Research Of Remote Sensing Image Parallel Segmentation Method Based On The New Technology

With the rapid development of space technology、information technology and sensor technology,thespatial resolution and time resolution of satellite remote sensing image improve greatly, so the remotesensing data quantity and quantity of the processing will increase one hundred times or even one thousandtimes. How to improve the mass of remote sensing data processing speed has been always one of theresearch content of remote sensing data processing. At the same time with the rapid development thecomputer parallel computing, the GPU computing power and memory bandwidth have been more than themainstream CPU, and the CPU multinuclear programming also become one of the ways to improve theprocessing power. The combination of computer parallel computing technology and mass remote sensingdata processing has been more and more close. For solving the remote sensing image data parallelcomputing processing and rapid display,the Institute of Remote Sensing Applications Chinese Academy ofSciences designed a remote sensing image classification organization mode called five layers fifteen levels.Therefore, base on the five layers fifteen levels,this article realizes image segmentation and the parallelcomputing acceleration based on CUDA platform, as well as the multinuclear platform, which greatlyimproves the speed of remote sensing image segmentation base on the five layers fifteen levels, and thishas important theoretical significance and practical significance.The main work in the following areas:Firstly, realizing the remote sensing image segmentation based on five layers fifteen levels. Fivelayers fifteen levels is a kind of remote sensing image organization way proposed by the Inistitute ofRemote Sensing Applications Chinese Academy of Sciences,this article researches the characteristics ofthe five layers fifteen levels, according to its characteristics to analysis and compare with the remotesensing image,and we get the relationship between the five layers fifteen levels and the remote sensingimage, then we realize the segmentation method of remote sensing images based on five layers fifteenlevels.Secondly, using CUDA technology and multinuclear technology to accelerate and optimize thesegmentation method.In the realization of segmentation method, we analysis the calculation characteristics and the processing time in the process of segmentation and organization, discussing the parallel processingof the right way。Through the GPU programming and multinuclear programming to optimize the program,to improve the processing speed、reduce the processing time and use computer resources reasonably.Thirdly, through the use of Multi-GPU and address mapping technology of CUDA technology tooptimize the segmentation procedure. According to the future development trend of Multi-GPU,weresearch how to allocate the task of segmentation and the task scheduling,not only to make thesegmentation procedure adapt to Multi-GPU environment,which can get a greater degree of efficiency, butalso to provide the design work in one computer with Multi-GPU or clusters with Multi-GPU design workfor reference in future. At the same time we use address mapping technology and asynchronous executiontechnology in GPU programming to realize the procedure, analysising the advantages and disadvantagesby comparing the different methods’ results.