NIR光谱法快速预测小麦籽粒干物质含量

NIR Spectra Method for Rapid Prediction of Dry Matter Content in Wheat Grain

通过采集百农201、百农207、百农307、百旱207、AK-58、冠麦1号、周麦18等7个不同品种完整小麦籽粒的近红外光谱(900～1700 nm)信息,经高斯滤波平滑(Gaussian Filtering Smoothing,GFS)、标准化校正(Normalization Correction)和卷积平滑(Savitzky-Golay Convolution Smoothing,SGCS)三种预处理后,利用偏最小二乘回归(Partial Least Squares Regression,PLSR)算法寻找光谱信息与小麦籽粒干物质含量之间的定量关系。结果显示,经GFS预处理的近红外光谱(100个波长)构建的全波段PLSR模型(PLSR)预测相关系数(RP)为0.952,预测误差(RMSEP)为0.158%,RMSEC与RMSEP绝对值差(ΔE)为0.082,预测效果优于其他两种预处理光谱。从GFS光谱中经PLSR-β法筛选获得17个最优波长,构建的优化模型(O-PLSR)RP为0.928,RMSEP为0.191%,ΔE为0.049,其预测效果接近于PLSR模型。试验表明,利用900～1700 nm光谱可被潜在用于快速无损预测小麦籽粒干物质含量。

Bainong 201, Bainong 207, Bainong 307, Baihan 207, AK-58, Guanmai 1 and Zhoumai 18 were used as materials. Gaussian Filtering Smoothing(GFS), Normalization Correction(NC) and Savitzky-Golay convolution smoothing(SGCS)were used to pretreat near-infrared spectral information(900~1700 nm) of whole wheat grain. Partial least squares regression was used to relate the spectral data to dry matter content of wheat grain. The results showed that the full-band PLSR regression model(PLSR) constructed by GFS pretreatment(100 wavelengths) had better performance(R_P= 0.952, RMSEP=0.158%,ΔE=0.082), than other two pretreated spectra.PLSR-β method was used to select 17 optimal wavelengths to optimize the PLSR model(building O-PLSR model), resulting in the RPof 0.928, RMSEP of 0.191% and ΔE of 0.082, similar to PLSR model. The whole experiment showed that ear-infrared spectra in the range of 900~1700 nm could be used to rapid and nondestructive prediction of dry matter content in wheat grains.