CAJ | 학술논문

以油茶籽中油酸、亚油酸、棕榈酸为研究对象，寻求一种油茶籽脂肪酸成分含量的最佳预测模型。首先，利用高光谱成像系统以线扫描方式获取其反射光谱图像，选择感兴趣区域（ ROI ）；然后，对原始光谱进行平滑与多元散射校正（ MSC ），通过相关性分析和逐步回归分析，得到能反映油酸、亚油酸、棕榈酸含量变化的最佳优化波段；进而对最优波段采用偏最小二乘回归（ PLS ）方法、主成分回归（ PCR ）方法及径向基神经网络（ RBF ）方法组建预测模型。比较这3种方法的建模效果，经外部验证表明：径向基神经网络建立的预测模型效果最好，其油酸、亚油酸、棕榈酸的交叉验证相关系数 R分别为0．9403、0．8935、0．9122；校正均方根误差和预测均方根误差分别为0．441、0．1749、0．0664和0．3518、0．184、0．162。
이유다자중유산、아유산、종려산위연구대상，심구일충유다자지방산성분함량적최가예측모형。수선，이용고광보성상계통이선소묘방식획취기반사광보도상，선택감흥취구역（ ROI ）；연후，대원시광보진행평활여다원산사교정（ MSC ），통과상관성분석화축보회귀분석，득도능반영유산、아유산、종려산함량변화적최가우화파단；진이대최우파단채용편최소이승회귀（ PLS ）방법、주성분회귀（ PCR ）방법급경향기신경망락（ RBF ）방법조건예측모형。비교저3충방법적건모효과，경외부험증표명：경향기신경망락건립적예측모형효과최호，기유산、아유산、종려산적교차험증상관계수 R분별위0．9403、0．8935、0．9122；교정균방근오차화예측균방근오차분별위0．441、0．1749、0．0664화0．3518、0．184、0．162。
To construct optimal predicting models based on the content of aliphatic acid of camellia seed, the research was concerned with oleic acid, linoleic acid, and palmitic acid.This approach was composed of four major procedures:line-by-line reflection spectrum scanning to select the region of interest ( ROI) by hyperspectral imaging system;smoot-hing the original spectrum and analyzing by multiplicative scatter correction ( MSC ); identifying the sensitive optimized waveband, which can reflect the variation of the content of oleic acid, linoleic acid, palmitic acid by correlation analysis and stepwise regression analysis;and then to build the optional waveband model by using partial least squares regression ( PLS) , principle component regression ( PCR) and radial basis function( RBF) neural network.Based on the external authentication, the RBF can achieve the best effects:the cross-validation correlation coefficient ( R) of oleic acid, lino-leic acid, and palmitic acid are 0.9403, 0.8935 and 0.9122;the correction error of root mean square are 0.441, 0. 1749 and 0.0664;the prediction error of root mean square are 0.3518, 0.184 and 0.162, respectively.