Remote-Sensing Recognition Of The Land-Use Types Of Urban Blocks Based On Pixel Set Units

The destination of this thesis is to improve the pattern recognition algorithm for the classification to the land-use of urban blocks on remote-sensing images. Since the original methods of remote sensing classification are working on the individual pixels, the result of recognition cannot reveal the class of the urban blocks directly. As a result, the researching in this field always focuses on optimizing the accuracy of pixel classification.As the following chapters describing, a new method to accomplish the recognition of urban blocks is designed. The conception of Pixel Set is defined to carry the character of urban blocks. With the Pixel Set as direct input of the classifier, a function of pattern recognition can return the class value of the urban blocks easily.A SPOT image of RGB visible band is used to apply this method of pattern recognition. The Pixel Sets on the image is going to be projected to the classes of Old Real Estates, Multi-Layer Apartments, Skyscrapers, Construction Yards and Green Lands respectively.In this thesis, the histograms of the Pixel Set are converted to vectors as the input parameters. Two mature classifiers, Nearest Neighborhood and Multi-layer Perceptron network, are involved in this task, as working with the original routine of pattern recognition.Furthermore, the algorithm of picking out and accessing the Pixel Sets from the whole image is described and another algorithm enhanced in speed is also provided. At the end of the thesis, a conclusion and the future of the method of pattern recognition are also mentioned.