Gaussian process machine learning-based surface extrapolation method for improvement of the edge effect in surface filtering

Liu, Ming Yu ; Cheung, Chi Fai; Feng, Xiaobing; Ho, Lai Ting ; Yang, Shu Ming

Abstract

Filtering for signal and data is an important technology to reduce and/or remove noise signal for further extraction of desired information. However, it is well known that significant distortions may occur in the boundary areas of the filtered data because there is no sufficient data to be processed. This drawback largely affects the accuracy of topographic measurements and characterizations of precision freeform surfaces, such as freeform optics. To address this issue, a Gaussian process machine learning-based method is presented for extrapolation of the measured surface to an extended measurement area with high accuracy prior to filtering the surface. With the extrapolated data, the edge distortion can be effectively reduced. The effectiveness of this method was evaluated using both simulated and experimental data. Successful implementation of the proposed method not only addresses the issue in surface filtering but also provides a promising solution for numerous applications involving filtering processes.

Citation

Liu, M. Y., Cheung, C. F., Feng, X., Ho, L. T., & Yang, S. M. (2019). Gaussian process machine learning-based surface extrapolation method for improvement of the edge effect in surface filtering. Measurement, 137, 214-224. https://doi.org/10.1016/j.measurement.2019.01.048

• Journal Article Type: Article
• Acceptance Date: Jan 12, 2019
• Online Publication Date: Jan 19, 2019
• Publication Date: 2019-04
• Deposit Date: Jul 1, 2020
• Publicly Available Date: Jul 13, 2020
• Journal: Measurement
• Print ISSN: 1536-6367
• Publisher: Taylor & Francis (Routledge)
• Peer Reviewed: Peer Reviewed
• Volume: 137
• Pages: 214-224
• DOI: https://doi.org/10.1016/j.measurement.2019.01.048
• Keywords: Instrumentation; Electrical and Electronic Engineering; Applied Mathematics; Condensed Matter Physics
• Public URL: https://nottingham-repository.worktribe.com/output/1487177
• Publisher URL: https://www.sciencedirect.com/science/article/pii/S0263224119300247?via%3Dihub