对原始高光谱数据进行光谱特征提取可以减少数据的维度,提高分析效率,同时保留关键信息,从而有利于提高模型的准确性和泛化能力。本研究采用3种典型的数据降维算法确定有效波长,分别是蒙特卡罗无信息变量消除(Monte Carlo uninformative variable elimination,MC-UVE)、CARS和SPA。
预测模型建立后,通过4个指标对模型的预测性能进行评价。具体的评价指标分别为训练集的决定系数R(decision coefficient of calibration set)、训练集均方根误差RMSEC(root mean square error of calibration,式中记为RMSEC)、R和RMSEP(式中记为RMSEP)。计算公式为
利用CARS算法的样本SSC特征波段选择结果如图6所示。在该过程中,使用50次蒙特卡罗采样迭代,并采用5倍交叉验证选择最终变量。交叉验证均方根误差 (root mean squared error of cross-validation,RMSECV)值越小,与样本相关的特征波段子集越好。由图6可知,随时间采样的变量数量,5倍RMSECV值以及每个变量的回归系数路径与运行一次CARS时进行的蒙特卡罗采样运行次数有关。图6(a)显示了CARS算法是如何结合快速选择和精细选择的,表明了选择的变量数在变得相对稳定时如何急剧下降。由于在图6(b)中,随着采样运行次数的增加,无信息变量逐渐被消除,因此模型的RMSECV值从采样运行1~34次开始缓慢下降,然后,由于一些重要变量被删除,RMSECV值开始上升。最小5倍RMSECV值用于建立样本SSC预测的最佳变量子集。图6(c)中的每一行显示每个变量在不同采样运行时的回归系数。回归系数较高的变量更有可能被选择。因此,回归系数的分析可用于确定每次样本运行的一组变量。最小5倍RMSECV值用带有星号的垂直线表示。通过CARS算法,提取了12个特征波段进行样本SSC的分析。
CHENY H, HUNGY C, CHENM Y,et al. Enhanced storability of blueberries by acidic electrolyzed oxidizing water application may be mediated by regulating ROS metabolism[J].Food Chemistry,2019,270:229-235.
[2]
QIAOS C, TIANY W, WANGQ H,et al.Nondestructive detection of decayed blueberry based on information fusion of hyperspectral imaging (HSI) and low-field nuclear magnetic resonance(LF-NMR)[J].Computers and Electronics in Agriculture,2021,184:106100.
[3]
LIJ L, SUND W, CHENGJ H.Recent advances in nondestructive analytical techniques for determining the total soluble solids in fruits:A review[J].Comprehensive Reviews in Food Science and Food Safety,2016,15(5):897-911.
ZHANGL X, ZHANGS J, SUNH X,et al.Determination of soluble solid content in peach based on hyperspectral combination with BPSO[J].Spectroscopy and Spectral Analysis,2024,44(3):656-662.
CHANGH J, MENGQ H, WUZ F,et al.Detection of soluble solids content in mango based on backpropagation algorithm neural network and hyperspectral imaging[J].Journal of Food Safety and Quality,2024,15(2):141-148.
[8]
QIAOS C, TIANY W, GUW J,et al.Research on simultaneous detection of SSC and FI of blueberry based on hyperspectral imaging combined MS-SPA[J].Engineering in Agriculture,Environment and Food,2019,12(4):540-547.
[9]
GAOS, XUJ H.Hyperspectral image information fusion-based detection of soluble solids content in red globe grapes[J].Computers and Electronics in Agriculture,2022,196:106822.
LIB, HANZ Y, WANGQ,et al.Research on bruise level detection of loquat based on hyperspectral imaging technology[J].Spectroscopy and Spectral Analysis,2023,43(6):1792-1799.
SONGZ Y, CHANGQ R, ZHENGZ K,et al.Estimation of kiwifruit leaf nitrogen balance index based on hyperspectral and successive projections algorithm[J].Jiangsu Journal of Agricultural Sciences,2024,40(7):1260-1267.
ZHANGY, JIAY, DUJ,et al.Rapid determination of flavonoids and antioxidant activity of fermented Astragalus membranaceus stems and leaves by partial least square method and near infrared spectroscopy technique[J].Feed Research,2024,47(18):70-75.
FANGX M, WANGH L, XUH,et al.Identification and detection of multi-component trace gases based on near-infrared TDLAS technology based on SVM[J].Spectroscopy and Spectral Analysis,2024,44(10):2909-2915.
[28]
YUK Q, ZHAOY R, LIUZ Y,et al.Application of visible and near-infrared hyperspectral imaging for detection of defective features in loquat[J].Food and Bioprocess Technology,2014,7(11):3077-3087.
[29]
HEX M, FUX P, RAOX Q,et al.Assessing firmness and SSC of pears based on absorption and scattering properties using an automatic integrating sphere system from 400 to 1150 nm[J].Postharvest Biology and Technology,2016,121:62-70.
[30]
XUEJ X, ZHANGS J, SUNH X,et al.Detection of shelf life of Malus asiatica using near-infrared spectroscopy and softening index[J].Transactions of the Chinese Society for Agricultural Machinery,2013,44(8):169-173.
[31]
LIUF, HEY, WANGL.Comparison of calibrations for the determination of soluble solids content and pH of rice vinegars using visible and short-wave near infrared spectroscopy[J].Analytica Chimica Acta,2008,610(2):196-204.
[32]
BALABINR M, SMIRNOVS V.Variable selection in near-infrared spectroscopy:Benchmarking of feature selection methods on biodiesel data[J].Analytica Chimica Acta,2011,692(1/2):63-72.
[33]
COSTAR C, DE LIMAK M G.Prediction of parameters (soluble solid and pH) in intact plum using NIR spectroscopy and wavelength selection[J].Journal of the Brazilian Chemical Society,2013,24(8):1351-1356.
[34]
ZHANGD Y, XUY F, HUANGW Q,et al.Nondestructive measurement of soluble solids content in apple using near infrared hyperspectral imaging coupled with wavelength selection algorithm[J].Infrared Physics & Technology,2019,98:297-304.
[35]
ZHANGC, GUOC T, LIUF,et al.Hyperspectral imaging analysis for ripeness evaluation of strawberry with support vector machine[J].Journal of Food Engineering,2016,179:11-18.
[36]
PANT T, CHYNGYZE, SUND W,et al.Pathogenetic process monitoring and early detection of pear black spot disease caused by Alternaria alternata using hyperspectral imaging[J].Postharvest Biology and Technology,2019,154:96-104.
[37]
RICCIOLIC, PÉREZ-MARÍND, GARRIDO-VAROA.Optimizing spatial data reduction in hyperspectral imaging for the prediction of quality parameters in intact oranges[J].Postharvest Biology and Technology,2021,176:111504.