Dissertation > Excellent graduate degree dissertation topics show

Study on Application of Parallel Computing to Seismic Damage Analysis from Remote Sensing Images

Author: ZhengYouHua
Tutor: WangXiaoQing
School: China Seismological Bureau, Institute of Earthquake Prediction
Course: Solid Geophysics
Keywords: Parallel computing Remote sensing Damage Analysis Earthquake Change detection
CLC: P315.61
Type: Master's thesis
Year: 2010
Downloads: 157
Quote: 0
Read: Download Dissertation

Abstract

An earthquake is a sudden natural disaster, brought great disaster to human society and civilization. Since the earthquake prediction is a worldwide scientific challenges before the earthquake defense and relief work after the earthquake, is currently the most effective way to reduce the loss of earthquake disaster. Remote sensing is a technology developed for Earth Observation in the 1960s, has unique advantages compared to the traditional means of observation: observation range, comprehensive, macro; images of large amount of information, technical means more multi-faceted and all-weather observation capabilities; fast access to information, the update cycle is short. Earthquake disaster information to obtain a long time, mainly rely on the artificial field survey, there is a heavy workload, low efficiency, and higher cost of inadequacies. Using remote sensing technology can quickly access the macro after earthquake disaster information quickly survey and loss assessment for disaster, emergency rescue and provide an important basis for the recovery and reconstruction. With the development of remote sensing technology, remote sensing image resolution was gradually increased, the amount of data along with the rapid growth, and the stand-alone processing capacity can not meet the massive remote sensing image data fast processing requirements. Remote sensing image processing speed in large part become a bottleneck restricting the application of remote sensing technology, remote sensing image processing technology has brought new challenges. In this thesis parallel computing technology in the mass remote sensing image data processing and seismic damage extract as a research topic, in order to greatly improve the speed and accuracy of processing of remote sensing Damage Analysis. This paper introduces the remote sensing technology in earthquake disaster information extracted from the status quo, the related concepts of parallel computing, parallel computer course of development, today's mainstream parallel programming environment overview, as well as the development of parallel computing technology trend - CPU GPU exclusive the constitutive cluster as well as MPI, OpenMP, CUDA and OpenCL hybrid parallel programming environment. On this basis, the analysis of the status quo of parallel computing technology and applications in remote sensing image processing; narrative Beowulf cluster based on Windows and MPI build process; correlation coefficient method and the ratio method commonly used change detection method as remote sensing, earthquake damage information The extraction method, the constructed clusters based on the main mode parallel algorithm, and using Wenchuan 8.0 earthquake in Beichuan county area earthquake before and after the earthquake remote sensing data conducted a change detection parallel calculation experiment, through the parallel computation of acceleration than and operational efficiency analysis, the effect of parallel computing. Based on the above study, this paper the following conclusions: (1) the correlation coefficient change detection method has good correspondence between the size of the correlation coefficient and Disaster of. The correlation coefficient smaller area Damage serious contrary Damage lighter. Similarly, the results of the calculation of the ratio method change detection has good correspondence between the results from the visual interpretation method. These results illustrate the parallel processing of the reliability of the method given herein and feasibility. (2) the correlation coefficient (9 × 9 window, for example) the speedup and efficiency of parallel computing analysis results show that the number of processes start each computing node in parallel computing is not greater than the number of processor cores on the node when using parallel computing correlation coefficient method to achieve better performance; impact cases do not consider non-calculation process, the speedup is more than 0.9 times the number of processes and efficiency are maintained at more than 90%; principle calculate the number of processes consistent with the number of the processor core node, the node speedup and efficiency is optimal; node to calculate the number of processes exceeds the number of processor cores, the speedup and efficiency of the node it will decline. (3) The the parallel computing speedup ratio method and efficiency analysis showed that, when the only consider node calculation process, parallel computing still has good speedup and computational efficiency. However, due to the smaller amount of the calculation of the ratio method, a higher proportion of data reading and writing and communication time, resulting in wait for the main process of receiving results from the process or wait for the main process sends pending data block, resulting in overall speedup and efficiency not too idealistic. This related with this paper, the main mode of parallel algorithms, the mode applies to the case of the large amount of computation, the computation of the ratio method is relatively small, the master-slave mode ratio calculations hard to realize the advantages of parallel computing. (4) analysis of speedup and efficiency under different window size of the correlation coefficient method, under the same circumstances, the window larger speedup and efficiency as possible. This is due to the different window, non-calculating process (mainly data access, the time variation of the consumption in the implementation of the data partitioning and data communication) is not large, but a large window have a larger increase in the proportion of the total running time of the computation time . Seen the master-slave mode is more suitable in the case with a large amount of computation. Remote sensing earthquake damage information extraction parallelized computational experiments, there are some problems with the lack of: (1) Feature Change after the earthquake on the remote sensing image, so that before the earthquake after earthquake image registration accuracy difficult to achieve sub-pixel level, and thus have an impact on the results of change detection. But does not affect the results of this paper parallel algorithms for performance analysis. (2) In this paper, the data partitioning strategy and program data buffer size fixed, the number of processes involved in the calculation did not take into account some extent caused by load balancing between nodes is not very satisfactory, thus affecting the performance of parallel computing improved. (3) In this paper, the main design patterns and inter-node process allocation method in the calculation of the amount is relatively small, leading to relatively large proportion of data access and communication time, which lead to the decline of the parallel computing performance. (4) The correlation coefficient of serial algorithm, a preliminary optimization space parallel program there can be optimized by further study of algorithms and programs, is expected to get practical remote sensing the earthquake damage change detection parallel computing program . In summary, the results of the work of this thesis has achieved the expected goals, is expected to further research and practice in the application of the results.

Related Dissertations

CLC: > Astronomy,Earth Sciences > Geophysics > Earth ( rock circles ) physics ( geophysics ) > Seismology > Seismometry > Earthquake observation techniques and methods
© 2012 www.DissertationTopic.Net Mobile