基于连续投影算法的小麦蛋白质含量近红外光谱预测分析

Prediction and analysis of wheat protein content by near-infrared spectroscopy based on successive projections algorithm

为了快速无损检测分析小麦蛋白质含量,构建近红外光谱最优小麦蛋白质定量检测分析模型。利用一阶S-G平滑算法+SNV算法对光谱进行预处理。使用连续投影算法(Successive projections algorithm, SPA)提取光谱中的特征波段点,使全谱图的141个波段点降低到17个特征波段点。在选择的17个特征波段点基础上分别建立偏最小二乘回归(Partial least regression, PLSR)模型、支持向量机(Support vector machine, SVM)模型、多元线性回归(Multiple linear squares regression, MLR)模型和主成分回归(Principal component regression, PCR)模型。在构建的4种小麦蛋白质含量预测模型中,MLR预测分析模型的验证集均方根误差(RMSEV)和校正集均方根误差(RMSEC)最小,验证集相关系数(r_v)和校正集相关系数(r_c)最大,其r_v=0.968,r_c=0.976,RMSEV=0.300,RMSEC=0.275。因此,相比于其他3种检测模型,建立的MLR小麦蛋白质含量检测模型最优,稳定性和精确性最高。

In order to detect the wheat protein content quickly and non-destructively, an optimal quantitative analysis model of wheat protein content was constructed. The first derivative S-G smoothing algorithm and the standard normal variable (SNV) were used to preprocess the spectrum. The successive projections algorithm (SPA) was used to extract the characteristic band points in the spectrum, so that 141 band points of the full spectrum were reduced to 17 characteristic band points. Partial least square regression(PLSR) model, support vector machine(SVM) model, multiple linear regression(MLR) model and principal component regression(PCR) model were established on the basis of 17 selected characteristic band points. In the four wheat protein content prediction models, the MLR model had the smallest root-mean-square error of the validation set ( RMSEV ), the smallest root-mean-square error of the calibration set ( RMSEC ), the largest correlation coefficient of the validation set ( r v) and the largest correlation coefficient of calibration set( r c). The correlation coefficients of validation set and calibration set were 0.968 and 0.976. The root mean square error of validation set and calibration set were 0.300 and 0.275. Compared with the other three detection models, the MLR detection model is the best, and the stability and accuracy are the highest.