| This thesis is mainly taking the HuangShui river basin as the research sample region. To explore the multi-scale segmentation technology of generating objects and the choice of the optimal scale, we get information from the extracted land use information of complicated terrain, according to the characteristics of the resolution Landsat 8 OLI image, using the object-oriented classification method, and combined with the analysis of the spatial texture characteristics. In order to seek a kind of high precision, high intelligent method to extract the information of land use can serve the complicated terrain. The main research conclusions are as follows:(1) Object-oriented classification method has the advantages of high classification accuracy, high speed, simple operation, etc. The selected images and the integrated global features to be extracted object types. To find the optimal segmentation scale, they can reduce the influence of noise and effectively avoid the phenomenon of "salt and pepper" in pixel classification method based on traditional. After the segmentation of basic processing unit is formed by a polygon object, considering the features in the image features such as shape, texture, it can make the classification result more objective.(2) Huangshui river basins are complex. The entire study area can be divided into different areas through geographical partition to extract information of land use area. Due to the difference of each sub regional geographical elements and spatial distribution, the classification level of the individual is established. It sets the appropriate segmentation scale for each classification level and determines the classification rules according to the characteristics of the object to be extracted. The results showed that the study area by means of geographic division will help extracted land use information in the study area.(3) In this paper, we discuss about the extract chooses OLI images of the optimal band selection and combination of the relevant discussion of the information about Huangshui river basin. Through a single-band information statistics, multi-band correlation statistics and OIF index to determine the optimal band images of the OLI, we can effectively reduce the information redundancy between the image of each band, and ultimately determine the B4, B5, B6 band is the best band selection; Considering OLI image B4, B5, B6 characteristic, USES the band B5 and B6, B4 respectively gives non standard false color composite scheme of RGB. The experimental results show that the proposed scheme is superior to the research area and the land use types of the study area are relatively high.(4) This thesis adopts multi-scale segmentation technology to generate multiple image object layer, and extracted from different levels respectively based on the characteristics of the surface features which are going to be extracted. Finally, this thesis utilizes a research method which is to extract different types of land use information from three classification levels. Through trial and error comparison, the writer discovers only if adopting the measures of setting an optimal segmentation scale for each classification layer, and extracting information of land use by using the method of object-oriented, can researchers get high precision. Among all these precisions, the overall classification accuracy percentage of water area is 84.83%, and the Kappa coefficient is 0.83; Shallow mountainous area’s overall classification accuracy is 86.60%, and the Kappa coefficient is 0.84; Brain mountainous area’s is 88.33%, and the Kappa coefficient is 0.86.(5) This thesis is based on pixels, the support vector machine(SVM) method to extract information of land use in the study area, and the results show that the overall classification accuracy percentage of water area is 77.56%, and the Kappa coefficient is 0.74; Shallow mountainous area’s overall classification accuracy is 81.05%, and the Kappa coefficient is 0.78. As a whole, the extraction accuracy of geographical divisions was significantly lower than object-oriented classification method, but can meet the requirements of classification.(6) By making the comparison of the two kinds of classification methods of extraction, these results can be gotten; the object-oriented remote sensing image classification method improves the efficiency and precision of the classification greatly, and this classification method has a great advantage for medium resolution OLI image. The object-oriented classification method can determine the various classification features through the calculation and statistics of typical image object spectrum, shape, texture, and custom features according to the characteristics of the image itself and the extraction of target characteristics. Furthermore, it can make full use of all kinds of features of objects and the relationships between object classes and different levels, and also can be introduced to the project data related to the land categories which are going to be extracted, such as NDVI, NDBI, MNDWI, DEM, Slope, and so on. The introduction of related auxiliary data layer according to the distribution features of the base class makes the classification more flexible, and at the same time further improves the accuracy of extraction, especially in the complex terrain of study areas. |
