples only a small fraction of them for processing and classification based on the polygon files of
tne agricultural fields in the GIS. And, consequently, this approach dramatically reduces ti.e data volume
;hich has to be processed even when spatial resolution is greatly improved. In addition, tne classification
accuracy is considerably improved due to the purity of the pixels chosen and high accessibility of ancillary
data available in GIS. -he field check strategy modified by this system can contribute the general accuracy,
too. The geometric accuracy can be as good as that .in the GIS which has been claimed to be considerably
higher than that can be derived from the most sophisticated corrected images.
/
1. INTRODUCTION
This paper is intended to discuss the basic concept
for implementing a GIS-based image processing
system for agricultural land cover and land use
purposes (GIPS/ALP). Firstly, the bottlenecks
in image classification which have to be resolved
for economical application of remotely sensed
data are briefly sketched. Secondly, the basic
concepts of GIPS/ALP are parsed by detailing
the implementing approaches. Finally, the unique
advantages of this new approach are concluded. 2
2. THE BOTTLENECKS IN IMAGE CLASSIFICATION
Three major bottlenecks which hinder the development
of a complete operational application of remote
sensing techniques to the monitoring and survey
in low to medium economic return agricultural
areas are the data volume, classification accuracy
and geometric accuracy.
It is argued by Allan (1984) that the smaller
the phenomena of concern and the more frequently
they need to be monitored the less likely it
is that the current remote sensing systems will
provide useful data. This is especially true
for the situation in Taiwan where the land use
patterns are complex and the crop fields are
rather small (usually less than 100 by 100 meters).
The average accuracy obtained for the first level
of land use/land cover classification (Anderson
et al 1976) is 83%>, and it decreases to 58% for
the second level (Tseng & Sung 1978).It is generally
recognized that in most geographic information
systems, remote sensing inputs do not represent
the primary source of data (Marble & Puequet
1983) because that remote sensing data themselves
are not thematic and their derived or classified
resulcs are generally less accurate in both position
and class level. It would be delightful if the
geometric accuracy of remote sensing outputs
is comparable or equivalent to that in the data
base of geographic information system.
The previous discussion means that if both
the classification and geometric results are
to be more accurate to meet the needs for the
management of agricultural lands in the environments
such as Taiwan, spatial, and spectral resolution
of the remote sensing systems should be greatly
improved. Suppose these two conditions are realized,
another issue will be accordingly raised. It
is that the volume of data which has to be processed
in a single project would be too large, in reality,
the limitation of spatial, spectral and temperal
resolution of remote sensing data for effective
operational applications to agricultural purposes
are basically lied in the data volume. This is
manifested in the diagram by Allan (1984) which
shows that at present stage the vr.iurae of remotely
sensed data is too large to be economically handled
for the purpose of crop discrimination.
A GIS—based processing approach is thus devised
to resolve these difficulties. Instead of handling
the whole lot of image data, this algorithm samples
only a small fraction of them on bas^s of the
polygon file of the agricultural fields in '-he
GIS. Further discussion is given in the following.
3. THE BASIC CONCEPT OF GIS BASED IMAGE PROCESSING
3.1 GIS and TSS data files
One of the data files in geographic information
system is the polygon file of which a polygon
is an ordered set of coordinate points enclosing
a crop filed. Each polygon can be referred by
a point called a POLYCENTRE or FIELDCENTRE locatec
at the geometric centre of the polygon as shown
in figure la. In the other hand, the data files
in remote sensing system are digital images.
An image consists of discrete picture elements
called PIXELS of which the location in space
is indicated implicitly by the relative position
of each pixel in the image as shown in figure lb.
The GIS polygon file records a relatively
permanent ground feature, i.e. the crop boundaries
or cadastral boundaries. However, the content
or the land use type of each polygon is changeable.
The land cover types in GIS stand only for the
condition of a particular time of a particular
season in the past. Neverthless, what a land
use planer or manager is interested in is the
extent of each cover type of present time or
in the near future as well as the extent of changes
between between two certain times. This information
can be obtained, if a sample of eacn crop field
