Information Extraction For Land Of Rural Areas Based On High-resolution Remote Sensing Image

With the spatial resolution of remote sensing images rising, the traditional technology based on pixel has been unable to properly extract Land Information, which has a very low accuracy,and the phenomena of " the same thing with different spectrums " and " different things with the same spectrum" cannot be distinguished.According to the characteristics of high resolution remote sensing images, the smallest operating unit for information extraction is object in this thesis. As the existing segmentation techniques for object has flaw of accuracy and low automation, segmentation technology using a multi-feature-based data fusion and incorporation of object is posed in this thesis. Before image data fusion, this method explore the edge features,spectral features,shape features and so on from the original images and then fuse them into an false color image, which retains the key features of the original image for reducing the original image data redundancy and enhancing the edge features, and then select the appropriate small-scale for object-oriented segmentation. Finally,An algorithm of image , based on color and spatial information, is used to merge the object of over-segmentation. The existing object segmentation technology create multi-layer network structure for segmentation, but the method of this thesis make various types of object into a separate layer with appropriate shape. Comparing with the two methods, the segmentation method of this thesis make the object is closer to the actual shape and size of land,which avoid the multi-scale segmentation selection in scales and improve the real-time of system, reduce operational complexity.After the segmentation of object, we not only consider the spectral difference between the regions, but also consider shape measures and texture measures as well to extract the object's features in this thesis, and then choose the optimum multi-object features to using the object-oriented classification for information extraction. Finally, Taking a typical region of rural in ChengDu District as the test area , a information extraction study on Quickbird 0.61 meter high-resolution remote sensing images is carried out at the end of this thesis. The accuracy evaluation showed that: the overall accuracy of land units was 92.6%, and that most accuracy of various land in the area is more than 90%. It is really show that this information extraction technology of this thesis can achieve high-resolution extract for land of rural areas.