近紫外-可见-短波近红外多光谱成像数据的糯玉米种子热损伤粒的无损快速鉴别

为对糯玉米种子热损伤粒进行无损快速鉴别并探索热损伤过程对糯玉米种子的影响，以糯玉米种子“京科糯2000”为例，用Videometer近紫外-可见-短波近红外多光谱成像仪分别以胚面向上和胚面向下方式采集糯玉米种子对照组及热损伤组多光谱成像数据，分别提取胚面向上胚部、胚面向上胚乳部、胚面向下胚乳部单点多光谱数据，并对胚面向上胚部和胚面向上胚乳部多光谱数据做初级融合；对多光谱数据进行基线校正预处理后计算各光谱数据样本标准差，进而通过光谱数据样本标准差的变化分析热损伤对糯玉米种子各部位的影响；基于多光谱数据采用偏最小二乘-判别分析算法建立糯玉米种子热损伤粒无损鉴别模型，对所建模型进行全交互验证，并与近红外光谱数据模型比较。结果表明，热损伤对糯玉米种子胚、胚乳有不同的影响，多光谱数据和近红外光谱数据表现出一致的变化趋势。采用多光谱数据建立热损伤粒鉴别模型，各模型前3主成分得分3D散点图中，对照组和热损伤组样品表现出一定的分离趋势，校正数据正确率在96%~100%之间，交互验证数据正确率在92%~100%之间，其中，糯玉米种子胚面向上胚部光谱和胚乳部光谱初级融合数据建模效果最好，校正数据正确率100%，交互验证数据正确率在98%~100%之间。作为对比，采用近红外光谱数据建立糯玉米种子热损伤粒偏最小二乘-判别分析模型，胚面向上、胚面向下以及二者初级融合数据模型的前3主成分得分3D散点图中，对照组和热损伤组样品表现出较好的分离趋势，各模型校正数据、交互验证数据正确率皆为100%。本研究表明，采用近紫外-可见-短波近红外多光谱成像对糯玉米种子的热损伤粒进行无损快速识别具有较好的可行性，多光谱成像数据各变量样本标准差和近红外光谱数据各变量样本标准差呈现一致的规律；采用胚和胚乳融合多光谱数据所建模型在各模型中具有更高的正确率。

Non-Destructive Identification of the Heat-Damaged Kernels of Waxy Corn Seeds Based on Near-Ultraviolet-Visible-Shortwave and Near-Infrared Multi-Spectral Imaging Data

In this research, took the waxy corn seed “Jingkenuo 2000” as an example to research the identification of the the heat-damaged kernels of waxy corn seeds quickly and non-destructively and explore the effect of heat damage on waxy corn seeds. The multi-spectral imaging data of the control group and heat-damaged group were collected by Videometer near-ultraviolet-visible-short-wave near-infrared multi-spectral imager with embryo facing up and embryo facing down respectively. The single-point multi-spectral data from the embryo with embryo facing up, endosperm with embryo facing up and down were extracted respectively, while the multi-spectral data from the embryo and endosperm with embryo facing up were fused primarily. The standard deviations of all spectral variables were calculated after the baseline preprocessing to the multi-spectral data to analyse the effect of heat damage on waxy corm seeds according to the change of standard deviation of the data. Based on the multi-spectral data, the non-destructive identification models of the heat-damaged waxy corn seeds were developed by partial least square - discriminant analysis (PLS-DA) algorithm, which was compared with the models developed based on near-infrared spectra data. The result indicated that heat damage results in different effects on embryo and endosperm, however, the multi-spectral data and near-infrared data show the same trend of change. Based on the multi-spectral data, the identification models of the heat-damaged kernels of waxy corn seeds are developed. In the 3D scatter score plots of each model’s first three principal components, the samples of the control group and the heat-damaged group show a certain separation trend. The accuracy of calibration data is between 96% and 100%, while the accuracy of cross-validation data is between 92% and 100%. The model developed by the fusion data of embryo and endosperm spectra with embryo facing up is of a higher accuracy, of which, the accuracy of calibration data is 100 %, and that of cross-validation data are between 98% and 100%. In contrast, the PLS-DA models of the heat-damaged waxy corn seeds are developed by near-infrared spectra data. In the 3D scatter score plots of the first three principal components of the models developed by the data of embryo facing up, embryo facing down, and the fusion of the two, the samples of the control group and the heat-damaged group show a good separation trend of which, the accuracy of the calibration and cross-validation are all 100 %. This research demonstrated that it is of good feasibility to identify the heat-damaged kernels of waxy corn seeds by near-ultraviolet-visible-short-wave near-infrared multi-spectral imaging technology rapidly and non-destructively. The standard deviation data of the multi-spectral variables are consistent with those of near-infrared spectral data. The calibration model by the fusion data of embryo and endosperm is of a higher accuracy for the multi-spectral data.