Extraction Of High-resolution Remote Sensing Imagery Information Of Dianchi Subbasin In Kunming Based On Object-oriented Method

In the past ten years, remote sensing (RS) technology which including satellite RS and aerial RS has been development of an unprecedented scale, and image information processing technology (including data processing and data anslysis and use) become the important aspect and core technology of the application of RS technology.With the development and wide application of high resolution satellite remote sensing imagery, the classification methods of images become the most concern among people. Because the traditional image classification methods are pixel-based rather than object-based classification, it exist lots of errors in practical applications and the accuracy is lower. High resolution images for example, there are much information of spatial data to be wasted. In order to avoid or solve these problems, a new high-resolution remote sensing image classification method came into being-the object-oriented classification method, here "object" described as the collection of a similar to the characteristics of a certain set of pixels. This method not only use the object-based classification, but also by using high-resolution remote sensing image to show the shape, space, texture or other characteristics, which make up the traditional classification methods some disadvantages.I do this research is just for the paper which belongs to the graduate student of major in physical electronic in the kunming university of science and technology, this paper include methods and results of the research. In this paper, the study is based on object-oriented classification, Kunming Dianchi basin will be a small part of QuickBird image as an experimental area, and with this new method I had an in-depth research and analysis to this area. Extracting the information which including unused land, woodland, arable land, water, grass and construction of this part by using Definiens developer 7.0, I get a very good classification results. The last overall accuracy and kappa coefficient is up to 98.19% and 0.9756 respectively. The results of the study can greatly promote the high-resolution satellite remote sensing images in many aspects such as urban construction, urban planning, agriculture, forestry, water, Dianchi Lake governance and the research methods and data can be also use to promote these industries.The research spread according to some aspects, which including data pre-processing, subbasin extraction, segmentation and classification methods research and analysis, information extraction, result comparison, and accuracy assessment.Because of the relatively large subbasin (36685×24552 pixels), before the classification, this subbasin will be divided into 40 small blocks based on 5000×5000 pixels, each small piece to do a separate extract, and obtain a good results of classification. In the data pre-processing, first of all, I extract one of Dianchi Lake Basin as the research area by using ArcGis hydrological analysis module.Before the start of the classification, image segmentation is carried out. The quality of image segmentation will directly affect the classification results. In this paper, image segmentation is also carried out in-depth analysis and research, if the results of segmentation are similar, in order to saving computing time, this article has been taken in the experiment with multiresoluation segmentation that combines quadtree segmentation method, and select the optimal segmentation scale, using space, texture, spectral characteristics of high-resolution images, and at last access to better classification information results.In order to verify the advantages of object-oriented classification, the classification results of new methods must to compare with traditional classification methods. Therefore, this thesis specifically selected a block of the image, using the traditional methods (maximum likelihood classification and ISODATA clustering method) to extract the information of this image, and the classification results will be compared to the object-oriented classification ones, and assess accuracy at the same time. The results showed that compare with the traditional pixel-based classification methods, object-oriented classification can not only effectively avoid the classification disturbance of different objects with the same spectrum and the same objects with different spectrums and "salt-and-pepper" noise, but its classification accuracy is also higher than the first two.