The commonly used spectral variable selection methods in near-infrared (NIR) spectroscopy were more theoretical and difficult to put into practice, due to a large number of optical filters with extremely narrow bandwidth at the desired wavelength was required for the spectral acquisition. In this study, a method of optimally selecting a set of the band-pass filter (BPF) to reduce the dimensionality of the spectral data was proposed and subsequently applied to the determination of theanine content in oolong tea. By utilizing 4 BPFs, the developed multiple linear regression, support vector regression and Gaussian process regression models produced R-squared values of 0.7971, 0.9036 and 0.9080, respectively, for prediction, indicating the beneficial potential of the proposed method for accurate prediction of the analytes with the lower cost of spectral acquisition in real practice.

近红外（NIR）光谱中常用的光谱变量选择方法更具理论性，难以实践，因为光谱采集需要大量在所需波长处具有非常窄带宽的滤光器。在这项研究中，提出了一种最佳选择一组带通滤波器（BPF）来降低光谱数据维数的方法，该方法随后被用于乌龙茶中茶氨酸含量的测定。通过使用4个BPF，所开发的多元线性回归，支持向量回归和高斯过程回归模型分别得出R平方值0.7971、0.9036和0.9080，以进行预测，