The Study On Remote Sensing Image Classification Technology Based On Textural Features

One of the important ways that the Remote Sensing technology uses widely isthe classification of Remote Sensing image. The classified precision influencesthe application level and practical value of the Remote Sensing data directly.How to classify some land use types and satisfy a certain precision is a keyproblem in the study of Remote Sensing image. It hasan important significance. According to the project of "Digital FuJian", this paper takes automaticclassification of TM Remote Sensing image as the research object and takesimproving the classified precision of the image as aim. Its major tasks are toselect the key spectral bands and extract the textural features form the image.And its key techniques are pretreatment of initial data, extraction of features andselection of classifier. At first the author uses the method of Rough Set theory to get key spectralbands and some sub-key spectral bands. By this way it can overcome the defectof traditional method, which gets the result by hand. After this the writer analysesthe spectrum feature of these optimum spectral bands and set up the model to getthe merging image, which not only reduce the data but also hold the basicinformation. Then some comparatively systematic research is done on Texture Spectrumand Fractal theory from aspect of texture analysis. The author chiefly solveseveral problems such as to reduce the texture spectrum, to improve algorithmsof DBC fractal dimension using the thought of Dynamic Programming, to extractvarious kinds of textural features and compare the results, etc. Finally according to the feature of the Remote Sensing image, the authorpresents an image classification method of Remote Sensing image. It extracts aset of textural features based on Texture Spectrum and Fractal theory, includingtexture spectrum, fractal dimension, lacunarity, multifractal and wavelet fractal.And it adopts FCM as classifier. This method takes advantages of texturalfeatures from different aspects and utilizes fully textural information of the - - IIimage. The experiment of classifying the TM Remote Sensing image has beendone and the result is satisfactory, whichverifies the effect of this method.