基于高光谱和集成学习的人参果维生素C含量无损检测方法

打开文本图片集

中图分类号：TP181;TS255.7 文献标识码：A 文章编号： 1000-4440（2025）09-1771-10

Abstract：Vitamin Ccontentservesas a crucial index forevaluating thequalityof ginseng fruits.Inthis study，rapid andnon-estructivedetectionof vitamin Ccontentinginseng fruits was conductedbyacquiringtheirhyperspectraldata.To ffectively eliminatetheimpactofdatanoise，spectral preprocessing methodssuchasmovingaveragesmoothing，multivariatescattercorrection，first-orderderivative，andleastsquaressmothingfilterwereemployed.Theoptimal spectral preprocessing method was determined bycomparing thepredictive performance of models built usingsupport vectorregresion.Addresing the issue of dimension reduction in hyperspectral data，competitiveadaptive reweighted sampling，sucesive pro

jections algorithm，and light gradient boostingmachine algorithm were utilized to extract feature wavelengths highlycorrelated with the vitamin C content in ginseng fruits.The selected feature wavelengths were then combined with support vector regression，random forest regression，multi-layer perceptron，and Stacking methods for

modelingandcomparisontoidentifythebestpredictivemodel.Theresults indicatedthattheStacking methodexhibitedthe best prediction performance，with coefficient of determination（ R2 ）ofO.9172，root mean square error（ RMSE ）of 15.053，and relative percent deviation（ RPD ）of 3.595 3 for the validation set. Thismethod enablesrapid and accurate prediction of vitaminCcontentin ginseng fruits，providing technical supportfortheevaluation，grading，andsorting of ginseng fruit quality.

Key words： ginseng fruit；vitamin C；hyperspectral；machine learning；ensemble learning

人参果是原产于南美洲的茄科类多年生水果和蔬菜两用型作物，在中国南方地区分布广泛[2]。（剩余13391字）