小米米粉碱消值的高光谱快速预测

小米米粉的主要成分是淀粉，其食味品质决定小米米粉的市场价值。糊化特性是小米米粉的重要物理特性，而碱消值是能够直接反应其糊化特性的主要特征指标。通过小米米粉碱消值的差异，可以间接反映直链淀粉含量，当碱消值降低时，相反，糊化温度和直链淀粉含量却很高，而小米米粉口感粘糯性越差。采用高光谱技术结合化学计量学方法，建立快速检测小米米粉碱消值预测模型，旨在探索一种快速、无损、低成本预测小米米粉碱消值的方法。实验采集小米米粉高光谱数据，在被测样品感兴趣区域(ROI)按像素点逐一选择，提取高光谱数据矩阵，并进行均值运算，得到每个样品在各个光谱波段的平均光谱值。利用粘度测定仪(RVA)测定小米米粉碱消值指标。光谱数据采用全波段、竞争性自适应重加权采样法(CARS)及随机蛙跳(RF)法选择特征波段处理，建立偏最小二乘回归(PLSR)模型；全波段建立预测模型R_p值最高为0.77，说明能够利用小米米粉高光谱反射率反演小米米粉的碱消值，而采用其他两种计算方法所得R_p值分别为0.72和0.7，与前者较为接近，也反映了采用CARS和RF建立的回归模型具有可行性。为提高预测精度，采用Savitzky-Golay(S-G)法、多元散射校正(MSC)和S-G+MSC对数据预处理。可以看出采用MSC预处理光谱数据建立PLSR模型性能较好(R_p=0.83)。对MSC预处理后的数据再次CARS和RF法选择特征波段，建立PLSR模型，与未进行预处理的回归模型相比，R_p值变化不大，这也说明CARS和RF具有一定的稳定性，可以作为小米米粉高光谱反射率预测碱消值的参考方法。结果表明：为实现对小米米粉碱消值的快速、无损检测，通过运用高光谱技术能够利用小米米粉高光谱反射率预测碱消值，进而为小米米粉品质评级、加工及碱消值传感器的开发提供参考依据和数据支撑。

Hyperspectral Technique Coupled With Chemometrics Methods for Predicting Alkali Spreading Value of Millet Flour

As the main ingredient of millet flour, the quality of starch determined the market price of millet flour. Gelatinization characteristic is one of the most important physical characteristics of millet flour, and the alkali spreading value is the main index that reflects the gelatinization characteristic directly. The differences in the alkali spreading the value of millet flour show the quality of amylose content. When the alkali spreading value becomes lower, on the contrary, the gelatinization temperature and amylose content become higher, eventually the lower the waxy of millet flour. This study employed the hyperspectral technique could with chemometrics methods to develop an approach for detecting the alkali spreading the value of millet flour, whose aim is to explore a rapid, nondestructive and low-cost method for predicting the alkali spreading the value of millet flour. First, the hyperspectral data of millet flour were collected, then the hyperspectral data matrix in the region of interest (ROI) in each pixel was computed. The results were meant in each wavelength of every sample. Then we used the rapid visco analyser (RVA) to measured the alkali spreading the value of millet flour. In the data processing, partial least square regression (PLSR) models were made after using competitive adaptive reweighted sampling(CARS) and random frog (RF) to extracted key wavelengths. The results showed that the highest predicted R_p was 0.77 in the PLSR of the full wavelengths, and that explained that the reflectance of millet flour could invert the alkali spreading the value of millet flour. The Rp in the other two methods were 0.72 and 0.7, and both were close to the previous result, these illustrated it was feasible to build the PLSR using CARS and RF. In order to improve the predicting accuracy, the full wavelengths were preprocessed by Savitzky-Golay (S-G), multiplicative scatter correction(MSC) and S-G+MSC. The performance of the PLSR model was better by using MSC predicted the full wavelengths (R_p=0.83). Then built the PLSR model again after extracting key wavelengths using CARS and RF, compared with the models without pretreatment, the R_p does not change much, which also shows that CARS and RF have a certain stability and can be used as reference methods for predicting the alkali spreading the value of the hyperspectral reflectance of millet flour. The results showed that the reflectance of millet flour could predict its alkali spreading value by using hyperspectral. This could supply a rapid, nondestructive and low-cost method of the alkali spreading value of millet flour, then provided the theoretical foundation for the rating, processing and alkali spreading value sensor of millet flour.