扫描次数对番茄叶漫反射光谱和模型精度的影响研究

Study on the Influence of Scan Number on Near-Infrared Diffuse Spectra of Tomato Leaf and Model Precision

近红外光谱分析技术因其诸多优点成为近年来发展最快的定量和定性分析技术之一,它是依据有机分子中的含氢基团(C—H,O—H,N—H)振动的合频与各级倍频的吸收获得信息。但是,近红外光谱易受到光谱仪的工作状态和扫描时各种参数的设置,如波长的准确性、仪器的分辨率、噪声、扫描次数、样品粒度的均匀性等的影响。为了给建立实际模型时选择最佳的测量条件提供依据,文章以番茄叶为研究对象,分析了扫描次数对番茄叶漫反射光谱和叶绿素定量模型精度的影响。结果发现扫描次数对番茄叶漫反射光谱和模型精度有较大的影响。随着扫描次数的增加,近红外光谱的均方根噪声的方差值逐渐变小。扫描次数越多,光谱的质量较高,但是仪器产生的系统误差也会增大;扫描次数越少,光谱的质量较差,但是光谱较为光滑,系统的误差较小。在128次扫描次数下,叶绿素定量模型的决定系数较高,但是模型不稳定。当扫描次数为32时,模型的决定系数较低,但该模型比较稳定,检测时间也较短。同时,不同扫描次数下所建模型都用于预测叶绿素含量,发现没有显著差异(α=0.05)。在采集光谱建立模型时,应考虑综合因素选择合适的扫描次数。

Near-infrared spectroscopy technique is non-destructive, simple, fast, highly efficient, cheap to implement, and very recurrent with no sample preparation, and has been a rapid and non-destructive modern qualitative and quantitative technique that has been widely used in many fields. As a powerful analytical tool in product quality determination, this technology is based on the measurement of vibration frequencies of chemical bonds in functional group such as C--C, C--H, O--H, C=O and N--H upon absorption of radiation. However, NIR spectra are affected by the status of spectrometer and the set of parameters when scanning, such as accuracy of wavelength, resolution of apparatus, noise, scan time and uniformity of sample size. To provide foundation with optimum test condition when modeling, the influence of scan number on NIR diffuse spectra of tomato leaf and chlorophyll prediction model precision was studied. 102 tomato leaf samples were used in this experiment. Partial least-squares （PLS） was used to develop models and evaluate and compare these models. The results show that scan number does have effect on NIR spectra and prediction models. Variance value of root mean square （RMS） noise of NIR spectra diminished gradually with the increment of scan number. The spectral quality with high scan number was high, however, the system error of instrument increased too. The spectral quality with low scan number was low, while the spectra were smooth and system error of instrument decreased too. The determination coefficient of chlorophyll calibration and prediction model was highest with 128 scan number, however, the model was not robust. But with 32 scan number, although the coefficient was low, the calibration and prediction model was robust and only a short test time was needed. At the same time, the difference of models to predict chlorophyll contents with different scan numbers was not distinct （a=0. 05）. Different influence factors should be considered when modeling.