GPU支持的宽幅遥感影像RPC校正并行处理

GPU Supported Parallel Processing for RPC Geometric Correction of Wide-range Remote Sensing Imagery

RPC校正是光学遥感影像生产流程中重要的环节,将遥感影像从图像坐标转换到地理坐标。为缩短RPC校正过程的时间,实现影像的近实时处理,使用当前性能一流的GPU(NVIDIA TITAN RTX)实现了宽幅遥感影像的并行RPC校正,GPU为每个需要计算的像素分配线程,完成坐标变换、重采样等计算过程,实现并行处理;并与CPU运算RPC校正过程的效率作对比。实验结果显示,与CPU处理不分块时的RPC校正相比,GPU并行的加速比与遥感影像的数据量有关,数据量越大,则并行加速比越高;并行处理在高分二号多光谱影像和全色影像上的运算加速比分别为3.71,8.27,整体加速比分别为1.59,5.07。该方法能够显著加快遥感影像RPC校正的速率。

RPC geometric correction is an important part of the optical remote sensing image production process,which converts remote sensing images from image coordinates to geographic coordinates.In order to shorten the RPC correction process and realize the near real-time processing of the image,the RPC geometric correction is realized by using parallel processing of wide-range remote sensing images with advanced GPU(NVIDIA TITAN RTX).The GPU allocates threads for each pixel that needs to be calculated,completes the calculation process of coordinate transformation,resampling,etc.,and realizes parallel processing.The efficiencies of RPC geometric correction using CPU and GPU are compared.Experimental results show that,as compared with RPC correction using CPU processing when the image is not divided into blocks,the acceleration of GPU parallel processing is related to the amount of remote sensing image data.The larger the amount of data,the higher the acceleration.The ratios of computation acceleration on spectral and panchromatic images are 3.71 and 8.27 respectively,and the overall acceleration ratios are 1.59 and 5.07 respectively.This method can significantly speed up the RPC correction of remote sensing images.