Optical Metrology with Deep Learning

Dear Colleagues,

Optical metrology utilizes the fundamental properties of light as standards or information carriers to realize high accuracy and non-contact measurements. Traditional optical metrology generally retrieves measurement results from temporal varying intensity signals or images based on physical models. However, due to the lack of straight forward models, error and/or noise accumulation and the ill-posed inverse problems, challenges appear, and new approaches have been introduced to optical metrology. Deep learning (DL) adopting deep neural network (DNN) is an active branch of artificial intelligence and has been applied to various engineering including optical metrology. It has succeeded in pre-processing including image enhancement and denoising, data analyses including phase retrieval, phase unwrapping, error compensation, and postprocessing including three-dimensional reconstruction et al.

This Special Issue on “Optical Metrology with Deep Learning” will welcome basic, methodological and applied cutting-edge research contributions, as regular and review papers, dealing with:

• Introduction and validation of new DL method or DNN to traditional optical metrology;
• Development or comparison research of current DL methods in optical metrology;
• Physics-informed DL methods for optical metrology;
• Metrology system design with DL guidance;
• Measurement data generation or acquisition for DNN training;
• Improvement of the generalization ability of DNNs in optical metrology;
• Interpretation of the DL method and traceability analysis of the results;
• Uncertainty estimation for measurements with DL methods.

Dr. Yao Hu
Guest Editor

Manuscript Submission Information

Manuscripts should be submitted online at www.mdpi.com by registering and logging in to this website. Once you are registered, click here to go to the submission form. Manuscripts can be submitted until the deadline. All submissions that pass pre-check are peer-reviewed. Accepted papers will be published continuously in the journal (as soon as accepted) and will be listed together on the special issue website. Research articles, review articles as well as short communications are invited. For planned papers, a title and short abstract (about 100 words) can be sent to the Editorial Office for announcement on this website.

Submitted manuscripts should not have been published previously, nor be under consideration for publication elsewhere (except conference proceedings papers). All manuscripts are thoroughly refereed through a single-blind peer-review process. A guide for authors and other relevant information for submission of manuscripts is available on the Instructions for Authors page. Photonics is an international peer-reviewed open access monthly journal published by MDPI.

Please visit the Instructions for Authors page before submitting a manuscript. The Article Processing Charge (APC) for publication in this open access journal is 2400 CHF (Swiss Francs). Submitted papers should be well formatted and use good English. Authors may use MDPI's English editing service prior to publication or during author revisions.

• optical metrology
• physics-informed deep learning
• metrology system design
• deep neural network training
• generalization ability of deep learning
• interpretation of deep learning
• uncertainty estimation

Title: Utility of an Image-based Generational Adversarial Network for Two-dimensional Phase Unwrapping
Authors: Caelan Stephens; Steve Marsh; Juergen Meyer
Affiliation: 1. University of Canterbury, Christchurch, NZ 2. University of Washington, Seattle, USA
Abstract: 2D Phase unwrapping is a common problem in interferometry and medical imaging applications. Existing methods of performing 2D phase unwrapping lack robustness in the presence of noise and more sophisticated approached can be complex and time consuming. We investigated the feasibility of utilizing an image-based generational adversarial network-based 2D phase unwrapper. The performance of the Artificial Intelligence (AI) based unwrapper was characterized and compared with (several standard methods) in terms of accuracy and time. It was found that the AI equaled or outperformed all the tested standard methods. The AI maintained accuracy even in the presence of significant noise, avoiding many of the image artefacts present in the outputs of standard methods. In summary, robust and efficient AI-based 2D-phase unwrapping is feasible.