光学 精密工程, 2018, 26 (9): 2269, 网络出版: 2018-12-16
噪声动态光散射数据Tikhonov与截断奇异值正则化反演
Inversion of tikhonov and truncated singular value decomposition regularization for noisy dynamic light scattering data
动态光散射 Tikhonov正则化 截断奇异值正则化 颗粒粒径反演 light scattering Tikhonov regularization truncated singular value decomposition regularizat particle size inversion.
摘要
Tikhonov与截断奇异(TSVD)正则化是动态光散射数据反演中的两种重要方法, 不同的正则化方法会对噪声DLS数据测量结果产生不同的影响。分别采用二阶差分矩阵的Tikhonov与TSVD方法, 在6种噪声水平下, 对宽窄不同的单峰与双峰分布颗粒进行了反演研究。结果表明: Tikhonov具有较好的光滑性; 对于单峰分布颗粒, TSVD峰值误差更小、对于窄分布以及强噪声宽分布颗粒系反演, 其抗噪性能更强、反演误差更小; 对于双峰分布颗粒, Tikhonov具有较小的反演误差、较强的双峰分辨能力与抗噪声能力; 对于窄分布颗粒的反演, 一般TSVD峰值误差更小。在同样噪声情况下, Tikhonov与TSVD的双峰分辨力与颗粒的粒径峰值比有关。Tikhonov双峰分辨力较强, 能够分辨出峰值比较低的颗粒。对实测200 nm单峰颗粒进行反演, Tikhonov、TSVD的反演峰值误差分别为3%和1.85%, TSVD峰值位置更准确, 能够验证模拟数据的结论。
Abstract
Tikhonov and truncated singular value decomposition regularization (TSVD) are two important methods in the dynamic light-scattering (DLS) data inversion. Different regularization methods will have different effects on the results of noisy measurement data. By using the Tikhonov of second-order differential matrix and TSVD method, the unimodal and bimodal distribution particles with different widths were retrieved under six kinds of noise levels. The results show that Tikhonov had better smoothness, the peak value error of TSVD is smaller for unimodal distribution particles, anti-interference ability is stronger, and the inversion relative error is smaller for a narrow distribution and the strong noise wide distribution particle system. For bimodal distribution particles, Tikhonov has smaller inversion error and higher bimodal peak resolution and anti-interference ability. For the inversion of narrow distributed particles, the peak value error of TSVD is generally smaller. Under the same noise condition, the bimodal peak resolution of Tikhonov and TSVD is related to the peak value ratio of the particle size. Tikhonov has a high bimodal peak resolution, and it can distinguish particles with lower peak value ratio. The measured data are retrieved for unimodal distribution particles of size 200 nm. The inversion peak errors of Tikhonov and TSVD are 3% and 1.85%, respectively. The results verify the conclusion of the simulated data.
王雅静, 袁曦, 申晋, 窦震海, 孙贤明. 噪声动态光散射数据Tikhonov与截断奇异值正则化反演[J]. 光学 精密工程, 2018, 26(9): 2269. WANG Ya-jing, YUAN Xi, SHEN Jin, DOU Zhen-hai, SUN Xian-ming. Inversion of tikhonov and truncated singular value decomposition regularization for noisy dynamic light scattering data[J]. Optics and Precision Engineering, 2018, 26(9): 2269.