| With the development of the satellite technologies, the resolution and timeliness of the remote sensing images have been greatly improved, remote sensing datas have become the important sources to collect the geographic informations.Those traditional automatic classifications almost use the gray of images, but the texture which describes the spatial structure of features was not or lessly been used in the automatic classification. This Resulted some misclassifications and inefficient using of the images,and directly affected the final classification accuracy. With the development of remote sensing technologies such as sensors, remote sensing platforms and so on, the geographic information provided by images is more and more, the texture have gradually been used deeper than ever, especially the improvement of image resolution makes more detailed structures of the surface features clear, the rich texture can be used to support the traditional gray automatic classification, and improve the image classification accuracy furtherly.This paper Bases on the Sino-German scientific cooperation and exchange programs (2007DFB70200) and Shandong Province Natural Science Foundation (Y2008E10), used spot-5 images, extracted quantitatively the texture by Fractal Theory, quantitatived the texture as a single band of grayscale image, and then supported the image classification as a single band image .The paper which relies on reading and analyzing some related references and previous studies , introduces the texture and the fractal theories, analyzes the pros and cons of various fractal methods, chooses the main method of extracting the texture .It analyzes the characteristics of remote sensing data and selects the research data ,processes the image with some image processing softwares, which includes: correction, integration, enhancement, image cut and normalized. It calculates the Differential box dimension by programimng, determines the sliding window size by trial experiments in matlab, calculates the fractal dimension of the study area on this basis, analyzes the fractal characteristics of the study area. It exports the quantitatived fractal dimension to a single-band image,carries out the traditional gray- classification and exture-classification using remote sensing software of the study area, analyzes and evaluates the classification accuracies of the both.The study obtains the Differential box dimension of surface features in sample area and quantitatived texture. It makes the texture as a gray band to support the traditional gray-classification, improves the final overall accuracy and Kappa coefficient. The paper although realizes remote sensing classification by texture, but there are still some problems, such as the study of the fractal dimension which is limited on ideal size of the image instead of random size image; ignores the edges of the image during the fractal dimension calculating; the selection of the sliding window size needs further research. |
