传感技术学报2025，Vol.38Issue(5)：809-816,8.DOI:10.3969/j.issn.1004-1699.2025.05.007
基于BP-Adaboost.M2的布里渊光时域分析传感系统温度提取技术研究
Temperature Extraction in Brillouin Optical Time Domain Analysis Sensors Based on BP-Adaboost.M2
摘要
Abstract
A novel method for extracting temperature from Brillouin gain spectrum(BGS)in Brillouin optical time domain analysis(BOT-DA)sensors is proposed,which relies on BP-Adaboost.M2,treating temperature extraction as a supervised single-label multi-classifica-tion problem.The feature dataset for training comprises BGS with Gaussian white noise,while the label dataset consists of corresponding temperature attributes.Back propagation neural network(BPNN)serves as a weak multi-classifier,and multiple BPNNs are aggregated into a robust multi-classifier by using the Adaboost.M2 ensemble algorithm.A new measured BGS can be categorized into temperature ranges based on the trained BP-Adaboost.M2 model.The influence of signal-to-noise ratio(SNR)and frequency scanning step on tem-perature extraction performance is explored through both simulation and experimentation,and the outcomes are compared with those of Lorentzian curve fitting(LCF),pseudo-Voigt curve fitting(pVCF),and BPNN.The results indicate that under SNR conditions below 11 dB and various frequency scanning steps,BP-Adaboost.M2 demonstrates the lowest root mean square error and uncertainty for tem-perature extraction.The processing time of BP-Adaboost.M2 for 1 000 data samples is merely 1.91 seconds,approximately 1.25 times faster than BPNN and more than 100 times faster than LCF and pVCF.关键词
布里渊光时域分析/布里渊增益谱/BP-Adaboost.M2/单标签多分类Key words
Brillouin optical time domain analysis/Brillouin gain spectrum/BP-Adaboost.M2/single-label multi-classification分类
信息技术与安全科学引用本文复制引用
郑欢,韩曼,舒涵,徐诺..基于BP-Adaboost.M2的布里渊光时域分析传感系统温度提取技术研究[J].传感技术学报,2025,38(5):809-816,8.基金项目
浙江省"尖兵领雁+X"科技计划项目(2025C02013) （2025C02013）
