分辨率对近红外光谱和定量分析的影响研究

近红外光谱分析技术是近年来发展最快的定性和定量分析技术之一，在较多领域得到了广泛应用。文章利用近红外光谱技术建立了杨梅汁酸度的定量分析模型，研究了不同分辨率对杨梅汁光谱和定量模型的影响。结果显示，分辨率对杨梅汁近红外光谱和其酸度的定量分析结果有较大的影响。高分辨率采集的光谱比低分辨率采集的光谱粗糙，扫描的速度较慢，所需的数据存储空间较大;不同分辨率下光谱的均方根噪声和平均吸光度在α＝0.05水平上差异显著;分辨率为4 cm-1时所建立的杨梅汁酸度模型的精度最高，模型的相关系数达到了0.994 04，校正均方根误差(RMSEC)和交互验证均方根误差(RMSECV)分别为0.023 3和0.153。但是考虑到扫描速度和数据存储空间，在不丢失样品信息的前提下采集光谱时建议将分辨率设为8 cm-1。

Study on the Influence of Resolution on Near Infrared Spectra and Quantitative Analysis

Near-infrared (NIR) spectroscopy technique is one of the qualitative and quantitative analysis techniques that were developed quickly in recent years.NIR spectroscopic analyses are non-destructive,simple and fast,and require no sample pretreatment,which makes this technology ideally suited for on line process monitoring and quality control. As a powerful analytical tool in product quality determination,this technology is based on the measurement of the frequencies of vibrations of chemical bonds in functional group such as C—C,C—H,O—H,CO and N—H upon absorption of radiation. A good and robust model is very important for NIR spectroscopy analysis. However,it was found that the status of NIR spectrometer and the set of parameters when scanning,such as accuracy of wavelength,resolution of apparatus,noise,scan time and uniformity of sample size,had effect on the spectrum quality. In the present paper,NIR transmittance spectra of bayberry juice were obtained and quantitative analysis was carried out. The influence of resolution on the NIR spectra and quantitative models was studied. Forty three bayberry samples with different varieties and ages were used in this experiment. Cross-validation was used to develop models and evaluate and compare these models. The results show that resolution does have effect on the NIR spectra and quantitative models. The spectrum with high resolution was rougher than that with low resolution. Moreover,the scan speed was lower and more data size was required when high resolution was used. The difference in root mean square noise and mean absorbance at different resolutions was distinct (α＝0.05). The best acid model was developed with 4 cm-1 spectral resolution,giving the relative high correlation coefficient of 0.994 04,and a root mean square error of calibration (RMSEC) and root mean square error of cross validation (RMSECV) of 0.023 3 and 0.153,respectively. However,if no information was lost,a resolution of 8 cm-1 was suggested considering the scanning time and data size.