The Object-oriented Classification Of Urban Remote Sensing Images And The Application Research

High resolution images possess of more and more abundant spatial information and texture information. Conventional information extraction method which based on pixel can't satisfy the increasing processing needs, and a new processing method is in urgent needed. The object-oriented information extraction method of remote sensing emerged as a reflection of the times. It can also avoid "pepper and salt" which is common in information extraction and guarantee the integrity of information extraction. In addition, it improved the classification accuracy.In the base of reading lots of literature of home and abroad, the author exposed the principle of the pixel-based method and the object-oriented method, has analyzed the superiority of the object-oriented information extraction method and the feasibility of used for the urban high resolution image, it has realized the remote sensing information extraction complete process in Ecognition. The author taked xi'an' yanliang's quickbird image for example, with two kinds of classify method to extract information, the object-oriented method showed the obvious advantage than others.The multi-scale segmentation is the innovation of the object-oriented method. Firstly, experiment are carried out which can get the suitable parameter of all types. Established a segmentation hierarchy containing three levels in studying area, and scale, smoothing index, color index for these are (85, 0.8, 0.5), (65, 0.9, 0.6), (30, 0.9, 0.5). Ecognition included the nearest-neighbor and the membership function, choosing out the best method according to the characteristic of all kinds. We can set the customized features so as to optimize the result. This dissertation extracted the road with the membership function method, extracted the water, vegetation, shadow with the customized feature and extracted the nudation and building with the nearest-neighbor method.Accuracy valuation aspect not only had the traditional method which based on the matrix error, it also estimated the object-oriented method with the best classification result and the classification stability which two kinds of method are based on vague mathematics.This dissertation taked high resolution image as data source, with the object-oriented information extraction's method applied to the city's classification, which provided for thenew style of city investigation. Its algorithm is comparatively mature, classification result iscomparatively entire, regardless of from visual effect or accuracy valuation aspect, indicatingthe object-oriented information extraction's method displaying the extremely advantageousprospect.