近红外光谱法预测大豆营养成分含量模型的建立和应用

Building and Application of Soybean Nutrients Content Model Predicted by Near Infrared Spectral Method

本试验旨在验证采用近红外光谱法(NIRS)预测大豆常规营养组分的可行性。180个样品来自黑龙江省,选用偏最小二乘法(PLS)为建模方法,采用二阶导数和Norris导数滤波法处理光谱数据后,建立了大豆中干物质(DM)、粗蛋白质(CP)、粗脂肪(CF)、粗灰分(Ash)等的近红外模型,为大豆提供科学的检测方法和高效的分析平台。结果表明:采用近红外光谱法得到的CP、CF、DM和Ash测定值与化学法测定值的决定系数(R2)分别为0.96、0.96、0.94和0.90,相对分析误差(SD/RMSEP)分别为4.52、3.75、3.47和2.47。本试验结果显示,用近红外光谱法可以准确快速测定DM、CP和CF含量,且预测的CP含量更为准确,而预测Ash含量的精度有待于进一步提高。

The objective of the experiment was to predict feasibility of soybean nutrients contents by near infrared spectral （NIRS） method. Samples were from HeiLongjiang province, and partial least squares method （PLS） was chose as modeling method, meanwhile second derivative and Norris derivative filter were made to treat spectral data, wave the first derivative and second derivative, therefor established predicted model of general nutrients by near infrared spectral （NIRS） method, and to provide detective methods scientifiely, and effective analysis platform for evaluating nutrients value quickly of soybean. Coefficient of determination of the determined value of DM, CP, CF and Ash using NIRS method and chemical method was 0.96, 0.96, 0.94 and 0.90, respectively and relative percent deviation （SD/ RMSEP） was 4.52, 3.75, 3.47 and 2.47 respectively. The conclusion obtained from this experiment was that DM, CP and CF contents can be obtained exactly and quickly by NIRS method, and predicted CP content was more accurate than other indexes, but accuracy of the predicted Ash content need to be improved.