Research On The Object-oriented Method Of Urban Building Extraction With Gf-2 Remote-sensing Imagery

With the development of remote sensing technology, a variety of satellites are constantly launched. Resolution is getting higher and higher. The GF-2 is one of the typical representative, the resolution of the panchromatic band reached 0.8 meters, multi-spectral resolution reached 3.2 meters. So far, GF-2 is only successfully launched for more than a year, research and application of GF-2 is still in the stage of a preliminary study. So this article chooses GF-2 remote sensing image to extract of urban building information, it has the certain research significance.Traditional extraction based on pixel of urban building, has obvious shortcomings, based on pixel only considers the spectral information. When the spectral difference is not obvious, or spectral information is not abundant, the result of the extraction can not meet people’s expectations.So the emergence of the object-oriented idea is undoubtedly a new thought and direction. Object oriented urban building extraction, fully consider the numerous features, not only is the spectral feature and shape feature, texture feature and so on. Especially for high sensing image, the number of bands is generally 4, with respect to hyperspectral, 4 multispectral spectral information is not enough to identify the housing information. At this time, the use of objectoriented thinking to consider a wide range of features building information extraction is an effective means.Object-oriented urban building extraction method, considering the characteristics of many sided, it also brings some problems, how to choose the features and select feature threshold for classification. The traditional idea is that the researchers are constantly experiment, constantly trying, constantly of trial and error, which will waste a lot of manpower and time. Based on the automation of feature selection and threshold selection, as well as the idea of object-oriented, this paper puts forward a new method, based on GF-2 object-oriented urban building of remote sensing image extracted, mainly divided into two stages. One is the image segmentation, the other is the classification of the image. Segmentation using multi-scale segmentation, the segmentation is an object-oriented segmentation, by considering the segmentation scale, shape, spectral, band weight of image segmentation and so on. By comparing the use of multi-scale segmentation on GF2, the segmentation effect is better. Classification using the CART decision tree algorithm to generate a rule set, cart decision tree algorithm using the idea of statistics and data mining makes the feature selection and the threshold selected to achieve automation, then rule sets as fuzzy classification reference. Fuzzy classification can synthetically consider many kinds of characteristic conditions, the threshold can be selected in a certain range of adjustment, at the same time, it will be more conducive to urban building information extraction. And then optimize the building extraction result and precision evaluation, The experimental results show that the method of this paper is better for urban building extraction.