The results from the neural network part in phase 2 and 
the comparisons here are given for the same two groups 
of stands. Stands that are "spectral" homogenetic have a 
lower correct classification rate of 34-53% compared with 
the 70-95% for the statistical classification . Some of the 
"spectral" heterogeneous stands are better classified with 
the neural network, but they still have too low 
classification rate (4-43%) to be compared with the 5-30% 
in phase 1. 
After the object-classification of the stands the differences 
are smaller. For the homogeneous stands two were 
changed from the correct classes. But for the 
heterogeneous stands a few more stands were put into the 
correct super class with the neural network classification. 
In the low-productive forest area none of the methods are 
useful depending on the dominating background- 
reflection. Small stands are another group that are not 
suitable for this kind of classification with the sensors we 
have available today. 
5. CONCLUSIONS AND REMARKS 
The results do not indicate that neural network classifier 
would be useful enough to this kind of forest stand 
classification. Especially when considering the problem of 
setting the variables for a non-specialist, the time 
consumption and the lack of confidence in the results 
when the behavior sometimes seems to be unpredictable. 
Out of 17 tests with larger neural networks (with 50 to 
100 nodes and from 30 minutes to 50 hours iteration 
tme) only two seem to reach some kind of global 
minimum. The other test ends up with a high average 
system error. For the best training results there are no 
intermediary results stored so it had not been tested to see 
if it was overtrained as is indicated by [Skidmore et al. 
1994] among others. 
ACKNOWLEDGEMENTS 
Special thanks are directed to the initiator of the project 
Assistant Professor Asbjórn Kjellsen, North-Trgndelag 
College for his encouraging support. 
REFERENCES: 
Dalen, 1995. Bruk av neurale nettverk for 
fjernmálingsklassifikasjon. Project report at The 
Norwegian Institute of Technology, Department of 
Surveying and Mapping. 
546 
Hagner, O. 1990. Computer aided forest stand delineatio; 
and inventory based on satellite remote Sensing, 
Proceedings SNS/IUFRO, Umeâ, Feb. 1990, 
Hagner, O. 1991. À prototype GIS/remote sensing work 
station for forest map revision and updating, 
Proceedings(?) SNS/JCR-LFA, Copenhagen 1991. 
Jansen, L.L.F., J. Van Amsterdam. 1991. An Object- 
Based Approach to the classification of remotely sengeq 
images. Proceedings IGARSS, Espoo, Finland 1991, 
Kolstad, J. 1993. Bruk av fjernmälte data ; 
skogbruksplanlegging og eiendomsforvaltning. Parti]. 
report la-c, Norskog, Norway, Feb. 1993. 
Lócherbach, T 1992. Interpretation of aerial and satellite 
images. Course compendium Institut für Photogrammetrie, 
Universität Bonn, First course in digital photogrammetry, 
Bonn, Germany, Oct. 1992. 
Nordahl, B. A, Kjellsen. 1994. Fjernmálingsprosjektet, 
Tolkning og praktisk anvendelse av en SPOT satellittscene 
i skogbruket. Preliminary report: NTDH, Steinkjer, Norway 
May 1994. 
Nordahl, B. 1995. Integration of SPOT-XS Data with 
Ancillary Data in a Postobject Classification for Forestry 
Registration. Proceedings ScanGIS’95, 
Trondheim,Norway. 
Nordahl, B. 1996. Skogregistrering med SPOT-Opptak - 
Project-rapport. Proceedings NKTF Kartdagene 1996 
Beitostglen, Norway. 
Richards, J.A 1993. Remote sensing digital image 
analysis - an introduction, second, revised and enlarged 
edition. Springer-Verlag. Chapter 8 part II and Chapter 
12. 
Skidmore, A.K, W. Brinkhof, J.Delaney 1994. Using 
Neural Networks To Analyse Spatial Data. 7th 
Australasian Remote Sensing Conference Proceedings 
:Poster and Post-deadline, Melbourne 1994 
Strand, G-H.1989. Remote sensing of forest - An 
inductive approach. Norwegian Computing Center report 
no. 824, 1989. 
Tharaldesen, C., M. Angeloff 1992. Kartlegging d 
bj?rkeskogen i Vang kommune ved hjelp av satellitt. 
NIJOS-report, Norway, July 1992. 
Zell, A., et al. 1994. SNNS User Manual, Version 3.5. 
University of Stuttgart, IPVR Report, No.3/94. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996