基于支持向量机的花生苗期抗旱指标筛选

Screening Drought Resistance Indexes of Peanut in the Seedling Stage Based on Support Vector Machine

筛选花生抗旱性指标,培育抗旱能力强的花生品种,传统线性筛选方法不能反映花生抗旱性的非线性特性,筛选得到的指标符合度有所局限。本研究基于支持向量机(SVM)构建了一种非线性筛选模型,利用SVM交叉测试逐个评估各个特征因子,以MSE最小原则逐步剔除对模型有不利影响的特征因子,并以29个花生品种为例,对其苗期的13个形态、生理特征进行再分析和甄别,得到主茎高、分枝数、地上部干重、叶片干重、比叶面积共5个特征指标,以此特征指标构建的线性与非线性模型留一法精度均高于经线性筛选特征的模型预测精度,实验结果表明了经非线性筛选得到的特征因子的准确性,最后经基于F测验的SVM回归显著性测验与单因子重要性分析进一步验证了保留指标的有效性。以此保留特征构建非线性预测模型能为花生抗旱育种工作提供有效指导。

The aims were to screen an index of peanut （Arachis bypogaea） drought resistance and breed new varieties of peanut with high drought resistance capability. Conventional linear selection method could not reflect nonlinear characteristics of drought resistance on peanut and the index conformity degree was limited. In this study, a kind of nonlinear feature selection model based on support vector machine had been set up, characterization factors were assessed individually by SVM cross validation, characterization factors which had adverse effects on model were eliminated by the minimum MSE. Taking 29 peanut varieties as examples, 5 characteristic indexes, including length of main stem, No. of branches, shoot dry weight, leaf dry weight and specific leaf area, were obtained by analyzing-and screening 13 configuration features and 13 physiological features in the seedling stage. Leave-one-out accuracy of nonlinear and linear feature selection model based on these characteristic indexes was higher than that of linear feature selection model. The experimental results showed the accuracy of characterization factors screened by nonlinear feature selection. Finally, test of SVM regression significance analysis based on F-test and analysis of one-factor importance validated the importance of indexes reserved. Nonlinear prediction model based on these indexes reserved could provide effective guidance for drought resistance and breeding of peanut.