间隔连续投影算法应用于近红外光谱苹果糖度模型的优化

Optimization of near infrared spectroscopy model for sugar content in apple by intervals successive projection algorithm

中文关键词:  近红外光谱法  间隔连续投影算法  偏最小二乘模型  变量筛选  快速检测

英文关键词:near infrared spectroscopy  interval successive projection algorithm  partial least squares  variable screen  rapid detection

中文摘要:

      目的 采用一种改进的连续投影算法(successive projection algorithm, SPA)筛选光谱区间变量, 优化苹果近红外光谱模型。方法 试验以半透射方式无损地获取134个苹果的光谱信息, 再以标准方法破坏性检测其内部糖度指标, 在光谱信息与糖度指标之间构建定量模型。区间连续投影算法(intervals SPA, iSPA)是根据各光谱区间之间的投影关系, 选择那些具有共线性小的区间变量来构建偏最小二乘模型(partial least square, PLS)。尝试以全区间光谱划分的间隔数量从5到60, 步长为5, 以优化共线性小的间隔组合。结果 当划分为20个间隔时, 构建的PLS模型相比于其他划分间隔时的模型, 具有较小的交互验证均方根误差和较少的入选变量, 此时对预测集的预测均方根误差为0.521, 优于常规连续投影算法线性回归和全区间PLS模型的预测性能。结论 区间连续投影算法可用于光谱区间变量的筛选, 结合偏最小二乘法可提高模型的预测性能。

英文摘要:

      Objective To optimize the apple model on the basis of near infrared spectroscopy technology with the spectral intervals by an improved successive projection algorithm (SPA). Methods The spectral information of 134 apples was obtained nondestructively by semi-transmission scanning mode, and the internal sugar content (SC) was measured by standard destructive method, and then a quantitative model was developed between the spectral information and SC. The interval successive projection algorithm coupled with partial least squares (iSPA-PLS) model was built by screening these intervals with small co-linearity based on the projection relationships between each spectral interval. In order to optimize the combination with small co-linearity attribute between spectral intervals, the whole spectra were divided into from 5 to 60 segments with 5 steps. Results When dividing into 20 intervals, the constructed PLS model had the relative lower root mean square error of cross validation (RMSECV), as well as the relative lower number of the screened variables, and samples in the prediction set were tested with root mean square error of prediction (RMSEP) of 0.521, which was better than the predictive performance of the conventional successive projection algorithm-linear regression (SPA-MLR), interval-based PLS and the full-based PLS model. Conclusion The iSPA method can be used to screen the spectral intervals, which can enhance the prediction performance of regression model combined PLS.