近红外光谱定性与定量分析技术在莲子无损检测中的应用

通过提取采后不同时期的莲子、莲仁的近红外漫反射光谱特征，以莲子的可溶性固形物(SSC)和干物质含量(DM)为指标进行定量和定性分析。利用偏最小二乘回归(PLSR)分析和距离判别分析(DA)计算所得的结果表明：SSC和DM含量与莲子、莲仁的吸收光谱特征具有明显相关。莲子SSC、DM的PLSR模型在5 941～12 480 cm-1谱区综合性能较好，预测相关系数(r₁)分别为0.74和82，校正相关系数(r₂)分别为0.82和0.84，留一交互相关系数(r₃)分别为0.72和0.71。莲仁SSC的PLSR模型在7 891～9 310 cm-1谱区综合性能较好，r₁为0.79，r₂为0.84，r₃为0.77。DM的PLSR模型在全光谱的综合性能较好，r₁为0.92，r₂为0.89，r₃为0.82。莲子在5 400～7 885 cm-1谱区的判别性能较好，正确率达84.2%，而莲仁在9 226～12 480 cm-1谱区的判别性能较好，正确率达90.8%。对不同年份和有膜有芯的干莲仁进行DA判别的精度可达98.9%。研究表明近红外检测技术可用于莲子和莲仁的SSC和DM含量的定量分析及储存期的定性判别，还可对不同年份和有膜有芯的干莲仁进行判别。

Application of NIR Spectroscopy for Nondestructive Qualitative and Quantitative Analysis of Lotus Seeds

By extracting the Near Infrared (NIR) diffuse reflectance spectral characteristics from the post-harvest lotus seeds in different storage periods, the quantitative and qualitative analysis were applied to lotus seeds with the Soluble Solids Content (SSC) and dry matter content (DM) as criteria. The results of the Partial Least Squares Regression (PLSR) and distance discrimination (DA) models showed that the absorption spectra of lotus seeds and lotus kernels has clear relations to their SSC and DM. The PLSR models of SSC and DM of lotus seeds had the best performance in 5 941～12 480 cm-1 spectral region in this study. Their correlation coefficients of prediction were 0.74 and 0.82, and the correlation coefficients of calibration were 0.82 and 0.84, and the correlation coefficients of leave one out cross validation were 0.72 and 0.71. The PLSR model of SSC of lotus kernels was better in 7 891～9 310 cm-1 spectral region. Its correlation coefficient of prediction was 0.79, and the correlation coefficient of calibration was 0.84, and the correlation coefficient of leave one out cross validation was 0.77. The PLSR model of DM of lotus kernels is better in the full spectral region. Its correlation coefficient of prediction was 0.92, and the correlation coefficient of calibration was 0.89, and the correlation coefficient of leave one out cross validation was 0.82. For lotus seeds, the DA model in 5 400～7 885 cm-1 spectral region is the best with a correctness of 84.2%. And for lotus kernels, the DA model in 9 226～12 480 cm-1 spectral region is the best with a correctness of 90.8%. For dry lotus kernels, the discriminant accuracy of the DA model is 98.9% in the optimal spectral region. All kernels with membrane and plumule were correctly discriminated. This research shows that the NIR spectroscopy technique can be used to determine SSC and DM content of lotus seeds and lotus kernels, as well as to discriminate their freshness and also to discriminate dry lotus kernels of different age and the kernels with membrane and plumule.