Community Research and Development Information Service - CORDIS

Abstract

Neural networks are now frequently offered as alternatives to statistical methods in pattern recognition. It is argued that the two methods are based on fundamentally different mathematical approaches which rather than being put in competition should be used in an integrated manner. A simple procedure is proposed to exploit a combined neural classifier and statistical classifier for the analysis of a multivariate spatial data set derived from remote sensing. The method involves parallel use of a neural and a statistical system followed by a second neural system for handling ambiguous samples. Results from a land cover classification experiment demonstrate that considerable gains in overall classification performances can be obtained using the integrated method. Furthermore it is noted that the model is extendable to other types of classifier and is applicable to a wide range of spatial data analysis problems.

Additional information

Authors: WILKINSON G G, JRC Ispra (IT);FIERENS F, JRC Ispra (IT);KANELLOPOULOS I, JRC Ispra (IT)
Bibliographic Reference: Article: Geographical Systems, Vol. 2 (1995) pp. 1-20
Record Number: 199511197 / Last updated on: 1995-10-10
Category: PUBLICATION
Original language: en
Available languages: en
Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top