| Due to the limitation of the current earthquake science, Scientists are not able to predict all destructive earthquakes accurately. Therefore, prevention is of great importance for reduction of seismic hazards. Many cases show that proper emergency responses can mitigate disasters, in particular reducing life losses. With rapid development of remote sensing(RS) technique, there are increasing available RS data of high resolution which can be applied to investigation of seismic damage, fast evaluation of seismic damage and automatic inspection of buildings. RS is characterized by large coverage, little effect of ground, and convenient acquisition of images. It plays an important role in post-earthquake management. In the past, the visual differentiation is the major means to recognize and interpret objects from RS data. This method has some shortages, such as criteria are not easy to learn, distinguishing results depend on workers, accuracies are poor, time consuming is large and efficiency is low. Thanks to the progresses in image processing and computer technique, there has been major improvement in theme information extraction and digital image recognition. It is possible to find an effective method of pattern recognition for fast estimation of seismic hazards in advance. It will provide evidence for decision making of emergency of the government and help reduce losses caused by earthquakes.Fast expansion of cities brings great demands for information of basal mapping of buildings. In this thesis I propose a method to extract information of three-dimensional attributes on spatial position and height of buildings. It uses the cast shadows and image features of constructions from single RS images. The cast shadow can reflect 3D information of objects on the ground, and play an important role in improving the accuracy and reliability of building detection. A algorithm is designed to find buildings' boundary points using the uniform gray degrees of roofs and determine further positions of these boundaries. By analyzing the relationships between shadow shape and sun elevation, sun azimuth, satellite elevation, satellite azimuth, as well as the constraints of building strike on shadow shape. I suggest a calculation formula for the buildings' height based on the shadow width in the vertical line of the building strike. This work uses segmentation to detect shadows, and calculates the shadow widths by the statistical averaging method, then determines the strikes of buildings based on shadows. In this way operation of searching basic points of buildings on RS images is not necessary. Meanwhile the accuracy of calculation is raised. It is simple and easy to get 3D information of buildings by using this method, and difficulty in traditional image matching is avoided. I employ this method to make recognition of buildings and their attributes on 5 towns of the Baoshan district of Shanghai from high-resolution images, and generate the vector map automatically. 41 random samples of buildings are investigated for test on site. Comparison between the calculated results and the actual data shows that the standard deviation error of buildings' height is 2.77 m, and that of the area is 68 m2. it indicates that thismethod is feasible. Building inspection based on RS images provides a great number of basic data for earthquake prevention and disaster reduction in cities. In conjunction with other data such as population and economy, I make a pre-estimation study of seismic hazards for the Shanghai area. A subsystem of pre-estimation for seismic damage is developed to accommodate the processing from basic data to result output.In the thesis, a algorism study for seismic hazard recognition is performed using single images after earthquakes. Considering the image features of underdeveloped rural regions in winter and regional properties that can be distinguished by object differences on the ground, I propose a layered recognition way for seismic damage by combining neighbor statistics of pixels and regional statistics of objects. Using the...
|
PDF Full Text Request