Opening the black box of neural networks with fuzzy set theory to facilitate the understanding of remote sensing image processing

Abstract

The purpose of this research is to facilitate the understanding of remote sensing image classifications based on the integration of neural networks with fuzzy expert systems, which is often known as neuro-fuzzy systems. A neuro-fuzzy system is basically a fuzzy system that uses a learning algorithm derived from or inspired by neural network theory to determine its parameters (fuzzy sets and fuzzy rules) based on sample data. The learning approach of neural networks can extract the fuzzy if-then rules and fine-tune the membership parameters based on empirical examples. The generated rules in symbolic form are comprehensible so that human expertise can be incorporated in the classification process. In this way, the black box of neural networks can be opened so that the decision process can be made transparent. This generalization ability can also be available to fuzzy systems. The combination of neural networks and fuzzy expert systems obtains the best of both worlds and compensates for the shortcomings of each.