深度学习融合模型共轭涡旋光干涉微位移测量

打开文本图片集

中图分类号：0439 文献标识码：Adoi：10.37188/OPE.20253315.2331

CSTR：32169.14.OPE.20253315.2331

Conjugate vortex beam interferometry for micro-displacement measurement with deeplearning fusion architecture

YANG Xuejiao¹，LIU Ji1*，WU Jinhui²，YUAN Tao1，WANG Shijie 2 ，JI Xiangfeng1，YULixia¹， ZHANG Boyang1， CHEN Xiang？

（1. School of Information and Communication Engineering，North University of China， Taiyuan 030051，China; 2. School of Instrument and Electronics，North University of China， Taiyuan O3Oo51，China; 3.China Mobile Communications Group Shanxi Co.，LTD.，Taiyuan O3Oo5l，China） * Corresponding author， E -mail： liuji6@nuc. edu. cn

Abstract：To overcome the limitations of traditional fringe-based displacement inversion algorithms in vor tex interferometry for micro-displacement measurement，a deep-learning fusion model based on conjugate vortex beam interference is proposed. A YOLOv8s-Seg segmentation network，incorporating a lightweight FasterNet backbone and a CARAFE dynamic upsampling module，is employed to segment petal regions in interference images accurately，thereby reducing the influence of background noise and beam distortion on phase information extraction. A14-layer convolutional neural network （CNN） is then used to perform multi-scale hierarchical feature extraction on the segmented petal regions，establishing a precise mapping between morphological variations and rotation angles to enable sub-nanometer displacement detection. Experimental results within a displacement range of（O-5OO）nm demonstrate a petal-region segmentation mean average precision （mAP） of 96.5% ， overall displacement accuracy better than 0.94nm ， and a mean absolute error（MAE）of 0.63nm . Owing to the dual-network collaborative architecture，the proposed method exhibits improved robustnessto fringe distortion and noise，providing marked advantages in both precision and stability for micro-displacement measurement.

Key words ： micro-displacement measurement; conjugate vortex light interferometry; YOLOv8s-Seg seg mentation；multi-scale hierarchical feature extraction

1引言

激光干涉测量技术凭借非接触性、高分辨率和宽动态范围等显著优势，广泛应用于精密测量领域，并成为高精度测量任务的首选方法[1]。（剩余13482字）