GPU并行计算在FD-OCT成像中加速插值计算的应用

Accelerated Application of GPU Parallel Computing in Interpolation Calculation in FD-OCT Imaging

FD-OCT(频域光学相干层析成像技术)成像中需要对相机采集到的光谱信号进行插值从而获得干涉信号,插值计算是FD-OCT成像重建的重要步骤,占据了三分之二的计算时间.因而如何加快插值计算的速度,是FD-OCT实时成像的主要问题.本文将基于GPU的CUDA语言引入FD-OCT系统成像的三次样条插值的计算过程中,把整个插值的计算过程并行化,并对其进行优化.从实验结果中可以看出,执行相同的数据处理任务,将GPU作为主要并行计算处理器,与传统基于CPU平台的串行计算模式比较,速度有大幅度的提高,可实现每秒52帧的实时成像.

In FD-OCT (frequency domain optical coherence tomography) imaging, the spectral signals collected by the camera need to be interpolated to obtain the interference signal. Interpolation calculation is an important step in FD-OCT imaging reconstruction, accounting for two-thirds of the calculation. time. Therefore, how to speed up the interpolation calculation is the main problem of FD-OCT real-time imaging. In this paper, the GPU-based CUDA language is introduced into the cubic spline interpolation process of FD-OCT system imaging, and the whole interpolation process is parallelized and optimized. It can be seen from the experimental results that the same data processing task is performed, and the GPU is used as the main parallel computing processor. Compared with the traditional serial computing mode based on the CPU platform, the speed is greatly improved, and 52 frames per second can be realized. Real-time imaging.