摘要
针对目前莜麦产量不高,市面上存在掺假等问题,基于仅仅依靠人工检测,存在有效率低、误差大的问题,提出了一种基于近红外光谱技术和决策树的莜面面粉识别方法。采用近红外光谱仪采集180份莜面和莜面掺玉米淀粉样本的近红外光谱,使用三种方法对获得的近红外光谱进行去噪处理,使用主成分分析方法对光谱进行降维处理,最终使用决策树建立莜面识别模型,研究三种不同光谱去噪方法对决策树分类效果的影响,对比了最终分类效果的准确率。其中使用一阶导数去噪的准确率为96.67%,使用二阶导数去噪的准确率为95.00%,使用多元散射校正去噪准确率为96.67%,在使用决策树的前提下用一阶导数或者多元散射校正去噪并结合主成分分析方法对原始光谱进行数据降维是识别莜面面粉的最佳方法。
Aiming at the problems of low yield of naked oats and adulteration in the market at present,and based on the problems of low efficiency and large error only relying on manual detection,this paper proposes a method of naked oats flour recognition based on near-infrared spectroscopy and decision tree.The near-infrared spectroscopy was used to collect 180 samples of naked oat flour and naked oat flour mixed with corn starch.Three methods were used to denoise the obtained near-infrared spectra.The principal component analysis method was used to reduce the dimension of the spectra.Finally,the decision tree was used to establish the naked oat flour recognition model.The impact of three different spectral denoising methods on the classification effect of the decision tree was studied,and the accuracy of the final classification effect was compared.Among them,the accuracy of using the first derivative to denoise is 96.67%,the accuracy of using the second derivative to denoise is 95.00%,and the accuracy of using multiple scattering correction to denoise is 96.67%.Under the premise of using the decision tree,using the first derivative or multiple scattering correction to denoise and combining the principal component analysis method to reduce the dimension of the original spectrum data is the best method to identify naked oats flour.
作者
王翊同
邹其
李子熠
张天宇
李鸿强
WANG Yi-tong;ZOU Qi;LI Zi-yi;ZHANG Tian-yu;LI Hong-qiang(Hebei Institute of Architecture and Civil Engineering,Zhangjiakou,Hebei 075000)
出处
《河北建筑工程学院学报》
CAS
2022年第4期170-175,共6页
Journal of Hebei Institute of Architecture and Civil Engineering
关键词
莜面
近红外光谱
去噪处理
主成分分析
决策树
naked oatmeal flour
Near infrared spectroscopy
Denoising
principal component analysis
Decision tree
作者简介
王翊同(1996-),男,硕士,研究方向:模式识别;通讯作者:李鸿强(1979-),男,副教授,研究方向:模式识别。
