基于GPU的分布式全息孔径数字成像技术研究

分布式全息孔径数字成像技术是利用数字全息记录各子孔径的复振幅信息,通过孔径间相位拼接实现综合成像的一种主动成像技术。在远距离成像中,大气湍流引入的子孔径内高阶相位误差和子孔径间低阶相位误差,以及孔径间的位置失配误差,都会影响成像质量。随机并行梯度下降算法(SPGD)是一种无波前探测优化控制算法,具有可以并行、快速收敛、高效可靠等优点,可用于校正系统孔径内高阶和孔径间低价相位误差。但是SPGD算法需要多次迭代,运算量巨大,难以满足实时性要求。文章基于GPU平台,对高、低阶相位误差校正进行了并行加速处理,运算速度较CPU平台分别提升26.42倍和36.47倍。此外,采用AKZAE算法校正各子孔径间的位置失配误差,完成了各子孔径复振幅的拼接,最终实现了分布式四孔径的综合成像。

Research on Distributed Holographic Aperture Digital Imaging Technology Based on GPU

Distributed holographic aperture digital imaging technology is an active imaging technology that uses digital holography to record the complex amplitude information of each sub-aperture, and realizes comprehensive imaging through phase stitching between apertures. In long-distance imaging, the high-order phase error in the subaperture introduced by atmospheric turbulence, the low-order phase error between the subapertures, and the position mismatch error between the apertures will affect the imaging quality. Stochastic parallel gradient descent (SPGD) is an optimal control algorithm without wavefront detection. With the advantages of parallelism, fast convergence, high efficiency and reliability, it can be used to correct high-order and low-cost phase errors within the aperture of the system. However, the SPGD algorithm requires multiple iterations and a huge amount of calculations, which is difficult to meet the real-time requirements. In this paper, parallel acceleration processing was performed based on the GPU platform for both high and low-order phase error correction, and the operation speed is 26.42 and 36.47 times higher than the CPU platform, respectively. In addition, the AKZAE algorithm was used to correct the position mismatch error between the sub-apertures and complete the splicing of the complex amplitudes of the sub-apertures, Finally, distributed four-aperture comprehensive imaging was realized.