并行化叠层成像算法研究

Ptychographical algorithm of the parallel scheme

叠层成像是定量相位恢复技术的重要研究方向,它通过照明探针的交叠式扫描,使用叠层迭代相位恢复算法对待测样品进行恢复,但成像效率与成像质量之间的矛盾等问题己成为其瓶颈之一本文从叠层成像迭代恢复算法的基本原理入手,提出了基于CPU和GPU的两种分块复振幅重建并行算法,并通过模拟实验研究了不同待测样品尺寸、不同分块、不同孔径数目对并行加速效果的影响模拟实验结果表明:两种并行算法可正确地恢复出样品的复振幅信息,并且显著提升了重建速度,使得重建耗时比传统叠层成像算法有了数量级的下降,在一定程度上解决了成像效率与成像质量之间的矛盾,有望实现准实时成像,为叠层成像在相关领域更广泛的应用提供了一定的技术指导,实验结果同时表明:在最优分块时,并行重建加速比与样品的大小有关,样品越大,加速效果越明显;同一个样品在不同分块下重建会得到不同的加速比,这与硬件设备密切相关,而成像中孔径的数目不会对并行加速比产生明显的影响.

Phychography is an important technique in the quantitative phase imaging research domain, which employs the illuminating probes to scan the specimen in an overlapped requirement, and the reconstruction is conducted by using the ptychographic iterative engine. But the contradiction between the imaging e？ciency and quality has become a bottleneck for its wide applications. In this paper, we start with the fundamental principle of the iterative algorithms for ptychographical imaging, and propose two parallel schemes based on CPU and GPU, besides the influences of the specimen size, the number of blocks and illuminating beams on the speedup of the two schemes are investigated via simulation experiment. The result shows that the complex amplitude of the specimen can be correctly reconstructed, meanwhile, the speed is significantly improved, which reduces the time consumed by one order of magnitude. This improvement solves the above contradiction, so that we can expect to achieve quasi-real-time imaging. The experimental data also indicate that 1） in optimal partition, parallel speedup is related to the size of the specimen, bigger size is corresponding to more obvious acceleration; 2） the same specimen under different partitions will speed up to different extents, which is closely related to the experimental hardware, however the number of illuminating beams has no significant effect on the speedup.