DEVELOPMENT OF SOFT COMPUTATIONAL SIMULATOR FOR OPTIMIZED DEEP ARTIFICIAL NEURAL NETWORKS FOR GEOMATICS APPLICATIONS: REMOTE SENSING CLASSIFICATION AS AN APPLICATION

Abstract

Artificial neural networks (ANN) obtain more importance after the innovation of deep learning (DL) approach. This research is oriented towards development of soft computational simulator for geomatics research using ANN supporting the deep approach. ANN seems to be a black box due to its sensitivity towards initialization, architecture, and behavior. This research gives a spotlight on the dull areas of ANN algorithm by developing a soft computational simulator for it. The applied examples are chosen to cover geomatics data. DANNDO (Deep Artificial Neural Networks Designer and Optimizer) software is developed to achieve the research objective. Multi-layer perceptron (MLP) architecture is applied in this simulator. Geomatics (remote sensing multi- spectral data) is selected to be a testing paradigm to insure the reliability of the developed simulator. The developed simulator proved the high performance of applying both shallow and deep ANN (DANN).