遥感影像正射纠正的GPU-CPU协同处理研究
GPU-CPU Cooperate Processing of RS Image Ortho-Rectification
-
摘要: 提出了一种基于CUDA的遥感影像正射纠正GPU-CPU协同处理方法,以实现重采样操作的GPU细粒度并行化。根据GPU的并行结构和硬件特点,采用执行配置优化技术提高warp占有率,利用共享存储器优化减少对效率低下的全局存储器中坐标变换系数的重复访问,通过纹理存储器代替全局存储器优化对原始影像数据的访问。实验结果表明,并行算法能够充分发挥GPU的并行处理能力,利用GeForce 9500 GT显卡,对大小为6 000像素×6 000像素的全色影像进行多项式纠正对比实验,最邻近灰度内插重采样和双线性灰度内插重采样的最终加速比分别能够达到8倍和10倍以上。Abstract: A fast ortho-rectification GPU-CPU cooperate processing algorithm is presented based on compute unified device architecture(CUDA),which realizes fine-grained parallel re-sampling using GPU in single instruction multiple thread(SIMT) pattern.On the basis of parallel architecture and hardware characteristic of GPU,the parallel algorithm introduces three speedup methods to improve the implementation performance: using execution configuration optimize technology to increase warp occupancy,using shared memory to reduce coordinate transform coefficients accessing times of low-speed global memory,using texture memory replace global memory to optimize the accessing of original image.The experiment result shows that using GeForce 9500 GT display-card to do multinomial rectification for gray image size of 6 000 pixel×6 000 pixel,the speed is more than 8 times and 10 times faster than CPU-based implementation each for nearest-neighbor interpolation and bilinear interpolation.
下载:
