Optimization of neural networks for multisource classification in a glaciated terrain

Abstract

In the present study, Artificial Neural Network (ANN) based semi-automatic multisource classification approach has been applied for the glacier terrain mapping in the Kashmir Himalayas (Kolahoi glacier) using integrated dataset comprising of multispectral Landsat TM data and several ancillary data layers (topographic attributes and transformed spectral bands). Terra ASTER data has been used for accuracy assessment. The study aims at selecting the best multisource dataset for classification and to investigate the impact of various neural network parameters on classification accuracy. The present study clearly demonstrates that selection of appropriate multisource dataset; network model and parameter values have a major influence on the performance of the ANN classification process.