近红外光谱结合随机森林的稻米粉种类快速鉴别

打开文本图片集

中图分类号：S511；TS213.3 文献标识码：A 文章编号：2095-5553（2025）10-0161-07

Abstract：Arapidand reliable methodforidentifying diferenttypesofriceflour species was established byutlizig near infrared spectroscopy（NIRS） technology.NIRS data（ 4000cm-1 to 12 000cm-1 ）from149 samples of rice flour were collected，andthreepredictivemodels，namelyPartialLeastSquares-DiscriminantAnalysis（PLS—DA），SupportVector Machine（SVM），and Random Forest（RF），were constructed.Theresults showedthat after first derivative preprocessing，the RF model achieved an accuracy of 98.06% on the full-band test set. The PLS—DA，SVM，and RF models wereoptimized by usingfour diferent feature wavelength selection algorithms.Theresultsof the test setrevealed thattheSVMmodel，inconjunctionwith standardnormal variabletransformationand non-informativevariableelimination algorithm，as wellas theRF model，combined withistandardnormal variable transformationandangularregression algorithm，achieved an accuracy，recall rate，and precision of 100% ，demonstrating the efficiency and accuracy of near infraredspectraltechnologyintherapid identificationof riceflour types.Thisresearchconfirmed thatusing near ifrared spectral technology combined with machine learning models was an effctive method for identifying rice flour types.

Keywords： rice flour; near infrared spectrum；machine learning;random forest；wavelength screening

0 引言

近红外光谱技术在食品分类鉴别等领域具有广泛应用[1]，其高效、迅速以及无损等特性2为品质分析提供了强有力支撑[3]。（剩余12127字）