Hardware based neural network data fusion for classification of Earth surface conditions

Abstract

A hardware-based neural network data fusion system is used for fast and accurate classification of surface conditions, based on SSMI satellite measurements. The system processes sensory data in three consecutive phases: (1) preprocessing to extract feature vectors and enhance separability among detected classes; (2) preliminary classification of Earth surface patterns at two separate and parallel acting classifiers, the backpropagation neural network (BP ANN) and the binary decision tree (BDT) classifiers, and (3) data fusion of results from preliminary classifiers to obtain the optimal performance in overall classification. Both the BDT classifier and fusion centers are implemented by neural networks. This system is implemented in a prototype of a massively parallel and dynamically reconfigurable modular neural ring coprocessor. It increases the detection accuracy to 94% compared with 88% for BP ANN and 80% for BDT classifiers. >