The Research On The Neural Network And Fuzzy Algorithm In Remote Sensing Classifications

With the increase of multi-band and multi-resolution remote sensing data, more effective algorithms are needed to improve the classification precision of the new remote sensing data. Traditional supervised classification methods require the assumption that data distribution obeys Gauss normal distribution, however, actual remote sensing data distributions do not always obey Gauss normal distribution, which is one of the reasons of low precise classification by using traditional methods. Therefore, the perspective of the dissertation is to select some nonlinear method and to achieve combined results. The basic elements used in combined methods are SOFM Neural Network Classification, BP Neural Network Classification and Fuzzy Mathematics. The algorithms are realized by programing.This paper takes the Landsat TM5,4,3 and SPOT Remote Sensing image of Guangdong Sanshui as source data. Its major tasks are to research Neural Network and Fuzzy Mathematics using in Remote Sensing image classification. According to the actual situation of the selected areas, we divide the land cover/use into five classes: artificial land(cities and towns,traffic land,mine project land), paddy field, arid land, woodland(forest,lawn, virescence land) ,water. In the experiment based on Neural Network Classification s, we use SOFM Neural Network and BP Neural Network Classification, and have a good result from comparing the result with the result of Maximum likelihood classification. In the experiment based on Fuzzy Mathematics,We analyze and compare the result of fuzzy classification with the result of Hard C-Means Clustering Method ,we can see that it has a good result and higher precision to use fuzzy technique classify remote sensing image.