Abstract The accuracy of classification models using spectral data decreases significantly with the increase numbers of categories. In order to overcome this problem, a middle layer is built based on the main factors of the sample. Herein, ten indicators, including chemical components, position and aroma styles, were selected to determine the identification of geographical origin and grade of flue-cured tobacco. Chemometrical algorithms were used to build quantitative prediction models based on labeled data. A voting algorithm was performed to determine the most likely geographical origin and grade of unknown samples. Experimental results show that the proposed method provides outstanding results for the independent test samples compared with traditional classify methods such as SIMCA and PLS-DA. The proposed method can be useful for origin traceability or adulteration detection of various agricultural products.

中文翻译：

---

使用近红外光谱同时识别烤烟的地理来源和等级

摘要 使用光谱数据的分类模型的准确性随着类别数量的增加而显着下降。为了克服这个问题，根据样本的主要因素建立了一个中间层。本文选取化学成分、位置和香气风格等10个指标来确定烤烟的地理原产地和等级。化学计量算法被用来建立基于标记数据的定量预测模型。执行投票算法以确定最可能的地理来源和未知样品的等级。实验结果表明，与SIMCA和PLS-DA等传统分类方法相比，所提出的方法在独立测试样本方面提供了出色的结果。