Fuzzy rule Base-Multispectral Images Classifications

Abstract

As it is the case in remote sensing images one the main problems in multi-spectral images is that land cover may be more frequent than sampling intervals between pixels in the image. Thus, the pixel representing mixture of land cover is called mixed pixels.The classic algorithm of classification is based on two values right or wrong. When there is overlapping areas of the future space, there will be mistake in the top of classification . thus, in recent years, the application of the fuzzy logic in remote sensing images witnessed rapid developments. Fuzzy set theory provides useful concepts and methods to handle interlocked information, where fuzzy classification is used to put a distinction line between the types, and to take the information from mixed pixels fuzzy classification plays a major role in carrying out full classification. What is done in this research is designing a complete program for supervised classification depending on fuzzy rule base and using trapezoidal membership function to represent prior knowledge.Such a program was applied to remotely sensed data recorded by the TM-sensor (thematic mapper) of landsat-5 satalites . the results were good comparing with the result obtained using the traditional ways such as maximum likelihood (ML) . and neural network such as probabilistic neural network (PNN) the outcome accuracy of classification is shown to be better than those produced by either the ML or PNN. This technique is implemented by using visual C++ 6.0 programming language.