Research On Information Extraction Of WorldView-2 Imagery With Object-Oriented Technology

In recent years, satellite remote sensing technology is developed rapidly. With the expansion of application, remote sensing technology is attracted more and more attention as an advanced spatial exploration technology. With the improvement of sensor systems, the spatial resolutions of images are higher which provides many diverse data for many fields. However, it is an urgent requirement and challenge of remote sensing for utilizing these high spatial resolution images that how to utilize fully the information in these data and how to identity and extract the information quickly and accurately.Traditional classification methods for remote sensing images are mainly based on the spectral characteristics of features, which are suitable for remote sensing images with medium or low spatial resolution. The high spatial resolution images are characterized by geometry, spatial location and texture information, but spectral information is relatively weak. Thus, the traditional pixel-based methods to extract high spatial resolution images can not effectively utilize the abundant information of spatial structure, and have problems such as, incorrect classification, "salt and pepper effect" due to the heterogeneous spectral reflectance in a specific entity, and these problems affect the classification accuracy seriously. Therefore, these pixel-based methods are difficult to apply for information extraction from high spatial resolution images. To tackle these problems, object-oriented analysis technology is introduced. It is received attention, because it can overcome the limitations of traditional methods and take full advantage of spectral and spatial information of the high-resolution images to extract the desired feature information quickly and accurately.This study takes Yushan town in Fuyang city as the study area, using WorldView-2 image to identify and extract information of land cover features. WorldView-2 system is launched by Digital Globe Company in October 2009. which is a new generation high spatial resolution commercial remote sensing system. The system provides 0.5 meter panchromatic images and 1.8 meter multispectral images. In addition to high spatial resolution and rich spatial information, the 8 spectral bands can be more valuable for object identification and information extraction.In the study, WorldView-2 image fusion is studied carefully. It is found that the Gram-Schmidt transformation method can preserve more spectral information and enhance the texture than other methods with qualitative and quantitative evaluation. The evaluation of this study provides a useful suggestion for the fusion of WorldView-2 imagery. By image multi-scale segmentation trials, suitable segmentation parameters are found for the study area:the segmentation scale of 175 is suitable for extracting the water body; and the segmentation scale of 130 is suitable for extracting the forestland, farmland, vegetation shadows, roads, factory and bare land; the segmentation scale of 55 is suitable for extracting rural settlements and buildings shadow. So the landuse features can be extracted in different scales by the multi-scale segmentation, and the appropriate classification hierarchy can be established. By statistical analysis of the image spectral characteristic, shape and texture information and relationships among objects, the best features combination are selected for the land cover features, reasonable classification rules are constructed with nearest neighbor and member functions for extracting information and land cover mapping. Overall accuracy by object-oriented information extraction method is 88.29%, outperforms 12.58% than the maximum likelihood method, and the Kappa coefficient increases from 0.7167 to 0.8633. Among them, the classification accuracy of forestland, bare land and buildings has improved obviously. With a comparative analysis for the result of pixel-based classification method, object-oriented classification method is more suitable for information extraction from high spatial resolution images.