欢迎访问《麦类作物学报》编辑部网站

作者	单位
丁静，梁琨，韩东燊，徐剑宏，沈明霞	（1.南京农业大学工学院/江苏省现代设施农业技术与装备工程实验室，江苏南京 210031；2.江苏省农业科学研究院农产品质量安全与营养研究所，江苏南京 210014）

摘要点击次数: 737

全文下载次数: 996

中文摘要:

为实现小麦呕吐毒素含量快速检测，采集了120个小麦赤霉病籽粒样本的高光谱图像，分别使用连续投影算法（SPA）、区间组合优化结合连续投影算法（ICO-SPA）对1 000~2 500 nm范围的光谱进行特征波段提取，结合偏最小二乘回归（PLSR）、多元线性回归（MLR）和最小二乘支持向量机回归（LS-SVR）模型比较了基于三种特征变量输入的模型预测效果。结果表明，ICO-SPA提取出的22个特征波段能够反映病粒样本中淀粉、蛋白质、脂肪、纤维素等大分子含量的差异，比单独使用SPA可多提取淀粉含量信息，少提取已经被控制在同一水平的水分含量信息，能更全面真实地反映小麦感染赤霉病后内部大分子成分含量的变化，同时减少水分含量信息对近红外模型的干扰。基于ICO-SPA所选变量的建模效果优于SPA，其中以ICO-SPA-MLR效果最优，预测集相关系数、均方根误差和相对分析误差分别为0.921、0.375 mg·kg^-1和 2.789。这说明基于近红外高光谱技术结合ICO-SPA-MLR进行小麦赤霉病籽粒呕吐毒素定量检测是可行的。

英文摘要:

In order to achieve rapid quantitative detection of vomiting toxin in the wheat,hyperspectral images of 120 wheat samples were collected in this paper. Two feature extraction methods of successive projections algorithm(SPA) and interval combination optimization combined with successive projections algorithm(ICO-SPA) were used to select the characteristic wavelengths from 1 000 nm to 2 500 nm. The two inputs of characteristic variables were compared by combining with the partial least squares regression(PLSR),multiple linear regression(MLR) and least squares support vector machine regression(LS-SVR) models. The experiment results showed that twenty-two characteristic bands extracted by ICO-SPA could reflect the difference in the content of macromolecules,such as starch,protein,fat and cellulose in samples. With the same water content level,information on starch content was extracted by ICO-SPA compared to SPA algorithm. The content information extracted by ICO-SPA was more fully and truly reflecting the changes of internal macromolecular composition content after wheat scab,and also reduces the interference of different moisture content on the near-infrared spectrum. The results obtained by ICO-SPA were always better than SPA under the three modeling methods,and the ICO-SPA-MLR model works best,in which r_p,RMSEP and RPD were 0.921, 0.375 mg·kg^-1 and 2.789,respectively. The feasibility was validated by this paper that vomiting toxin content in wheat could be detected based on near-infrared hyperspectral technology combined with ICO-SPA-MLR model.

查看全文查看/发表评论下载PDF阅读器

关闭