| Neural networks, which make no assumption about data distribution, have been adopted to classify complex remote sensing data, and achieved improved results compared to traditional statistical methods. The attractions of neural networks also include their ability to learn from empirical examples and simulate any nonlinear decision function. However, a neural network is a black box and it is difficult to determine how a particular classification has been reached. Fuzzy expert systems, on the other hand, are able to represent classification decisions explicitly in the form of fuzzy if-then rules. The weakness of fuzzy expert systems is their inability to learn from empirical examples. The construction of a knowledge base is a tedious and subjective process, a problem often referred to as the knowledge acquisition bottleneck of a fuzzy expert system.; The purpose of this study is to build a neuro-fuzzy system based on the synergism between neural networks and fuzzy expert systems, which provides the best of both technologies and compensates for the shortcomings of each. The learning algorithms of neural networks can be used to extract fuzzy if-then rules for a fuzzy expert system. The rules obtained, in symbolic form, also facilitate the understanding of a neural network based image process. Based on the analysis and evaluation of three existing neuro-fuzzy systems, a hybrid neuro-fuzzy image classification system was proposed and implemented based on a fuzzified learning vector quantization (LVQ) network. The system generated comprehensible fuzzy if-then rules that involved the use of a simple fuzzy averaging operator. In addition, the center of each data cluster and its associated fuzzy boundary in the feature space were obtained. By incorporating human expertise, the hybrid neuro-fuzzy image classification system produced a significantly better image classification than traditional statistical methods and standalone neural network models.; The results of this study indicate that the integration of the learning algorithms of a neural network and the symbolic representation of a fuzzy expert system opened the black box of the neural network and simultaneously broke the knowledge acquisition bottleneck of the fuzzy expert system. |
