近红外光谱法对不同地域糖蜜品质的快速鉴定

Rapid identification of the quality of molasses in different regions by near-infrared spectroscopy

研究旨在通过近红外光谱法替代传统的莱因·埃农滴定法,实现快速测定不同地域的糖蜜总糖含量。选取广西、广东、云南、海南的甘蔗糖蜜513批以及新疆、内蒙古的甜菜糖蜜359批进行建模以及预测,以糖蜜原液与一级水质量比1∶5做参比获取近红外光谱,通过K最邻近分类算法(KNN)联合偏最小二乘法(PLSR)建立总糖含量定量模型。结果显示:根据拟合优度(R^(2))、校正均方根误差(RMSEC)以及预测均方根误差(RMSEP),判断KNN-PLSR算法所建甘蔗糖蜜模型稳健性以及精确度良好,R^(2)为0.86,RMSEC为1.05%。PLSR算法所建甜菜糖蜜模型稳健性以及精确度良好,R^(2)为0.71,RMSEC为0.69%。研究表明,试验所用两种算法均可分别快速定量检测甘蔗糖蜜和甜菜糖蜜中总糖含量。

The study was to replace traditional Rein-Enon titration method by near-infrared spectroscopy for rapid determination of total sugars of molasses. 513 batches of cane molasses are from Guangxi, Guangdong, Yunnan and Hainan, and 359 batches of beet molasses are from Xinjiang and Inner Mongolia were selected for building models and treated as prediction sets. The near-infrared spectra were obtained by using the solution prepared by mass ratio of raw molasses and primary water 1∶5 as references. The quantitative model of total sugar content was developed by KNN classification algorithm combined with PLSR. The results showed that based on R^(2), RMSEC and RMSEP, the robustness as well as accuracy of sugarcane molasses model which was built. KNN-PLSR algorithm was good, its R^(2) value was 0.86,RMSEC value was 1.05%, and the robustness as well as accuracy of beet molasses model built by PLSR algorithm was good, its R^(2) value was 0.71, RMSEC value was 0.69%. The study indicates that the two algorithms can quickly and quantitatively detect the total sugar content in sugarcane molasses and beet molasses, respectively.