The Research On The Technologies For The Analysis Of Remote Sensing Image Based On The Saliency

Under the constant advancement of remote sensing technology, more and more high spatial and spectral resolution and high time frequency data are becoming available, which not only greatly increase our capability and accuracy in information extraction and monitoring but also open new application opportunities. But the"three high"characters make the data size of the remote sensing image huger and huger. In this situation, we think there are two problems should be considered, which are how to gain the valuable remote sensing data and how to process the remote sensing data efficiently.In this thesis, a frame for gaining and analyzing remote sensing image based on saliency is proposed. In which, building area is extracted, building in high-resolution remote sensing image is detected, and contour of the building is constructed. The analysis mentioned above will help us to recognize remote sensing image objects and to build an image database and a corresponding relevant knowledge base to improve data sharing and facilitate investigations in this field.The main contributions in this thesis are as follows:(1) A frame for gaining and analyzing remote sensing image based on saliency is proposed, which is the architecture of analysis remote sensing image under the direction of the knowledge. It consists of two parts that is top-to-down and bottom-to-up. Bottom-to-up consider that what feature be abstracted, while top-to-down consider that how to gain knowledge that help to auto-interpretation of remote sensing image.(2) An algorithm for extracting building area from remote sensing images based on wavelets transformation and mathematical morphology is proposed. In the remote sensing images, different ground form has different spectrum feature. The frequency of building area is higher than that of other area. First, wavelets transformation is applied on a remote sensing image to transform different information of ground form into different frequency band. Then, the high frequency information is merged together and the area where the amplitude changes rapidly in the high frequency information merged is regarded as a peak, while the area where the amplitude changes smoothly is regarded as a basin. Finally, the building area is extracted accurately and quickly based on morphological reconstruction.(3) A method for detecting buildings based on the information of their shadow in high-resolution remote sensing images is proposed. Get the threshold based on statistical approach, apply Susan edge detector on the shadow area, and then locate the building based on the perpendicular edge.(4) How to rebuild the contour of the detected building is discussed in this thesis. A multi-scale edge detecting algorithm is applied to building image. If a point whose value in every scale edge image equals 1 which named"seed". First, the directions of the edge on the seeds are calculated. Then they are clustered based on the direction and the distance. Finally, a line is found to stand for a cluster. The direction of the edge is corrected based on priori information to create the contour of the building.Finally, a prototype system is designed and implemented and some suggestion about our research is proposed at the end of this thesis.