壳斗科化石是北半球新生代地层的优势分子，然而由于其属种众多，叶片属种间特异性不明显，该科叶片化石的鉴定是古植物学研究的难点之一。面对化石标本不完整的性状特征信息，如何剔除冗余特征，重点考虑具有鉴定意义的标本特征，就显得非常必要。本研究通过调研多种降维算法，考虑到植物特征编码及赋值的数值特性，选用主成分分析法，以壳斗科青冈亚属植物叶片为例，对可能在化石叶片上观察到的22个性状特征(变量)进行降维处理，挑选出对于壳斗科化石叶片分类鉴定起主要作用的10个性状特征，并将其应用于化石鉴定进行验证。结果表明，仅考虑经过主成分分析法压缩挑选的10个主要性状特征，仍然能够实现壳斗科化石植物的准确鉴定。主成分分析法应用于壳斗科性状特征的降维处理效果良好，剔除冗余特征对标本鉴定结果没有影响。

Abundant fossil records show that the Fagaceae has remained a dominant component in the Northern Hemisphere since the Cenozoic.However,due to the large number of living species,it is not easy to identify leaves to a particular species.Consequently,the identification of fossil leaves belonging to the Fagaceae is problematic. In the face of incomplete characteristics information about fossil specimens,it is necessary to eliminate redundant features and to focus on the features which can determine the identification of fossil specimens.Based on an investigation of various dimensionality reduction algorithms,in which the numerical characteristics of encoding and assign-ment of plants were considered,it is shown that principal component analysis is the best method to choose primary identifying characteristics.In this study,taking Quercus subgenus Cyclobalanopsis(Fagaceae)as an example,we use principal component analysis to eliminate subordinate features from the 22 traits which may be observed in the fossil leaves.The result demonstrates that 10 characteristics play a major role in the classification and identification of Fagaceae fossil leaves.The outcome is applied to the identification of Quercus subgenus Cyclobalanopsis fossil specimens collected from the Late Miocene of Tiantai,Zhejiang,eastern China.The results show that even if only the 10 main characters picked out by principal component analysis compression are utilized,these can achieve an accurate identification of fossil Fagaceae.Adding the redundant features does not improve the taxonomic resolution.The principal component analysis method is verified as the most effective method to eliminate the subordinate Fagaceae leaf traits.

陕西省自然科学基础研究计划项目(2016JQ4018)；国家自然科学基金项目(41604113)资助