Using The Fractal Method To Extract Remote Sensing Image Spatial Structure Information Applied Research

With the development of remote sensing, the resolution ofspatial and spectrum of remote sensing images increases quickly, andthe computer technique's promotion also make it possible to processvast data. Although use higher spectrum resolution data can get somehigher precision of auto classification, but limited by autoclassification technology itself, not all potetntial of high spatialresolution date have been used. For making the best use of the spatialstructure information provided by remote sensing image, to increasethe precision of auto classification, this study use the model ofDiscrete Fractal Brownian Random field (DFBR) to extract localspatial information of gray-scale image of each band as its texturefeature index. Comparing the results of only using spectruminformation to classify and using spectrum information with itsspatial information to classify, it can be found that as follow:a. Using spatial structure information can increase the precision ofauto classification, special for some landscape, for example, city andpaddy field, etc.b. The window size of local area can affect the result of classification.If it was too small, it could not reflect the spatial structure exactly.And if it was too large, the stability of DFBR would be damaged. Sohow to select the window size should be determined by actual statusof images.c. Window should blur texture feature at the edge of landscape andaffect the precision of classification, sometime even decrease theprecision.