The Simulation Of Satellite Imagery By Parallel Ray-tracing On SMP-CMP Cluster

Simulating the real image of the satellite is very improtant for setting the orbit parameters before launching it, at the same time, the imagery can be used in the ground system. In order to accurately simulate the satellite imagery, in this paper the polular methods in simulating the satellite imagery are compared, then the Ray-tracing method is be chosed . In order to ensure the accuracy of simulation result, in this paper the time integration, spatial integration and TDI integral methods are used to simulate the real principle of the satellite which carrys three linear-array CCD camera.In the simulation of satellite imagery, changing the serial method into parallel one can effectively reduces the computing time and improves the efficiency of calculation. The algorithm in this paper is running in the SMP-CMP cluster, which is the most popular architecture in the area of high-performance computing. According to the characteristics of the SMP-CMP cluster architecture and the algorithm used in this paper , the model which mixes the message passing and shared variable is chosed from the paraller models used on SMP-CMP cluster.In this model, proceses on different nodes communicate with each other through the MPI, and in each process the OpenMP programming method is used to implement the parallelization.The ground scene information is organized in DEM or DSM which are different in resolution and complexity. In this paper according to the grid point form of DEM, the uniform grids structure is used, and combined with the local characteristic of the data and ray to reduce the computing time in ray-tracing. On this basis, pixel-oriented parallel implementation is used to partition task to each MPI process, then the processes on different nodes finish its own task parallelly, and the ideal speed up that nearly linear is achieved. At the same time, according to the defect of poor of useing the pure MPI model in SMP-CMP cluster, the finer-grained mixed parallel model combined with OpenMP and MPI parallel method is used to improve the real capability, the experimental results show that the mixed model is better in the expansibility when there are many nodes in the cluster. According to the irregular triangles form of DSM, the kd-tree structure is used to organize the DSM scene, and on this basis its parallel implementation method is explored.