Information Extraction Technology Research Based On GF-1 Image Of The East Dongting Lake Wetland

In this paper, GF-1 images were used as the data source and research area is located in the East Dongting Lake area in the northern part of Hunan Province. Based on land use and land cover types, the spectrum characteristics of images were analyzed for a proper water index. In order to obtain an effective method for texture extraction of images, the comparison of the classification results were carried out between the original images and images added the local binary operator of three different radius. Furthermore, spatial and temporal distribution regularities of wetland resource was studied by analyzing feature extraction results of different phases with transition matrix. The research was aimed to explore an effective method of information extraction based on high spatial resolution remote sensing images, to analyze spatial and temporal distribution regularities of wetland resource, and also to provide theoretical and methodological reference for wetland protection and restoration. The main points and conclusions were as following:(1) The first band of GF-1 is blue which is particularly sensitive to water quality. It is not only a chlorophyll absorption band of green plants, but also could be used to distinguish between soil and vegetation. Band calculation and analysis showed that by replacing green band with blue band, water could be well-classified by subtracting SAVI(soil adjusted vegetation index) from NDWI(normalized difference water index) with the accuracy of lake classification in June reached 96.97%.(2) The method of LBP(Combine Local Binary Pattern) combined with GLCM(Gray-level Co-occurrence Matrix) was used to improve classification accuracy of texture extraction. In small-scale experiments, the accuracy of classification reached as the highest as 86.58% when the radius of LBP was 3 pixels, which is better than the method of gray level co-occurrence matrix. So in the method, LBP transform radius is one of the most important parameters, and it was found out that it exist the correlations between the radius and spatial resolution combined with the actual size of classification features.(3) Using the above methods to apply in high resolution image information extraction, classification accuracy of large-sized features, such as rivers, lakes, marsh lands, cultivated lands, Mud flats, and bare lands etc., was close to or higher than 90%. While classification accuracy of small-sized features, such as construction lands, forest lands and farm ponds, was about 80%.(4) Transition matrix method was used to analyze the classification of land use and land cover types in January and June, which realize to extract comprehensive information based on different time phases. Comprehensive information presented variable characteristics of wetland resources in spatial distribution, which benefit not only to reveal the spatial and temporal regularity of wetland, but also to provide decision support for wetland utilization and management.