文献类型：期刊文献

中文题名：人体皮肤表面的近红外光谱检测及额温光谱分析模型的建立

英文题名：Determination of Near Infrared Spectra on Surface of Human Skin and Spectral Analysis Model Establishment of Forehead Temperature

作者：兰树明[1,2];张燕[1];王红鸿[1];吴婷[1];杜一平[1]

机构：[1]上海市功能性材料化学重点实验室,华东理工大学化学与分子工程学院,上海200237;[2]无锡迅杰光远科技有限公司,江苏无锡214000

年份：2025

卷号：45

期号：S1

起止页码：451

中文期刊名：光谱学与光谱分析

外文期刊名：Spectroscopy and Spectral Analysis

收录：;EI(收录号：20260620043259);WOS:【SCI-EXPANDED(收录号:WOS:001710525600008)】；北大核心:【北大核心2023】；

基金：国家自然科学基金面上项目(22576067)资助。

语种：中文

中文关键词：近红外光谱;体温;模型;偏最小二乘法

外文关键词：Near infrared spectroscopy;Human temperature;Model;Partial least squares

摘要：皮肤是人体重要器官,是防止人体水分流失的重要屏障。人体皮肤的检测项目很多,如含水量、水分散失、油脂、弹性、光泽度、摩擦力、酸碱度、温度等,这些指标通常采用物理方法进行检测。近红外光谱是近年来发展非常迅速的一项无损快检技术,在皮肤表面水分含量检测方面已经开展了较多的研究工作,展现了良好的发展前景。本文尝试采用近红外光谱技术,对人体体温进行无损、快速检测,扩大近红外光谱在皮肤检测中的应用范围。采用自行搭建的皮肤近红外光谱检测装置,在人体前额进行近红外光谱测量,同时用红外体温计检测前额的体温,采用多元校正方法建立额温的近红外光谱分析模型。分别采用平滑、求导、多元散射校正和标准正态变换等方法对光谱进行预处理,还对光谱进行波长选择,建立了相应的偏最小二乘模型,对模型进行了系统评价。研究发现,前额皮肤上采集的近红外光谱噪声比较大,经平滑处理获得较好的效果。与原始光谱及各种预处理方法处理后的光谱进行比较,平滑结合多元散射校正方法获得了最佳模型,在潜变量数为7时,交互检验均方根误差RMSECV为0.269,校正集和预测集均方根误差RMSEC和RMSEP分别为0.202和0.297,校正集和预测集相关系数R_(c)和R_(p)达到0.894和0.728。对比波长选择后所建立的模型,发现波长选择并不能改善模型的预测精度,可能是因为体温没有化学意义明显的特征谱带吸收,而是与全波长信号均具有关联。该研究揭示了用近红外光谱预测人体体温的可行性,为拓展近红外光谱技术在皮肤检测中的应用奠定了基础。
Skin is an essential human organ and serves as a critical barrier to prevent water loss.Numerous skin parameters—such as surface moisture content,transepidermal water loss,sebum level,elasticity,glossiness,friction,pH,and temperature—are typically assessed using physical methods.Near-infrared spectroscopy(NIR)has rapidly developed in recent years as a nondestructive and fast analytical technique.It has been widely applied in the determination of skin moisture and has demonstrated promising potential in this field.In this study,we explored the feasibility of applying NIR spectroscopy to the nondestructive and rapid measurement of human body temperature,thereby extending its application in skin assessment.A self-assembled NIR detection device was used to collect NIR spectra from the forehead,while forehead temperature was simultaneously measured using an infrared thermometer.Multivariate calibration models were then developed to relate NIR spectra to the temperature.Spectral pretreatments including smoothing,derivatives,multiple-scatter correction(MSC),and standard normal variate(SNV)were applied,and wavelength selection was also performed.Partial least squares regression(PLSR)models were established and systematically evaluated.The results showed that the raw spectra contained substantial noise,and smoothing significantly improved spectral quality.Among all pretreated spectra,the combination of smoothing and MSC yielded the best model performance.With seven latent variables,the model achieved an RMSECV of 0.269,with RMSEC and RMSEP of 0.202 and 0.297,and correlation coefficients(R_(c) and R_(p))of 0.894 and 0.728 for the calibration and prediction sets,respectively.Compared with full-spectrum models,models with wavelengths selected did not improve prediction accuracy,likely because body temperature lacks distinct chemically meaningful absorption features in the NIR region and is correlated with the global spectral response.Overall,this study demonstrates the feasibility of predicting human body temperature using NIR spectroscopy and provides a foundation for further expanding the application of NIR techniques in skin analysis.