Scholarlink

<p>White Light Interferometry (WLI) is among the most advanced non-contact measurement technologies currently available, capable of capturing surface morphology and roughness with nanometer-level precision. This technique is particularly well-suited for micro-structural measurements and is widely applied across various fields. One of the major challenges in applying WLI lies in the reconstruction of three-dimensional surface profiles from multi-frame interferometric images, which are often significantly affected by noise, optical distortions, and the intricate features of the measured samples. In this study, we propose an innovative approach that departs from traditional methods relying on single-frame processing. Instead, we utilize multiple frames captured at the same scan position and apply deep learning techniques to train a model that enhances the quality of interference fringe images under noisy conditions. This leads to improved accuracy in 3D surface reconstruction. An experimental setup was developed to validate the approach, with results demonstrated on surfaces machined using a diamond turning process.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing the quality of three-dimensional reconstruction from white-light interferometry based on deep learning

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Hai Le Hoang,
  • Nhu Le Van,
  • Thong Nguyen Doan

摘要

White Light Interferometry (WLI) is among the most advanced non-contact measurement technologies currently available, capable of capturing surface morphology and roughness with nanometer-level precision. This technique is particularly well-suited for micro-structural measurements and is widely applied across various fields. One of the major challenges in applying WLI lies in the reconstruction of three-dimensional surface profiles from multi-frame interferometric images, which are often significantly affected by noise, optical distortions, and the intricate features of the measured samples. In this study, we propose an innovative approach that departs from traditional methods relying on single-frame processing. Instead, we utilize multiple frames captured at the same scan position and apply deep learning techniques to train a model that enhances the quality of interference fringe images under noisy conditions. This leads to improved accuracy in 3D surface reconstruction. An experimental setup was developed to validate the approach, with results demonstrated on surfaces machined using a diamond turning process.