Community Research and Development Information Service - CORDIS


This paper describes a mixture of connectionist and statistical techniques to perform land cover/land use classification of urban areas on satellite imagery. The developed system consists of 4 main stages: feature extraction; feature coding; feature selection; classification. Feature extraction is carried out via a Tree-Structured Self-Organizing Map algorithm. For feature selection a classification and regression trees (CART) based algorithm has been developed. The classification with the selected features is carried out with a multilayer perceptron model. The proposed hybrid system is tested on satellite images which cover the area around the city of Lisbon in Portugal, with promising classification results.

Additional information

Authors: HEIKKONEN J, Helsinki University of Technology, Laboratory of Computational Engineering (FI);VARFIS A, JRC Ispra (IT);WILKINSON G, JRC Ispra (IT);KANELLOPOULOS I, JRC Ispra (IT);FULLERTON K, JRC Ispra (IT);STEEL A, JRC Ispra (IT)
Bibliographic Reference: Paper presented: EANN 97, Stockholm (SE), June 16-18, 1997
Availability: Available from (1) as paper EN 40376 ORA
Record Number: 199711078 / Last updated on: 1997-09-16
Original language: en
Available languages: en