基于最小角度回归模型的NIR光谱果品品质分析方法

Non-destructive measurement of fruit internal quality by NIR Spectroscopy based on the least angle regression model

[目的]本研究旨在提高果品内部品质检测的时效性和准确性。[方法]本文提出基于最小角度回归(Least Angle Regression,LAR)模型的果品品质分析方法。[结果]与现有的非线性LS-SVM(Least Squares Support Vector Machines)和线性的PLSR(Partial Least Squares Regression)模型进行对比表明,在预测准确度上,LS-SVM模型达到了最优的预测性能,而LAR模型明显优于常用的线性的PLSR模型;在计算复杂度上,LAR和PLSR模型明显优于LS-SVM模型;在模型的可解释方面,LAR模型要优于PLSR模型。[结论]LAR模型虽然在预测精度上稍逊于LS-SVM,但在模型的实现和计算复杂度以及可解释方面都具有明显的优势,因此提出的LAR模型更能有效地应用于基于NIR光谱的果品品质分析中。

[Objective]The objective of present study was to develop a new method to improve the effectiveness and accuracy of fruit internal quality measurement.[Methods]The model for non-destructive detection of internal quality of fruit based on the least angle regression(LAR)was proposed and the major parameters were compared with the existed models.[Results]Two existed non-liner and liner models,i.e.,LS-SVM(Least Squares Support Vector Machines)and PLSR(Partial Least Squares Regression)were selected to evaluate the effectiveness of new developed model.The comparison results indicated that the proposed LAR model generated the best prediction results and performed better than conventional PLSR,while LAR and PLSR were better than LS-SVM model in term of computational complexity,and the proposed LAR was superior to PLSR model considering the aspect of interpretability.Although the precision rate of LAR was lower than LS-SVM,it possessed advantages for model realization,computation complexity and interpretability over LS-SVM.[Conclusion]The proposed LAR model can be applied effectively in the determination of internal quality of fruit based on NIR(Near infrared)spectroscopy.