Object-oriented land cover information extraction in emigration area of Zigui county using high resolution imagery

Abstract

Zigui county is in the east of the reservoir area of the Three Gorges Project, where the Three Gorge Dam is located. Strong human activities and policy impact result in the type and structure of land cover have changed in great degree. Meanwhile, the characteristics of surface features in Zigui are very complicated because of fragmented surface condition and hilly topography. Thus, it is meaningful to get the latest land cover data for government decision and scientific research. Remote sensing technique provides us with a fast information extraction method of land cover data. High resolution remote sensing images have abundant spatial, geometric and texture information, which can be used for producing more accurate land cover data. However, traditional remote sensing information extraction methods emphasize on pixel-based classification. Those methods only make use of spectral characteristics of features, but ignore the spatial characteristics inherent within the high resolution images. In addition, it will very difficult to distinguish some features only using one-stage images. For example, the characteristics of paddy field and upland in study area are close in SPOT5 images (in winter), and the characteristics of paddy field and bare soil are close in SPOT5 images (in spring). Paddy field will be confused with bare soil or upland only in one-stage SPOT5 images. It results in the classification accuracy of paddy field in one-stage image is very low. In this study, object-oriented image analysis method is made use of to carry on land cover information extraction in emigration area of Zigui county. The advantage of two-stage of SPOT 5 images (spring and winter) and subsidiary data (DEM, Slope, etc) are fully taken into consideration. The paper introduces the method of image pre-processing, multi-resolution image segmentation, etc. It illustrates object-oriented image analysis method and its application in study area in detail. Accuracy assessment is performed after land cover information extraction using field samples. The result shows the accuracy was as high as 95%, and object-oriented image analysis and the use of two-stage remote sensing images have great potential for extracting land cover information from high resolution imagery.