强激光与粒子束2026，Vol.38Issue(4)：1-11,11.DOI:10.11884/HPLPB202638.250370
基于卷积神经网络的激光自混合干涉微位移重构
Laser self-mixing interference micro displacement reconstruction based on convolutional neural network
摘要
Abstract
[Background]Laser self-mixing interferometry(SMI)is a highly sensitive and non-contact technique widely used for micro-displacement measurement.However,traditional displacement reconstruction methods typically involve complex phase unwrapping calculations,which increases computational difficulty and limits the efficiency of signal processing in practical applications.[Purpose]This study aims to propose a novel micro-displacement reconstruction method for semiconductor laser SMI based on convolutional neural networks(CNN).The objective is to achieve direct and accurate reconstruction of micron-scale displacement while bypassing the tedious phase unwrapping process.[Methods]The proposed method involves segmenting the SMI signal and using the window-averaged displacement as the label for training the CNN.The architecture of the network consists of three sets of convolutional layers,pooling layers,and Rectified Linear Unit(ReLU)functions.Specifically,the convolutional layers are utilized to extract local displacement features from the SMI signal,the pooling layers are designed to compress feature information and enhance noise immunity,and the ReLU functions help highlight critical displacement features within the signal.[Results]In theoretical simulations,SMI signals with 10 dB noise were input into the trained CNN,resulting in a displacement reconstruction RMSE of 5.3×10-8.In experimental tests,SMI signals containing system noise were processed by the network,yielding a reconstructed displacement RMSE of 2.1×10-7.The simulation and experimental results demonstrate consistent performance.[Conclusions]Both theoretical and experimental results indicate that the convolutional neural network can effectively achieve micron-level displacement reconstruction by analyzing the temporal segments of SMI signals.This method provides an efficient alternative for semiconductor laser self-mixing interference systems by eliminating the need for complex phase-based algorithms.关键词
激光自混合干涉/微位移重构/卷积神经网络/特征提取/半导体激光器Key words
laser self-mixing interference/displacement reconstruction/convolutional neural network/feature extraction/semiconductor laser分类
信息技术与安全科学引用本文复制引用
李鑫涛,刘晖,乔硕,杨一帆,吕杨,刘霞,熊玲玲..基于卷积神经网络的激光自混合干涉微位移重构[J].强激光与粒子束,2026,38(4):1-11,11.基金项目
陕西省自然科学基础研究计划项目(2025JC-YBMS-770) （2025JC-YBMS-770）
陕西省秦创原"科学家+工程师"队伍项目(2023KXJ-129) （2023KXJ-129）
