Symposium on Remote Sensing for Resources Development and Environmental Management/Enschede/August 1986 
© 1987Balkema, Rotterdam. ISBN 90 6191 674 7 
1053 
Conclusions and recommendations of the working groups 
Working Group 1: Visible and infrared data 
F.Quiel 
Chairman 
Working Group I conducted one plenary session with 
4 papers, a business meeting and participated with 
18 papers in two poster sessions. Influence of pre 
processing on classification results, image segmen 
tation, efficient interactive procedures and map 
production from Thematic Mapper data were some of 
the main topics. 
Topographic maps at a scale 1:50,000 were produced 
in Canada based only on Landsat TM data. In The 
Netherlands geocoded Thematic Mapper images at a 
scale 1:25,000 proved useful for extracting thematic 
land use information —especially in combination 
with topographic maps— and for updating existing 
maps. Landsat data were also used for cartographic 
control in the production of airphoto mosaics in 
developing countries and proved superior to the 
slotted template method mainly due to savings in 
time and a more consistent accuracy. 
Studies of the influence of various preprocessing 
techniques on classification accuracies indicate, 
that for some applications bilinear resampling or 
smoothing algorithms can improve the accuracies. The 
use of topographic information and multitemporal 
data improved classification accuracies in forestry. 
Advantages and disadvantages of image enhancement 
and classification techniques and their complimen 
tary character were also demonstrated. 
A highly interactive environment was presented 
combining the display of image data and two dimen 
sional histograms of training areas with a topomap 
attached to a digitizer. This system was used to 
display the actual distribution of data for training 
areas, to interactively modify the class definition 
in the feature space or the image space and to add 
features to the classification using photointerpre 
tation methods. 
Classification of a multiple data set with air 
borne and spaceborne multispectral data, a digital 
elevation model and geological, soil and vegetation 
maps was performed in Austria. The sometimes strong 
discrepancies between maps and the classification 
results can be attributed —among other reasons— to 
cartographic generalization, differences in class 
definition and spectral subclasses within mapping 
units. 
Available spatial information was used in differ 
ent ways. In one case polygon files with field boun 
daries from a Geographic Information System were 
overlaid with satellite data and only one sample 
point per field was used for the classification of 
that field. In another approach edge preserving 
smoothing, edge detection and tracking using prede 
fined filters was employed for image segmentation. 
Then spectral properties, form and position of the 
segments were determined and used for a per field 
classification. In an attempt to better characterize 
the different structures of the landscape a basis 
for stratified sampling the size, compactness and 
orientation of land use fields in Belgium were com 
pared. 
To extract subpixel linear features, e.g. hedge 
rows in England, in satellite data, differences be 
tween the right, center and left parts of 13 x 3 
elements windows were determined and successfully 
used with simulated SPOT data. 
A new approach to generate digital elevation data 
from contour lines which are digitized with a vidicon 
camera was presented. The new Vertical Measurement 
Module for the stereo Zoom Transferscope to measure 
spot and object heights and terrain elevation was 
shown. 
To clarify the goals and the activities of the 
Working Group, a name change to "Spatial Information 
Extraction" was recommended. The Working Group will 
concentrate on techniques to extract spatial informa 
tion from remote sensing data using e.g. texture and 
context in addition to spectral information. To 
facilitate the evaluation of remote sensing data the 
development of an expert system shell to guide the 
evaluation was also recommended. 
To improve the comparison and test of evaluation 
techniques and procedures, a common test data set was 
recommended. This set might consist of TM and SPOT 
data, a Digital Elevation Model and available landuse 
and other ground truth information in the Freiburg 
area in West Germany. 
An additional meeting of the WG is planned in the 
fall of 1987 in Sweden with an emphasis on evaluation 
techniques for high resolution spaceborne (TM and 
SPOT) and airborne data, e.g. the use of context and 
texture in the evaluation. A tutorial on expert sys 
tems during that meeting is under consideration.