基于Si-cPLS的小麦种子发芽率近红外模型优化研究

发布时间：2018-07-17 20:13

【摘要】：为了提高近红外光谱技术快速测定小麦种子发芽率的准确度和稳健性,比较分析了基于全光谱的单一偏最小二乘(PLS)模型和多模型共识PLS模型(cPLS)性能,并提出了基于特征光谱的多模型共识PLS模型(Si-cPLS)。实验收集84份小麦种子,通过SPXY法将样本集划分为训练集样本66个,预测集样本18个。从训练集中随机抽取50个样本作为校正集建立一系列PLS子模型,选取其中性能较好100个子模型作为成员模型建立cPLS模型,取成员模型预测结果的均值来分析未知样本。在此基础上,采用组合间隔偏最小二乘法(SiPLS)筛选特征谱区建立多模型共识PLS模型(Si-cPLS)。各模型均采用均值中心化预处理方法,PLS模型、cPLS模型以及Si-cPLS模型对预测集的小麦种子发芽率进行50次重复预测的平均相关系数r分别为0.935,0.949和0.967,平均预测均方根误差RMSEP分别为13.735%,12.533%和10.273%,RMSEP的标准偏差分别为1.144%,0.096%和0.08%。实验结果表明cPLS模型较单一PLS模型更加稳定可靠,而基于特征光谱的Si-cPLS模型则进一步提高了cPLS的稳定性与预测精度,为建立性能优秀的小麦种子发芽率近红外模型提供了新思路。
[Abstract]:In order to improve the accuracy and robustness of fast determination of wheat seed germination by near infrared spectroscopy (NIR), the performance of single partial least squares (PLS) model based on full spectrum and consensus PLS model (cPLS) based on multiple models were compared and analyzed. A multi-model consensus PLS model (Si-cPLS) based on feature spectrum is proposed. 84 wheat seeds were collected. The samples were divided into 66 samples from training set and 18 samples from prediction set by SPXY method. A series of PLS submodels were established by randomly selecting 50 samples from the training set as correction sets, and 100 sub-models with good performance were selected as member models to establish cPLS model. The average value of predicted results of member models was taken to analyze unknown samples. On this basis, a multi-model consensus PLS model (Si-cPLS) is established by combining interval partial least square (SiPLS) method to filter the characteristic spectral region. The average correlation coefficients of 50 times repeated prediction of seed germination rate of wheat seed from the predicted set were 0.935 ~ 0.949 and 0.967, respectively. The mean correlation coefficients were 0.935 ~ 0.949 and 0.967 for wheat seed germination rate prediction by means of cPLS model and Si-cPLS model respectively, and the mean root mean square error (RMS) was predicted by the mean centralization pretreatment method. The standard deviations of RMSEP were 13.735% 12.533% and 10.273%, respectively. The standard deviations of RMSEP were 1.144% and 0.08%, respectively. The experimental results show that the cPLS model is more stable and reliable than the single PLS model, while the Si-cPLS model based on the characteristic spectrum can further improve the stability and prediction accuracy of cPLS. It provides a new idea for establishing the near infrared model of wheat seed germination rate with excellent performance.
【作者单位】：北京工商大学计算机与信息工程学院;土壤植物机器系统技术国家重点实验室;中国农业机械化科学研究院;
【基金】：国家国际科技合作专项项目(2014DFA31660) 土壤植物机器系统技术国家重点实验室开放课题(2014-SKL-05) 北京工商大学两科基金培育项目(19008001110)资助
【分类号】：S512.1;O657.33

【相似文献】