高光谱技术结合特征波长/光谱指数对冬枣成熟度可视化判别

Discrimination of Winter Jujube's Maturity Using Hyperspectral Technique Combined with Characteristic Wavelength and Spectral Indices

为了给冬枣采收后成熟度分级提供理论指导,运用高光谱技术获取特征波长和计算光谱指数对其成熟度可视化分级。采集三类成熟度冬枣(未成熟果、白熟-初红果、半红-全红果)样本共336个并获取其高光谱信息,通过Savitzky-Golay(S-G)平滑对原始光谱降噪后再用Kennard-Stone(K-S)方法将样本分为训练集(226个)和测试集(110个)。选用连续投影法(SPA)和竞争性自适应重加权采样法(CARS)选择特征波长(CWs);同时从水果生理成分变化角度引入7个光谱指数(SIs)。基于SPA和CARS选取的CWs和引入的SIs分别建立偏最小二乘判别分析(PLS-DA)模型,并比较了3个模型的分级效果。结果表明:基于SPA和CARS选择的特征波长和引入的SIs建立的PLS-DA模型判别精度分别为:97.27%,95.45%和98.18%。为了直观展现判别结果,选用SIs建立的PLS-DA回归系数拟合判别向量Y的回归方程,依据Y中最大值元素所在类别为该样本预测类别的规则,将结果用不同颜色直观显示。该研究为冬枣成熟度可视化分级提供了思路,引入的SIs参数为开发适于多种水果成熟度分级的设备提供了技术支撑。

In order to provide theoretical guidance for the grading of winter jujube maturity after harvesting,this study applied hyperspectral technique to obtain the characteristic wavelengths and calculate the spectral indices to achieve its maturity visualsort.A total of 336 samples of jujube with three types of maturity（immature fruit,white ripeness and primary red fruit,half red and red fruit）were collected and their hyperspectral information wereacquired.The samples were divided into training set（226）and testing set（110）using Kennard-Stone（K-S）method after the original spectral noise was reduced by Savitzky-Golay（S-G）smoothing algorithm.The characteristic wavelengths（CWs）were selected withsuccessive projections algorithm（SPA）and Competitive adaptive reweighted sampling（CARS）.At the same time,7 Spectral indices（SIs）were imported from the perspective of fruit varied physiological components.Three partial least squares discriminant analysis（PLS-DA）models were established based on the CWs selected by SPA and CARS and the introduced SIs,and the classification results of three models were compared.The results show that the discrimination accuracy of PLS-DA models based on two kind of CWs（selected by SPA and CARS,respectively）and SIs wereseparately 97.27%,95.45%,and 98.18%.For the purpose of showing the discriminant results intuitively,a regression equation of the discriminant vector Y was fitted with SIs joint its PLS-DA regression coefficients,and the discriminant results were visually displayed by different colors in accordance with the rule that the corresponding category of the maximum value in Yis the sample belonging category.This study will contribute some proposals to visual grading of winter jujube maturity,and the imported SIs parameters will provide technical support for the manufacture of device that suitable for multiple fruits maturity sorting.