|
Remote sensing, including satellite remote sensing and aerial remote sensing technology in the past 10 years has been the development of an unprecedented scale, which the production and use of remote sensing images related to the image information processing technology (including depth analysis and use of data processing and data) become important aspects of the use of remote sensing technology and core technology. With the development and wide application of high-resolution satellite remote sensing images (High resolution satellite remote sensing imagery), the classification of images has increasingly become an important aspect of social and economic development. Since the conventional image classification method is based on the pixel, rather than the object-based classification, there are large errors in the practical application, are extracted from the image information of lower accuracy, can not meet the needs of rapid development of social economy; For high resolution images, there are a large number of spatial data information (Spatial information data) is 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 (Object-oriented classification), the \collection. This approach will not only traditional pixel-based classification into image-based object classification, can take advantage of high-resolution remote sensing images show the characteristics of shape, space, texture, and other information to classify the images, which make up the traditional image pixel classification The methods of the great deficiencies. Study Kunming University of Science and Technology College of Physical Electronics direction Graduate Thesis written in the paper, the method and the results of the study. This thesis is based on object-oriented classification, Kunming Dianchi Lake Basin, a sub-basin (part of QuickBird images) as an experimental area, and conducted in-depth research and analysis of the region, and then using Definiens developer 7.0 this molecule watershed feature information extracted successfully extracted the data of six types of thematic information of unused land, woodland, arable land, waters, land for construction, and grass, get good classification results. The object-oriented classification classification results overall accuracy and kappa coefficient of 98.19% and 0.9756, respectively. Use of the results of the study can greatly promote the depth of use of high-resolution satellite remote sensing image in urban construction, urban planning, agriculture, forestry, water supply, environmental protection, and the Dianchi Lake and other industries or aspects of the research methods and data can above industry promotion of use. The study data preprocessing, the sub-basin extraction, image segmentation, image classification (information extraction), the results of the comparison, precision analysis of several aspects. Data preprocessing, firstly with the ArcGis hydrological analysis module, from Dianchi Lake Basin to extract a sub-basin (a sub-basin of the Dianchi Lake Basin), throughout the course of the study as an experimental area of ??the sub-basin. As the sub-basin is relatively large (36685x24552 pixels), classification, size of 5000 × 5000 pixels of the sub-basin is divided into 40 small pieces, and every small piece of information classification separate. The second step of image segmentation (Imagery segmentation). Image segmentation technique is the key technology in the new method, image segmentation is good or bad will directly affect the results of the classification. Image segmentation technique-depth analysis and research, in the case of a similar segmentation results, in order to save the time of the split operator taken in the experiment multiscale quadtree segmentation (Quad tree based segmentation) combined split (Multiresoluation segmentation) method to select the optimal partition scale and parameters. Image Classification (Imagery classification) is the most critical step of this study, spectral information in the study, the use of high-resolution images, and the use of the image contains a wealth of space, texture information for image classification, the experimental area good feature information classification results. Finally, in order to verify the advantages of object-oriented method, and the classification result of the new method were compared with the traditional classification method. Therefore, this paper specifically select a small image of the sub-basin, using the traditional method (largest classification results and object-oriented image classification to extract this block likelihood classification and ISODATA clustering method) will come Classification results were compared, and accuracy assessment. The results show that the traditional pixel-based classification, object-oriented classification method can not only effectively avoid synonyms spectrum, the same spectrum of foreign body \the former two.
|