一种大规模样本数据的特征提取方法

A METHOD OF EXTRACTING FEATURES FROM LARGE SCALE PATTERNS

针对沙尘暴样本数据的特点 ,根据专家经验依次通过聚类分析、建立典型模式类、计算中心场 ,再以样本与中心场的距离作为样本的特征 ,成功地在每个样本的几百个数据中提取到 40个特征。通过对提取特征的统计检验 ,说明了本文所用方法的有效性 ;同时 ,提出了一种建立在K L变换基础上的特征综合方法 。

The statistic and analysis of high-dimension data is the important difficulty in the present application and theory of statistics. According to the characteristic of patterns data and expert's experience, clustering and selecting typical patterns and calculating central fields are performed in sequence. Then using the distances between every sample and the central fields as features of the sample, 40 pieces of features are extracted from hundreds of data successfully. The validity of the method through which features are extracted is proved by the results of statistic analysis for the features. At the same time, the method to integrate features based on K-L transform is brought forward. which help us to fulfill the test to the difference of multi-dimension features between pattern classes.