应用高光谱成像技术对不同保藏温度的灵武长枣的可溶性固形物含量进行预测模型建立。提取图像中感兴趣区域的平均光谱数据, 经过不同光谱预处理后, 利用连续投影法（SPA）选择特征波长, 对4 ℃冷藏光谱提取13个特征波段（421, 426, 512, 598, 641, 670, 675, 723, 814, 906, 944, 978, 982 nm）, 对常温保藏光谱提取12个特征波段（425, 507, 555, 598, 673, 680, 685, 718, 809, 910, 954, 978 nm）。对于MSC处理、MSC+SPA处理、Savitzky-Golay平滑处理和SNV 4种预处理方法, 筛选出的最优预处理方法是冷藏采用MSC处理、常温采用MSC+SPA处理。对应这两种最优预处理方法, 分别建立偏最小二乘法（PLSR）、支持向量机（SVM）、主成分回归(PCR)3种预测模型。在以上获得的6个预测模型中, 得出冷藏、常温保藏的最优模型分别为MSC-PLSR模型（R2C: 0.852, RMSEC: 0.940; R2P: 0.857, RMSEP: 0.894）和MSC+SPA-PLSR模型（R2C: 0.872, RMSEC: 0.866; R2P: 0.787, RMSEP: 1.007）。结果表明: 利用高光谱成像技术, 结合多种预测模型建立, 能够测定不同保藏温度下的灵武长枣可溶性固形物含量, 实现对灵武长枣准确快速的无损检测。

The hyper-spectral imaging technology was applied to build a prediction model for soluble solid content of Lingwu jujube at different preservative temperature. The average spectra data were extracted from the area-of-interest of the image. After pre-treatment of different spectrum, the succession projection analysis (SPA) was used to select characteristic wavelength. 13 characteristic wavebands under 4 ℃ temperature condition (421, 426, 512, 598, 641, 670, 675, 723, 814, 906, 944, 978, 982 nm) and 12 characteristic wavebands under normal temperature condition (425, 507, 555, 598, 673, 680, 685, 718, 809, 910, 954, 978 nm) were extracted. By adoption of MSC treatment, MSC+SPA treatment, Savitzky-Golay smooth treatment and SNV treatment, both MSC treatment and MSC+SPA treatment out of 4 above were screened out as the optimum pre-treatment method afterwards. In corresponding to these 2 optimum pre-treatment methods, 3 prediction models like partial least squares regressions (PLSR), support vector machine (SVM) and principal component regression (PCR) model were built, respectively. Among the aforesaid 6 prediction models, 2 optimum modes such as PLSR model after treated by MSC (RC2: 0.852, RMSEC: 0.940; RP2: 0.857, RMSEP: 0.894) and PLSR model after treated by MSC+SPA (RC2: 0.872, RMSEC: 0.866; RP2: 0.787, RMSEP: 1.007) were acquired. The results show that the content of soluble solids of Lingwu jujube at different preservative temperature can be forecasted by utilization of hyper-spectral imaging technology in combination of building multiple prediction models, so that the nondestructive testing (NDT) can be achieved for Lingwu jujube in accurate and rapid manner.