复杂轨迹合成孔径雷达后向投影算法图像流GPU成像

摘要

Compared with the frequency domain algorithms based on Fourier transform, back projection (BP) algorithm is more applicable for high-precision synthetic aperture radar(SAR) imaging with com-plex trajectory because of its time-domain coherent integration point by point. However,BP algorithm has a problem of huge calculation quantity,which restricts its application to fast SAR imaging with large scene and big data. Graphics processing unit( GPU) has a strong ability in floating-point calculation and parallel processing,and provides an effective approach for large-scene fast imaging with BP algorithm. According to GPU parallel processing, this paper proposes a new processing method of large-scene SAR fast imaging with BP algorithm,based on complex motion of image flow. Depending on the character that the image pix-els in BP algorithm are processed independently,this method uses the parallelization of image pixels and image-flow processing to design a scheduling scheme between apertures and image caches to improve the imaging efficiency of SAR with large scene and big data. Simulation and airborne experimental results dem-onstrate the effectiveness of the method,and fast imaging with the large scene of 8 192 ×8 192 has been re-alized in the condition of limited GPU memory,and the speed-up ratio can be improved 300 times com-pared with the traditional CPU single-thread processing.%相对于基于傅里叶变换的频域成像算法，后向投影( BP)算法因采用时域逐点相干积累，更适合于复杂轨迹合成孔径雷达( SAR)高精度成像。但BP算法计算量巨大，限制了其应用于SAR大场景大数据量快速成像。图形处理器( GPU)具有强大浮点运算和并行处理能力，为大场景BP算法快速成像实现提供了途径。结合GPU并行处理，提出了一种基于图像流的复杂运动SAR大场景BP快速成像处理方法。该方法借助BP算法中图像像素点相互独立处理的特性，采用图像像素点并行及图像流程处理，设计了孔径与图像缓存调度方案，提高SAR大场景大数据BP算法成像效率。仿真和机载实测数据结果验证了方法的有效性，在有限GPU显存条件下实现了8192×8192大场景快速成像，并且成像加速比相对于传统CPU单线程处理可达300倍以上。