SEMI-AUTOMATIC IDENTIFICATION OF REVISION OBJECTS IN HIGH RESOLUTION SATELLITE DATA 
Camilla Mahlander, Dan Rosenholm, Katarina Johnsson 
Dept. of Geodesy and Photogrammetry 
Royal Institute of Technology, KTH 
S - 100 44 Stockholm 
Sweden 
Commission IV, 
Working Group 3 
KEY WORDS: rule-based classification, fuzzy logic, clear-cut detection, road detection, SPOT pan 
ABSTRACT 
The methodology for topographic map revision is changing. With the storage of topographic maps in digital databases, maps can be 
produced “as required”. To ensure production of maps which are accurate and complete, the databases must undergo continuous 
revision. A hierarchical system for topographic database revision, based on multiple resolution satellite data, is envisioned. Medium 
resolution data, for example from the RESURS satellite, would be used for initial change detection and signalling, while high 
resolution imagery, such as SPOT panchromatic data, would be 
used to identify and possibly delineate new objects in those areas 
where changes have been detected. The development is currently focused on the forested parts of Sweden (> 60% of the country), 
and revision objects under consideration are new roads and clear-cuts. A rule-based methodology for identification of revision 
objects in SPOT Pan images is under development. The rules are made “soft” by the application of fuzzy logic. Existing map data 
helps in differentiating between new objects and objects which have been previously mapped. In the test area, roads have been 
successfully detected based on spectral reflectance and shape. New clear-cuts have also been detected, although with some confusion 
with other objects. Further development of the rules is ongoing. 
1. INTRODUCTION 
The methodology for topographic map revision is changing. 
Until now map sheets have been revised regularly, e.g. with five 
or ten year intervals. After revision, new maps were printed and 
distributed. With the storage of geographic information in 
databases it is no longer the maps that need revision, it is the 
databases. In a situation where maps are being produced “as- 
requested” from a database it is of importance that the database 
is continuously updated. 
The National Land Survey of Sweden (NLS) is responsible for 
the topographic mapping of Sweden. The topographic maps of 
scale 1:50,000 are stored in a digital database: "Geographical 
Data for Sweden" (GSD). The database is organized by map 
sheet. At present there are 619 topographic map sheets which 
cover most of the country, except for some of the northern parts. 
The latest version of the topographic map (the so called T5- 
version) is more detailed than previous versions, particularly in 
forested areas, where it includes, for example, rock outcrops, 
wetlands, tracks and clear-cuts. Approximately 1/4 of the 
topographic map sheets in Sweden are of this new type, and the 
upgrading of older sheets to T5 is ongoing. 
Once completed, the T5 database will require continuous 
revision. Three main types of revision objects will be regarded; 
clear-cuts, roads and urban development. Entirely new features 
as well as extensions of existing features will be considered. 
Development and implementation of cost-efficient methods for 
operational revision of the T5 map database is of highest 
priority to the NLS. For this reason, the NLS and the Swedish 
National Space Board have initiated a programme for 
“Topographic map revision from remote sensing data”. Within 
this programme methods for semi-automated database revision 
are being developed at the Dept. of Geodesy and 
Photogrammetry, Royal Institute of Technology in Stockholm. 
The programme is currently focused on semi-automated 
methods for revision in forested areas. The chances of 
successful automation are expected to be higher here than in 
areas of more complex landuse. Since more than 60 % of the 
country is covered by forest, automation of this part would 
mean a significant contribution to the cost-effectiveness of the 
revision process. In addition, two out of the three object types 
which are considered to be of interest for revision of the T5 
maps are more or less linked to forests. (Logging roads are the 
most commonly “constructed” roads in Sweden). 
This paper presents a semi-automatic rule-based method for 
detection and identification of forest revision objects for the T5 
topographic maps from SPOT Pan images. In addition, a 
framework for operational revision of the nation-wide 
topographic map database from multi-resolution satellite data is 
briefly discussed. 
2. OPERATIONAL REVISION OF MAP DATABASES 
2.1 Map revision from satellite data 
The revision objects under consideration (new roads and clear 
cuts) are characterized by strongly contrasting spectral 
characteristics to the surrounding forest. Previous studies have 
proved that it is possible to visually detect and identify the 
revision objects from as well Landsat TM, SPOT XS and SPOT 
Pan ( e.g. Ahern and Leckie, 1987; Pilon and Wiart, 1990; 
Malmstrom and Engberg, 1992; Olsson and Ericsson, 1992). 
It has been shown that panchromatic (SPOT Pan) and multi- 
spectral (SPOT XS) satellite data are just as suitable as aerial 
photographs (9,200m) for detection and identification of roads 
and clear-cuts (Malmstrôm and Engberg, 1992). SPOT 
534 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996 
  
pan 
top 
bee 
revi 
ApJ 
But 
resc 
and 
stro 
are 
like 
fron 
data 
initi 
iden 
has 
wou 
high 
very 
the 
of h 
to h 
22 
obje 
In fc 
the 
char 
iden 
For 
simi 
requ 
inter 
for 
info 
of tl 
abili 
visit 
ever 
In ai 
or g 
iden 
char. 
base 
land 
ancil 
soil 
infoi 
kno 
expe 
kno 
Thes 
THE 
more 
deve 
trans 
mem 
hum 
Srini 
Luck