为检测面粉中非法过量添加的超量滑石粉，提出一种近红外多光谱定量检测方法。首先，选取三种品牌面粉，分别制备滑石粉质量分数为0%、0.5%、1%、3%、5%、10%、15%和20%的样本，采集样本在900~1700 nm的原始光谱，随滑石粉质量分数增大，样本光谱曲线在1160~1700 nm范围内幅值逐渐下降，足量掺杂滑石粉的样本光谱曲线在1393 nm出现与纯滑石粉光谱曲线一致的微小吸收峰。对比7种不同预处理方法所对应全波长模型的预测效果，选取标准正态变量为建模最优预处理方法。采用三种变量优选方法提取最优波长并分别建立面粉中滑石粉含量的多光谱定量检测模型。研究结果表明：竞争自适应重加权采样（CARS）方法对应多光谱模型的检测效果最好，验证集R2p为0.998，RMSEP为0.282%；而连续投影算法（SPA）方法选择的波长数量最少且彼此间共线性最小；三种波长优选方法所构建多光谱模型的检测限均可达0.5%，可为便携式或在线式检测仪器的开发提供理论基础。

In order to detect excessive talc powder illegally added in flour, a near infrared multi-spectral quantitative detection method was proposed. Firstly, the samples with 0%, 0.5%, 1%, 3%, 5%, 10%, 15% and 20% talc powder mass fraction were prepared. The original spectra of the samples within the wavelength range of 900~1700 nm was acquired. With the increase of talc powder mass fraction, the amplitude of the sample spectrum curve decreased gradually in the range of 1160~1700 nm, and small absorption peaks appeared at 1393 nm in the spectrum curve of the samples with sufficient doping talc powder as compared with that of pure talc powder. By comparing the prediction results of the full-wavelength model corresponding to seven different pretreatment methods, the standard normal variables were selected as the optimal pretreatment method. Furthermore, three variable optimization methods were used to extract the optimal wavelengths and then the corresponding multispectral quantitative detection models for talc content in flour were established. The results showed that the competitive adaptive re-weighted sampling (CARS) method had the best detection effect among the three multispectral models. The verification set R2p was 0.998 and RMSEP was 0.282%. The continuous projection algorithm (SPA) method had the least number of wavelengths and the smallest collinearity among them. The detection limits of the three multi-spectral models corresponding to the three wavelength selection methods can all reach 0.5%, which can provide a theoretical basis for the development of portable or on-line testing instruments.

国家自然科学基金面上项目(31772062)；“十三五”国家重点研发计划项目(2018YFC1603506)；中国农业大学2019“URP”科研训练计划项目