欢迎访问《麦类作物学报》编辑部网站

作者	单位
刘洁琼，罗斌，张晗，康凯，陈泉，邱朝阳	（1.新疆农业大学机电工程学院，新疆乌鲁木齐 830052； 2.北京市农林科学院智能装备技术研究中心，北京 100097）

摘要点击次数: 73

全文下载次数: 41

中文摘要:

小麦籽粒类胡萝卜素含量关系到小麦制品的颜色外观和商品价值，是小麦育种过程中的重要指标之一。目前检测小麦籽粒类胡萝卜素含量的方法主要有紫外分光光度法、薄层色谱法、高效液相色谱法等化学方法，成本高且耗时耗力。为实现小麦籽粒类胡萝卜素含量的快速无损预测，将近红外光谱技术与化学计量技术相结合，利用留出法（hold-out method，HOM）、K折交叉验证（K-fold cross-validation，KFCV）和时间序列划分（time series split，TSS）3种样本集划分方法，Savitzky-Golay平滑（Savitzky-Golay smoothing，SG）、多元散射校正（multivariate scatter correction，MSC）、标准正态变量变换（standard normal variable transformation，SNV）和趋势校正（trend correction，TC）4种光谱预处理方法，方差阈值特征选择（variance threshold feature selection，VTFS）、SelectKBest特征选择（SelectKBest feature selection，SKB）、递归特征消除（recursive feature elimination，RFE）分别与主成分分析（PCA）算法融合的3种特征选择方法，建立偏最小二乘回归（partial least squares regression，PLSR）、支持向量机回归（support vector machine regression，SVR）、梯度提升回归（gradient boosting regression，GBR）三种模型，比较分析了不同模型预测精度。结果表明，样本集最佳划分方法为留出法，光谱最佳预处理方法为SG预处理，最佳特征选择方法为PCA-SKB，最优模型为PCA-SKB-GBR，校正集和预测集决定系数R²分别为0.99和0.89，均方根误差RMSE分别为0.03和0.34 μg·g^-1，剩余预测偏差RPD为3.01。因此，基于留出法划分样本集、SG光谱预处理和PCA-SKB特征选择方法，建立PCA-SKB-GBR模型，可实现小麦籽粒类胡萝卜素含量快速高效预测。

英文摘要:

The carotenoid content in wheat kernels serves as a crucial determinant of color appearance and commercial value, representing an important indicator in wheat breeding. While conventional detection methods(ultraviolet spectrophotometry, thin-layer chromatography, and high-performance liquid chromatography) demonstrate accuracy, their operational costs and time requirements prove prohibitive for large-scale applications. This study provided a rapid non-destructive prediction method through the integration of near-infrared spectroscopy(NIRS) with chemometric techniques. We systematically evaluated:(1) three dataset partitioning methods hold-out method(HOM), K-fold cross-validation(KFCV), time series split(TSS); (2) four spectral preprocessing techniques ［Savitzky-Golay smoothing(SG), multiplicative scatter correction(MSC), standard normal variate transformation(SNV), trend correction(TC)］， three feature selection methods were obtained by fusing variance threshold feature selection(VTFS), SelectKBest(SKB), and recursive feature elimination(RFE) with principal component analysis(PCA), respectively. Subsequent modeling employed partial least squares regression(PLSR), support vector machine regression(SVR), and gradient boosting regression(GBR) were used to build different prediction models. The results showed that HOM achieved optimal dataset partitioning, SG preprocessing provided superior spectral enhancement, and PCA-SKB feature fusion delivered maximum information retention. The optimized was PCA-SKB-GBR. In the validation set and prediction set, R² were 0.99 and 0.89 with RMSE of 0.04 and 0.34 μg·g^-1, respectively. The residual prediction deviation(RPD) reached 3.01. Therefore, based on the hold-out method of sample set, SG spectral pretreatment and PCA-SKB feature selection method, the PCA-SKB-GBR model can be used to predict the carotenoid content of wheat grain quickly and efficiently.

查看全文查看/发表评论下载PDF阅读器

关闭