The International Archives of the Photogrammetry. Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
To test the effectiveness of the addition of topographic data, an
additional classification, using the same methods, landscape
types and training data, was implemented using only the
IKONOS data. The results of these two classifications were
then compared.
For evaluating the accuracy, the master landscape map was
converted from vector data to 4 meter square raster data, and
stratified random sampling was employed to choose a total of
5000 points from each category. Using the master landscape
map as a base, producer’s accuracy, user’s accuracy and Kappa
index were calculated for each landscape category in each of
the two classification methods.
Method: Thematic attributes
Paramter: Lowland
Lowland only
Lowland
Ô
Ô
Upland
Method: Nearest neighbor Method: Nearest neighbor
Parameter: Mean(NIR, R, G, B) Parameter: Mean(NIR, R, G, B)
Ò
Ò
Ô Ô
Figure 5. Decision tree for object-based classification using topographic GIS data
| Conifer Plantation
1 Evergreen Broad-leaved Forest
] Deciduous Broad-leaved Forest
| Bamboo Grove
Grassland
] Wetland Vegetation
] Paddy Field
j Bare Ground
i Rural Residential
S Urban Residential
| Open Water
i Conifer Plantation
\ Evergreen Broad-leaved Forest
j Deciduous Broad-leaved Forest
j Bamboo Grove
Grassland
Wetland Vegetation
| Paddy Field
Bare Ground
j Rural Residential
| Urban Residential
| Open Water
j£2i
Figure 6. Result of object-based classification with topographic
data
Figure 7. Result of object-based classification only
(Kamagata et al., 2006)
9 Conifer Plantation
| Evergreen Broad-leaved Forest
] Deciduous Broad-leaved Forest
| Bamboo Grove
] Grassland
| Wetland Vegetation
] Paddy Field
] Bare Ground
3 Rural Residential
| Urban Residential
| Open Water
(a) Landscape Map (b) Object-based + Topographic data (c) Object-based only
Figure8. Expanded comparison of master map with results of two classification methods
1361
■
