| In this paper, Spot5 high-resolution satellite images as data sources, experimental research area is Changting County, Fujian Province. We studied the information extraction of Spot5 high-resolution images of Changting in 2010.First, based on the characteristics of satellite Spot5 remote sensing images, we conducted pre-treatments on the images, including atmospheric correction, geometric correction, image fusion, image stitching and tailoring.Then we conducted the multi-scale image segmentation. We determined the most appropriate segmentation scale and form factor, compact factor, and obtained appropriate extraction object of each information category. According to the various segmentation scales, we determined the classification level of each category.Finally, we got the classification results based on fuzzy classification method and nearest classification method, and analyzed the classification accuracy using the confusion matrix. We got the following conclusions:(1) When the multi-scale segmentation parameter being set to 150, rivers, roads and industrial buildings present as whole and homogeneous objects; when it was set to 50, buildings, farmland and woodland are ideal objects for segmentation.(2) During the segmentation parameter test, when the shape factor was set as 0.2 and compactness factor to 0.5, we got the most ideal segmented image objects.(3) The fuzzy rules of information extraction are:Normalized Difference Vegetation Index (NDVI)>= 0.06 & Brightness <89 for woodlands,65<Mean layer3<78.3 for water body,0.3<Density<1 for roads,-0.006<NDVI<0.09 for farmlands,1<Border index<1.82 & 0.03< Gray Level Co-Occurrence Matrix Homogeneity (GLCM Homogeneity)<0.13 for residence. The others categories were extracted using nearest classification method.(4) Finally, we evaluated the accuracy of the extraction results based on Changting County Land Data in Year 2011. The overall classification accuracy was 93.73% and the Kappa coefficient was 0.77. |
PDF Full Text Request