Abstract This paper is conducted to identify the authenticity, quality, and origin of saffron using hyperspectral imaging and multivariate spectral analysis. Reflectance spectra were extracted from hyperspectral images of saffron. Successive projections algorithm, genetic algorithm, uninformative variable elimination, and competitive adaptive reweighted sampling were used to select characteristic wavelengths. Back propagation neural network model was established based on the selected wavelengths. Results showed that the model combining competitive adaptive reweighted sampling with back propagation neural network achieved the best performance. Its prediction accuracy of the one-adulterated, three-domestic and two-imported saffron was 100, 95, 94, 100, 83, and 96%, respectively.

中文翻译：

---

使用高光谱成像和多变量光谱分析鉴定藏红花的真实性、质量和来源

摘要 本文旨在利用高光谱成像和多变量光谱分析鉴定藏红花的真伪、质量和来源。从藏红花的高光谱图像中提取反射光谱。使用连续投影算法、遗传算法、无信息变量消除和竞争性自适应重加权采样来选择特征波长。根据选定的波长建立反向传播神经网络模型。结果表明，将竞争性自适应重加权采样与反向传播神经网络相结合的模型取得了最佳性能。其对一掺、三国产、两进口藏红花的预测准确率分别为100%、95%、94%、100%、83%和96%。