A STUDY OF NEURAL NETWORK PARAMETERS AFFECTING IMAGE CLASSIFICATION

Abstract

Our planet, "Earth" is endowed with rich natural resources that are widespread but at times inaccessible. The advancements in the field of remote sensing has made access to these areas somewhat easy with the use of images captured through satellites. However, data made available by the satellites has to be put through a process called digital image classification before it can be put to any meaningful use. Image classification has been a tool in the hands of scientists and engineers in the field of remote sensing for analysis and classification of remotely sensed images. However the conventional methods of classification have limitations in providing satisfactory results particularly in heterogeneous areas where classes are mixed in nature. Amongst various techniques, which are currently under research, Artificial Neural Network (ANN) promises good hopes. The remote sensing literature reports many classification works using neural networks but with varying classification accuracies achieved. The variation in the classification accuracy has been attributed to various ANN parameters involved. The present study seeks to study the effect of some of these ANN parameters on classification accuracy.