| 王国梁,赵媛,刘敏,郭二虎,王瑞,范惠萍,李瑜辉,张艾英.利用特征波段提取及结合机器学习对小米淀粉的高光谱检测研究[J].中国粮油学报,2024,39(4):149-157 |
| 利用特征波段提取及结合机器学习对小米淀粉的高光谱检测研究 |
| Research on the Hyperspectral Detection Methods of Starch of Millet by Feature Bands Extraction Combined with Machine Learning Agorithm |
| 投稿时间：2023-03-01 修订日期：2023-06-18 |
| DOI： |
| 中文关键词: 小米淀粉 高光谱检测 特征波段提取联用 机器学习 |
| 英文关键词:millet starch hyperspectral feature bands extraction sequential combination machine learning algorithm |
| 基金项目:杂粮种质资源创新与分子育种国家实验室(202204010910001-13)，国家现代农业产业技术体系建设专项(CARS-06-14.5-A21)，山西省现代农业产业技术体系谷子体系(2023CYJSTX04-04) |
|
| 摘要点击次数: 63 |
| 全文下载次数: 0 |
| 中文摘要: |
| 运用高光谱检测技术实现小米淀粉的快速检测在小米定级、定价及降低加工成本中具有重要意义。本研究基于高光谱检测技术，采用化学计量学及机器学习相关知识对小米直链、支链淀粉含量进行检测，并提出特征波段提取联用预处理方法及Logistic结合COOT （coot optimization algorithm）优化算法。结果表明采用特征波段提取联用算法建立的PLSR（partial least squares regression）模型能够在减少波段冗余情况下不影响模型预测精度，其中直链淀粉较好模型为MSC（multiplicative scatter correction）-RF（random frog）-IRIV（iteratively retains informative variables）-PLSR，支链淀粉较好模型为MSC-CARS（competitive adaptive reweighted sampling）- IRIV-PLSR。为了进一步提高模型预测精度，基于最佳预处理算法结合Logistic-COOT建立BP（back propagation）预测模型能够较好的预测小米直链、支链淀粉的含量，模型评价直链、支链淀粉相关系数 (correlation coefficient, R)、均方根误差(root mean squared error, RMSE)、相对分析误差(relative percent deviation, RPD)分别为0.74，1.19，1.51；0.72，5.25，1.40。这一结论能够为小米其他营养成分的高光谱检测及产品分类、定级等提供理论参考。 |
| 英文摘要: |
| The rapid detection of millet starch by hyperspectral technology is of great significance in millet grading, pricing and reducing processing costs. In this paper, based on hyperspectral detection technology, the content of amylose and amylopectin in millet was detected by using chemometrics and machine learning algorithm, and the pretreatment methods feature bands extraction sequential combination and Logistic combined with coot optimization algorithm optimization algorithm were proposed. The results show that partial least squares regression model established by feature bands extraction sequential combination can reduce the bands redundancy without affecting the prediction accuracy of the model, the better prediction model for amylose was MSC-RF-IRIV-PLSR, and the better prediction model for amylopectin was MSC-CARS-IRIV-PLSR. In order to further improve the accuracy of the model prediction, BP model based on the best pretreatment method combined with Logistic-COOT could predict the content of amylose and amylopectin in millet, R (the correlation coefficient), RMSE (root mean squared error) and RPD (relative percent deviation) of amylose & amylopectin were 0.74, 1.19, 1.51; 0.72, 5.25, 1.40, respectively. This conclusion can provide theoretical reference for hyperspectral in other nutritional components of millet and product classification or grading. |
| 关闭 |
|
|
|
