可见-近红外光谱的小麦硬度预测模型预处理方法的研究

发布时间：2018-10-09 11:38

【摘要】：硬度是评价小麦品质的一个重要质量参数,对小麦的分类、最终用途以及小麦籽粒组成的研究都非常重要。为实现小麦硬度的快速、准确检测,在详细分析小麦籽粒成分对红外光吸收特性的基础上,研究建立径向基函数(RBF)神经网络模型实现对未知样品硬度的准确检测,并着重分析了不同光谱信号预处理方法对模型预测精度的影响。从各小麦主产区收集111个小麦样品,扫描样品获得可见-近红外光谱,采用马氏距离判断并剔除异常光谱;利用优化后的SPXY划分样品集合,得到校正集84个样品,预测集24个样品;利用连续投影算法(SPA)从262个光谱波点中提取47个特征光谱;分别使用一阶导数、二阶导数和标准正态变量变换(SNV)及其不同组合对光谱进行预处理,验证不同预处理方法之间的相互影响,寻找最优的预处理方法组合。校正集预处理后的特征光谱数据作为RBF模型的输入,采用硬度指数法测定的对应样品硬度作为输出建立模型。预测结果显示当采用SNV和SPA处理光谱数据时模型的效果达到最优,评价指标判别系数(R~2)、预测标准差(SEP)和相对分析误差(RPD)可分别达到0.90,3.02和3.11,表明基于可见-近红外光谱的RBF神经网络模型能够准确地预测小麦的硬度,与传统检测方法相比具有方便、快速、无损等优点,为小麦硬度的检测提供一条更为便捷与实用的方法。
[Abstract]:Hardness is an important quality parameter to evaluate wheat quality. It is very important to study the classification, end-use and grain composition of wheat. In order to detect wheat hardness quickly and accurately, a radial basis function (RBF) (RBF) neural network model was established to detect the hardness of unknown samples on the basis of detailed analysis of infrared light absorption characteristics of wheat grain components. The influence of different spectral signal preprocessing methods on the prediction accuracy of the model is analyzed. 111 wheat samples were collected from the main wheat producing areas, the visible near infrared spectra were obtained by scanning the samples, the abnormal spectra were judged and eliminated by Markov distance, 84 samples were obtained by the optimized SPXY. A continuous projection algorithm (SPA) is used to extract 47 characteristic spectra from 262 spectral wave points, and the first derivative, second derivative and standard normal variable transform (SNV) and their different combinations are used to preprocess the spectrum, respectively. Verify the interaction between different pretreatment methods and find the best combination of preprocessing methods. The pre-processed characteristic spectral data of the calibration set is used as the input of the RBF model, and the hardness of the corresponding sample measured by the hardness index method is used as the output to establish the model. The prediction results show that when the spectral data are processed by SNV and SPA, the effect of the model is optimal. The predictive standard deviation (SEP) and the relative analysis error (RPD) were 0.90 and 3.11, respectively, which indicated that the RBF neural network model based on the visible and near infrared spectra could accurately predict the hardness of wheat. Compared with the traditional testing method, it has the advantages of convenience, rapidity and nondestructive, which provides a more convenient and practical method for wheat hardness detection.
【作者单位】：黑龙江省电子工程高校重点实验室黑龙江大学;农业部谷物及制品质量监督检验测试中心(哈尔滨);
【基金】：哈尔滨市青年科技创新人才研究专项基金项目(2012RFQXN119) 国家现代农业技术体系任务书项目(CARS-3-1-6)资助
【分类号】：O657.33;TS210.7

【相似文献】