The Technology And Realize Method Of RS Image Parallel Process Base On GPU

This dissertation introduces a RS image GPU parallel process mode, based on the stringency demand of parallel image process and GPU's parallel process ability, which can be seen as a Single Instruction and Multiple Datastream (SIMD) parallel stream processor. The followings are what I have done in this paper.1. We first discussed the stringency; realize possibility and research status of RS Image Parallel Process Field, summed up the parallel modes of image processing and typical system, dissertated the development history, system framework, program model and the main application fields of GPGPU (General Purpose of GPU).2. On the basis of analysis of GPU's parallel process characteristic, we introduced a basic GPU parallel process flow, designed and realized a framework of GPU image process, and studied at length the key technologies; optimize measures and performance estimate also.3. Considering the characteristic of GPU and the complexity of image process, we studied the GPU parallel processing of pixel level arithmetic. Designed and realized the parallel process of some typical operation, such as: supervise classify, convolve operation, geometry transform and disperse wavelet transform etc, to validate the strong parallel compute ability of GPU.4. Designed and realized a prototype system of RS image parallel processing based on GPU, which can be called: GPUParImage. Then gived the main function and corresponding running effect image of prototype system.We can make a conclusion that: GPU has very obvious advantages in process pixel level data parallel problems, especially for high compute density one, such as: supervise classify etc. But limited the specially program model of GPU, it is very difficult even impossible to realize some complex image process.