Finding optimal neural networks for land use classification

Abstract
In this letter we present a fully automatic and computationally efficient algorithmbased on the Minimum Description Length Principle (MDL) for optimizing multilayer perceptronclassifiers. We demonstrate our method on the problem of multispectral Landsatimage classification. We compare our results with a hand designed multi-layer perceptronand a Gaussian maximum likelihood classifier where our method produces betterclassification accuracy with a smaller number of hidden units.1...

This publication has 7 references indexed in Scilit: