Abstract Three experimental setups, based on near infrared technology (NIR), were tested for rapid “at-line” assessment of sunflower oil adulteration by mineral oil. Experimental setups included a commercial portable NIR, coupled to both reflexion (S1) and immersion probes (S2), and a prototype of a multichannel Quasi Imaging Visible NIR spectrometer (QIVN) coupled to an immersion probe (S3). Independent calibration and validation samples sets were prepared with mineral oils (MOs) content up to 10% (w/w), and calibrations were developed using partial least square (PLS) regressions. Root mean square error of prediction (RMSEP) ranges from 0.23 to 1.26% (w/w) MOs, depending on the NIR setup. The best performances were obtained with S1, which provides satisfactory calibrations, and low number of false positives starting from levels of mineral oil around 1%. S3 still provides acceptable calibrations, and could be practically used to detect mineral oil at concentrations higher than 2.5% (w/w) MOs.

中文翻译：

---

应用 NIR 技术在线快速检测掺有矿物油的葵花籽油

摘要 测试了三种基于近红外技术 (NIR) 的实验装置，以快速“在线”评估矿物油掺假的葵花籽油。实验装置包括与反射 (S1) 和浸入式探头 (S2) 耦合的商用便携式 NIR，以及与浸入式探头 (S3) 耦合的多通道准成像可见近红外光谱仪 (QIVN) 的原型。使用最高 10% (w/w) 的矿物油 (MO) 含量制备独立的校准和验证样品组，并使用偏最小二乘 (PLS) 回归开发校准。预测均方根误差 (RMSEP) 的范围从 0.23 到 1.26% (w/w) MO，具体取决于 NIR 设置。使用 S1 获得了最佳性能，它提供了令人满意的校准，从大约 1% 的矿物油水平开始，误报率很低。S3 仍然提供可接受的校准，并且可以实际用于检测浓度高于 2.5% (w/w) MOs 的矿物油。