International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
studies, highlighting the potential of ANNs to classify 
vegetation of a high spatial variation. In the example of the 
ANNs consisting of 28 hidden nodes, overall accuracies were 
calculated for both approaches. On average the overall 
accuracy of the remote site data increased by 38%, e.g. for ANN 
E from 22.6% to 64.4%. It showed that the generalisation 
ability and thereby performance of the ANN could be improved 
by adding additional data to the training process. The 
integration of geographic label also comes as no additionally 
cost and provides a simple option to improve the classification 
performance significantly. 
4. CONCLUSION 
This study analysed the suitability of Artificial Neural Networks 
as mapping methodology for regional areas. The ANNs were 
trained using limited amounts of training data of one site of the 
image. The trained ANNSs classified the training and validation 
data of the training site to accuracies comparable to the 
traditional Maximum Likelihood classification. However when 
the ANNs were applied to unseen data of a remote testsite the 
overall accuracy significantly decreased. The ANNs performed 
poorly and did not result in a generalisation ability high enough 
to transfer the learned knowledge to unseen data. The 
performance of the ANNs could be improved if additional 
information of each site, in this case a geographical label, was 
added. It increased the overall classification accuracy of the 
unseen data of the remote site to the same magnitude as of the 
validation data of the training site. It was concluded that ANNs 
as classification methodology across regional areas, and 
therefore also as multi-temporal approach, had failed to 
perform. 
On the other hand it was concluded that the specific 
characteristics of upland vegetation influenced the 
generalisation ability of the ANNs. Upland vegetation consists 
of a high spatial variation, resulting in a spectral variability of 
land cover classes depending on the topographic location and 
interaction of different upland species. 
Simplified ANN classification schemes promise a higher 
transferability. More research is needed to improve the 
transferability of ANNs as classification methodology. The 
choice of more ancillary data or incremental learning offer new 
opportunities to improve its application across large 
geographical areas and as multi-temporal classification 
approach. . 
5. REFERENCES 
Atkinson, P.M., Cutler, M.E.J. and Lewis, H.G., 1997. Mapping 
sub-pixel proportional land cover with AVHRR imagery. 
International Journal of Remote Sensing, 18, pp. 917-935. 
Atkinson, P.M. and Tatnall, A.R.L., 1997. Neural networks in 
remote sensing. /nternational Journal of Remote Sensing, 18(4), 
pp. 699-709. 
Backshall, J., Manley, J. and Rebance, M., 2001. The upland 
management handbook. English Nature Science No. 6. English 
Nature, Peterborough. 
Benediktsson, J.A., Swain, P.H. and Ersoy, O.K., 1990. Neural 
network approaches versus statistical methods in classification 
of multisource remote sensing data. IEEE Transactions on 
Geosciences and Remote Sensing, 28(4), pp. 540-551. 
Benediktsson, J.A. and Sveinsson, J.R., 1997. Feature 
Extraction. for Neural Networks Classifiers. In: J. Austin 
(Editor, Neurocomputation in Remote Sensing. Springer- 
Verlag, Berlin, pp. 97-104. 
Bernard, A.C., 1998. The Identification of Sub-Pixel 
Components from Remotely Sensed Data: An Evaluation of an 
Artificial Neural Network Approach. PhD thesis, University of 
Durham, Durham, UK. 
Bishop, C.M., 1995. Neural Networks for Pattern Recognition. 
Clarendon Press, Oxford, 482 pp. 
Cherrill, A.J., McClean, C., Lane, A. and Fuller, R.M., 1994. A 
Comparison of Land Cover Types in an Ecological Field Survey 
in Northern England and a Remotely Sensed Land Cover Map 
of Great Britain. Biological Conservation, 71, pp. 313-323. 
Duda, R.O., Hart, P.E. and Stork, D.G., 2001. Pattern 
Classification. John Wiley & Sons, Inc., New York, 654 pp. 
Dunne, R.A. and Campbell, N.A., 1994. Some practical aspects 
of pruning multi-layer "perceptron models applied to remotely 
sensed data. Technical Report 94/06, Murdoch University. 
Egmont-Petersen, M., de Ridder, D. and Handels, H., 2002. 
Image processing with neural networks - a review. Pattern 
Recognition Letters, 35, pp. 2279-2301. 
ERDP, 2000. England Rural Development Programme. North 
East ERDP Chapter. 
http://Awww.maff.gov.uk/erdp/docs/nechapter/nesection] 2/ne121 
uplands.htm (accessed 3rd October 2000). 
Foody, G.M., McCulloch, M.B. and Yates, W.B., 1995. 
Classification of Remotely Sensed data by an Artificial Neural 
Network: Issues Related to Training Data Characteristics. 
Photogrammetric Engineering & Remote Sensing, 61(4), pp. 
391-401. 
Foody, G.M., Lucas, R.M., Curran, P.J. and Honzak, M., 1997. 
Non-linear mixture modelling without end-members using an 
artificial neural network. /nternational Journal of Remote 
Sensing, 18(4), pp. 937 - 953. 
Foody, G.M., 1999. The significance of border training patterns 
in classification by a feedforward neural network using back 
propagation learning. /nternational Journal of Remote Sensing, 
20(18), pp. 3549-3562. 
Hanley, H.B. and Fraser, C.S., 2001. Geopositioning accuracy 
of IKONOS imagery: indications from 2D transformations. The 
Photogrammetric Record, 17, pp. 317-329. 
Haykin, S., 1999. Neural Networks - A comprehensive 
Foundation. Prentice-Hall, Inc., New Jersey, 842 pp. 
Heppner, G.F., Logan, T., Ritter, N. and Bryant, N., 1990. 
Artificial neural networks classification using a minimal 
training ‘set: comparison to conventional supervised 
classification. Photogrammetric Engineering & Remote 
Sensing, 56(4), pp. 469-473. 
Kavzoglu, T. and Mather, P.M., 1999. Pruning artificial neural 
networks: an example using land cover classification of multi- 
sensor images. International Journal of Remote Sensing, 
20(14), pp. 2787-2803. 
Lees, B.G., 1996. Neural network application in Geosciences: 
an introduction. Computers & Geosciences, 22(9), pp. 955-957. 
Lippmann, R.P., 1987. An Introduction to Computing with 
Neural Nets. IEEE ASSP Magazine, April, pp. 4-22. 
Mehner, H., Cutler, M.E.J. and Fairbairn, D., 2003. Issues 
concerning the transferability of Artificial Neural Networks. In: 
Scales and Dynamics in Observing the Environment. Annual 
Conference of the Remote Sensing and Photogrammetry 
Society, Nottingham, UK, 
Mehner, H., Cutler, M.E.J., Fairbairn, D. and Thompson, G., in 
press. Remote sensing of upland vegetation: the potential of 
high spatial resolution satellite sensors. Global Ecology & 
Biogeography. 
Op 
ge 
RS 
htt] 
20( 
Sla 
Im: 
27% 
Tal 
Per 
(Ed 
Ma