基于高光谱成像和化学计量学对大米中的水分和脂肪酸含量进行无损检测

摘要：水分含量 (MC) 和脂肪酸含量 (FAC) 是衡量大米品质的关键指标。快速、无损地检测 MC 和 FAC 对大米储前 检查和储藏监控至关重要。本研究采用高光谱成像结合化学计量学对大米中的 MC 和 FAC 进行无损检测。采用竞争自 适应加权抽样 (CARS) 和顺序投影算法 (SPA) 方法选择重要变量，并利用偏最小二乘回归 (PLSR) 建立大米和精米中 MC 和 FAC 的预测模型。此外，还分析了稻壳对模型性能的影响。最后，实现了大米样品 MC 和 FAC 的可视化。结 果表明，SPA 方法在选择 MC 和 FAC 的重要波长方面优于 CARS 方法。 “大米-水分”、“精米-水分”、“大米-脂肪酸”、 “精米-脂肪酸”数据集的最佳模型中的RP 2和RMSEP分别为0.9650、0.9567、0.8573、0.8436和0.0031、0.0033、1.6956、 2.0270。稻壳对FAC模型性能的影响显著。本研究表明利用高光谱成像技术检测大米的MC和FAC是可行的。

关键词：高光谱成像；水分含量；脂肪酸含量；稻壳；精米；可视化；序贯投影算法；偏最小二乘回归

期刊：Journal Pre-proof

DOI：https://DOI.org/10.1016/j.jfca.2023.105397

设备：GaiaField

Non-Destructive Detection of Moisture and Fatty Acid Content in Rice using Hyperspectral Imaging and Chemometrics

Abstract：Moisture content (MC) and fatty acid content (FAC) are critical indicators in terms of rice quality. Rapid and non- destructive detection of MC and FAC is crucial for rice pre-storage inspection and storage monitoring. In this study, hyperspectral imaging combined with chemometrics was used to detect MC and FAC in rice non-destructively. Competitive adaptive reweighted sampling (CARS) and sequential projection algorithm (SPA) methods were employed to select significant variables, and partial least squares regression (PLSR) was utilized to establish prediction models of MC and FAC in rice and polished rice. Also, the effect of rice husk on the performance of models was analyzed. Finally, visualization of MC and FAC of rice samples was achieved. The results indicate that the SPA method is superior to the CARS method in selecting significant wavelengths of both MC and FAC. The RP 2 and RMSEPs in the best models for the "rice-moisture", "polished rice-moisture", "rice-fatty acid", and "polished rice- fatty acid" datasets are 0.9650, 0.9567, 0.8573, 0.8436, and 0.0031, 0.0033, 1.6956, 2.0270, respectively. The effect of rice husk on the performance of the FAC model is significant. This study demonstrates that it is feasible to use hyperspectral imaging technology to detect both MC and FAC of rice.

Keywords：Hyperspectral imaging; Moisture content; Fatty acid content; Rice husk; Polished rice; Visualization; Sequential projection algorithm; Partial least squares regression