Extended Methods for Classification of Remotely Sensed Images Based on ARTMAP Neural Networks

Authors:
Norbert Kopco;Peter Sincak;Howard Veregin
Affiliations:
-;-;-
Venue:
Proceedings of the 6th International Conference on Computational Intelligence, Theory and Applications: Fuzzy Days
Year:
1999

Citing 4
Cited 0

A massively parallel architecture for a self-organizing neural pattern recognition machine

Computer Vision, Graphics, and Image Processing
ARTMAP: Supervised real-time learning and classification of nonstationary data by a self-organizing neural network

Neural Networks
Gaussian ARTMAP: a neural network for fast incremental learning of noisy multidimensional maps

Neural Networks
Remote Sensing Digital Image Analysis: An Introduction

Remote Sensing Digital Image Analysis: An Introduction

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper deals with two aspects of the application of ART-MAP neural networks for classification of satellite images obtained by remote sensing of the Earth. The first part contains an analysis of the influence of data representation and cluster determination method on classification accuracy. Three types of representation/determination are analyzed. Best results are obtained for Gaussian ARTMAP, an ARTMAP neural network using gaussian distributions for identification of clusters in feature space. In the second part, a method for evaluation of the classification quality is described. This method introduces a confidence index which is assigned to each individual pixel of the image, thus allowing generation of a confidence map for the classified image. The confidence index is computed conveniently exploiting features of the Gaussian ARTMAP learning algorithm. Using a threshold determining the minimal required confidence, this method allows one to generate a map which shows only pixels with prescribed minimal confidence, effectively creating an additional class containing pixels classified with subthreshold confidence.