stress.
51-70%, labeled Marginal (M), which warrants further
investigation.
71-100%, labeled High (H), and was considered a
satisfactory development.
The above categories were chosen for illustrative purposes
only. The results are summarized in Table 1.
Finally, the percentage of correctly classified pixels in
each polygon together with the agreement code were entered
into the database.
In the polygon specific statistics generation method the
image statistics were computed within every plantation
polygon to be assessed, one-by-one. These statistics were
then compared with those obtained from the training
samples that matched the age class of the polygon being
assessed. For example, the image statistics obtained within
a polygon with attribute PL1-77 was compared to the
training statistics obtained for the PL1-77 age class.
Ranges were set in terms of the standard deviation (0) of the
training samples for the agreements between the polygon
and the training values for the mean, median, and mode
statistical expressions, to judge the degree of compliance of
the individual polygons with the normal development of the
plantation. Table 2 shows the number of polygons whose
image statistics agreed with the control value at the 30
level.
An additional testing mechanism devised was the ratio by
which the 36 ranges of the training and the tested polygons
overlapped. The count of the polygons that passed this test
above the 0.6 ratio are also shown in Table 2, under the
heading: Ratio.
A comparison of Tables 1 and 2 indicates a good
agreement of the results obtained by the image statistics
based assessment at the 360 level, with the 70-100% class
agreement in the per-pixel classification. The 36
agreement level was, however, chosen for illustrative
purposes only.
4. CONCLUSIONS
Polygon specific image analysis in a GIS is a powerful tool
for monitoring changes and assessing prevailing land cover
conditions. It has numerous advantages:
The monitoring is performed within known geographic
boundaries which are stored in the GIS and can be queried
through the database management system.
* This monitoring scheme provides for an immediate update
of the database.
* Problems caused by overlaps of spectral response patterns
of the various themes, which often plague traditional
image classification, are largely overcome.
* In the per-pixel classification method, the parallelepiped
classification performed on a few carefully selected bands
can provide satisfactory results. This classifier is
computationally the least demanding.
* Satisfactory results can also be expected if image
classification is bypassed entirely and change detection
is based solely on comparing image statistics.
Table 1
NUMBER OF POLYGONS FALLING INTO EACH OF THE
FOUR CATEGORIES OF AGREEMENT WITH APRIORI
EXPECTATION IN THE MODIFIED PER-PIXEL
CLASSIFICATION MONITORING METHOD.
Polygon Total
Code 0-30% 31-50% 51-70% 71-100% Polygon
PL1-77 1 2 6 9 18
PL1-78 1 1 6 22 30
PL1-79 -- 1 -- 9 10
PL1-81 3 2 11 22 38
PL1-87 S -- 2 9 16
PL1-90 1 -- 3 15 19
PL1-93 1 1 4 -- 6
PL2-77 -- -- 1 7 8
PL2-78 -- -- -- 15 15
PL2-79 -- 1 3 20 24
PL2-80 5 2 -- 3 10
PL2-81 -- -- 1 23 24
PL2-92 2 -- 1 4 7
Table 2
NUMBER OF POLYGONS IN AGREEMENT WITH APRIORI
EXPECTATION WITH 30 RANGE IN VARIOUS IMAGE
STATISTICS COMPARISONS.
Polygon Total
M Rati i M Polygon
PL1-77 9 9 13 16 18
PL1-78 27 23 28 28 30
PL1-79 9 7 9 9 10
PL1-81 35 20 35 35 38
PL1-87 12 10 11 11 16
PL1-90 18 18 18 17 19
PL1-93 5 1 5 4 6
PL2-77 7 7 8 8 8
PL2-78 15 13 15 :415 15
PL2-79 24 17 24 22 24
PL2-80 3 3 3 3 10
PL2-81 24 22 24 23 24
PL2-92 5 4 5 5 7
The objective of this paper was to introduce and explain
the concept of polygon specific image analysis as applied
to monitoring changes. The example of monitoring the
development of forest plantations was presented only to
illustrate this concept, and the quality of the results should
not be judged from a forest management point of view. The
real success of this method depends on the decision rules set
at the assessment stage. These rules must be formulated by
discipline oriented experts, on a case-by-case basis through
careful consideration of the specific project. Further testing
and refinement of the ideas presented here are in progress.
Polygon specific image analysis in a UGIS is an
advantageous approach in a wide range of applications such
as monitoring land use changes, updating forest
inventories, agricultural crop classification on a yearly
basis, monitoring compliances with environmental
regulations, etc.
214
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
Dat
res
De]
Dei
sup
En;
De
het:
Dei
ISF
49(
Gel
sca
dat
Em
Hoi
Ca
Lec
Syr
Spe
Rui
for
unc
Int
Ste
En,
