Deterministic-teacher-supervised fringe-width regression for compact dual-channel line-array conoscopic holography displacement measurement

作者：Zhuo Chen, Xiaobin Lan, Liuwei Zheng, Zhanyang Xiang, Hong Zhang · 发表于：Measurement Science and Technology · 年份：2026 · DOI：10.1088/1361-6501/ae9630 · 研究领域：Digital Holography and Microscopy、Optical measurement and interference techniques、Photorefractive and Nonlinear Optics

Abstract Conoscopic holography provides a coaxial and non-contact sensing mechanism for long-working-distance linear displacement measurement. In the linear-fringe, dual-channel line-array implementation considered in this work, the detector output is not a displacement value directly. Instead, the displacement information is encoded in two polarization-separated fringe waveforms recorded by line-array detector channels, so the measurement accuracy depends strongly on fringe-width demodulation, calibration and signal robustness. In industrial acquisition conditions, deterministic fast Fourier transform (FFT)-based demodulation remains interpretable but can be affected by channel misalignment, spectral peak uncertainty and repeated signal-processing operations, whereas direct black-box displacement regression lacks metrological traceability.

This paper presents a deterministic-teacher-supervised fringe-width regression framework for dual-channel line-array conoscopic holography. In this framework, the teacher is not a neural network but a fixed Period-Constrained Cross-Correlation (PCCC) and FFT demodulation pipeline that provides continuous fringe-width labels in pixel units. A compact one-dimensional convolutional neural network then estimates the fringe width from a 2 × 2048 dual-channel waveform, and the final displacement is reconstructed through an empirical calibration model. The audited dataset contains 500 cleaned file-level records and 4310 valid wav...