| The automatic extraction of roads from high-resolution remote sensing images is of great significance for disaster emergency response,military operations,and real-time updates of GIS road network data.However,due to the large amount of remote sensing image data,and most road extraction algorithms based on remote sensing images are complex and computationally intensive.As a result,road extraction methods based on remote sensing images are relatively slow in processing speed,while greatly limits its practical application.In order to solve the problem of low processing efficiency of road extraction based on remote sensing imagery,this paper makes use of high-performance computing methods such as GPU and cluster computing to parallelize road extraction related algorithms,realize fast road extraction processing based on remote sensing images,and develop a Practical value of road extraction software system.The research work of this paper is as follows:(1)The road extraction algorithms are discussed in terms of road extraction,road feature analysis and road vectorization.In road extraction algorithm,a road extraction algorithm based on fuzzy C-means is implemented.In the aspect of road width estimation,the algorithm of stroke width analysis in character recognition is used to realize a road width estimation algorithm based on edge gradient.In the aspect of road vectorization,the vectorization method based on feature point tracking is studied.The cubic B-spline curve fitting method is used to optimize the vectorization results.(2)Parallel analysis of the processing algorithms for road extraction,road width estimation,and road vectorization.Based on the CUDA architecture,GPU parallel optimization design is performed for three algorithms,which improves the road extraction algorithm execution for individual computer nodes.effectiveness.(3)Aiming at the disadvantages of GPU parallel computing while requires redesign of the algorithm,a generic parallel processing model for road extraction is designed based on cluster computing.The specific implementation of the model was studied from the aspects of data partitioning and task assignment.In view of the weak disaster tolerance of the MPI cluster model,a task recovery mechanism was designed to improve the disaster recovery capability of the processing model.(4)Set up a high-performance road extraction system based on cluster parallel processing model.Develop a system client for user interaction and integrate the cluster parallel processing model in the software system to perform fast parallel processing of tasks submitted by the client.The system performance was tested and analyzed through experiments.From the experimental results,the system showed better stability and disaster tolerance.Compared to the road extraction process based on CPU single node,the system achieved a great deal of efficiency in processing.With a large upgrade,the speedup ratio is 3 times faster in the CPU cluster processing mode and 15 times faster in the GPU cluster processing mode. |
PDF Full Text Request