在近红外光谱分析应用中, 温度扰动导致的光谱变化对于定量分析的准确度影响较大。 针对温度扰动识别及定量分析的需要, 研究了基于双光谱二维相关谱(2T2D-COS)的光谱扰动分析方法。 选择葡萄糖水溶液为样本, 根据人体血糖浓度范围和在体组织温度范围设计实验, 测量样本在浓度扰动和温度扰动下的透射光谱。 对其进行基线校正和滤波预处理后, 通过2T2D-COS分析得到浓度扰动和温度扰动下的异步谱。 结果表明, 温度扰动引起的交叉峰出现在强氢键结合水对应的特征吸收波长(1 474 nm)和弱氢键结合水对应的特征吸收波长(1 410 nm)附近, 而葡萄糖浓度扰动引起的交叉峰出现在水分子对应的特征吸收波长(1 450 nm)和葡萄糖分子对应的特征吸收波长(1 595 nm)附近。 为了定量分析样品温度, 进一步提取了温度扰动异步谱交叉峰1 410 nm波长下的切片谱, 其在1 410～1 600 nm波段的相关峰强度随温度的升高而增大, 与样品温度之间具有较好的相关性； 选择波长(1 475±4) nm范围内的切片谱, 对谱峰强度积分后进行线性拟合, 建立样品温度的线性回归模型, 对温度的预测均方根误差可以达到0.125 9 ℃。 以上结果表明, 双光谱二维相关谱只需要对两条光谱进行运算, 就可以通过异步谱中交叉峰的位置判别扰动来源, 还可以根据交叉峰的强度对温度进行定量分析, 且只需要较窄特征波段的光谱就可以建立高精度的温度预测模型, 为简化在体光谱测量系统的设计, 减少在体温度变化的影响提供了重要参考。

The near-infrared absorption features are easily affected by temperature changes, which reduce the accuracy of quantitative analysis. This paper presents the methods for analysing the source and amplitude of temperature disturbance based on the theory of two-trace two-dimensional correlation spectroscopy (2T2D-COS). The absorption spectra of glucose solution, with a concentration of 0～200 mg·dL-1, are measured in the temperature range of 34~38.5 ℃ with a stepwise of 0.5 ℃. After baseline correction and noise filtering, the concentration-dependent and temperature-dependent spectra are analyzed with the 2T2D-COS algorithm. In the asynchronous spectrum of temperature-dependent spectra, the cross-peaks are observed at 1 474 and 1 410 nm, corresponding to strong and weak hydrogen-bonded water, respectively. While in the asynchronous spectrum of concentration-dependent spectra, the cross peaks at 1 450 and 1 595 nm are related to the characteristic absorption of water and glucose. From the slice spectrum at 1 410 nm, the intensity of the peaks in the range of 1 410n~1 600 nm is increased with temperature, demonstrating good correlation between the cross peaks and temperature variance. The calibration model is established for the quantitative analysis of sample temperature with the cross peak intensity range of (1 475±4) nm. The root mean square error achieved 0.125 9 ℃ for sample temperature prediction with linear regression. The experimental results demonstrate that the temperature disturbances could be identified and qualified with distinctly separate cross peaks in the asynchronous spectrum, obtained using only a pair of spectra. The proposed approach also provides a reasonable way for researchers to simplify the in vivo measurement system and establish a more stable calibration model.