中国农学通报
中國農學通報
중국농학통보
CHINESE AGRICULTURAL SCIENCE BULLETIN
2013年
15期
31-36
,共6页
花生%抗旱性%支持向量机%筛选
花生%抗旱性%支持嚮量機%篩選
화생%항한성%지지향량궤%사선
peanut%drought resistance%support vector machine%screening
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。
篩選花生抗旱性指標，培育抗旱能力彊的花生品種，傳統線性篩選方法不能反映花生抗旱性的非線性特性，篩選得到的指標符閤度有所跼限。本研究基于支持嚮量機(SVM)構建瞭一種非線性篩選模型，利用SVM交扠測試逐箇評估各箇特徵因子，以MSE最小原則逐步剔除對模型有不利影響的特徵因子，併以29箇花生品種為例，對其苗期的13箇形態、生理特徵進行再分析和甄彆，得到主莖高、分枝數、地上部榦重、葉片榦重、比葉麵積共5箇特徵指標，以此特徵指標構建的線性與非線性模型留一法精度均高于經線性篩選特徵的模型預測精度，實驗結果錶明瞭經非線性篩選得到的特徵因子的準確性，最後經基于F測驗的SVM迴歸顯著性測驗與單因子重要性分析進一步驗證瞭保留指標的有效性。以此保留特徵構建非線性預測模型能為花生抗旱育種工作提供有效指導。
사선화생항한성지표，배육항한능력강적화생품충，전통선성사선방법불능반영화생항한성적비선성특성，사선득도적지표부합도유소국한。본연구기우지지향량궤(SVM)구건료일충비선성사선모형，이용SVM교차측시축개평고각개특정인자，이MSE최소원칙축보척제대모형유불리영향적특정인자，병이29개화생품충위례，대기묘기적13개형태、생리특정진행재분석화견별，득도주경고、분지수、지상부간중、협편간중、비협면적공5개특정지표，이차특정지표구건적선성여비선성모형류일법정도균고우경선성사선특정적모형예측정도，실험결과표명료경비선성사선득도적특정인자적준학성，최후경기우F측험적SVM회귀현저성측험여단인자중요성분석진일보험증료보류지표적유효성。이차보류특정구건비선성예측모형능위화생항한육충공작제공유효지도。
The aims were to screen an index of peanut (Arachis hypogaea) 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.
