用于时间序列聚类分析的小波变换和特征量提取方法  

Wavelet Transform and Statistical Characteristics Extraction Applied to Times Series Clustering Analysis

作　　者：刘汉丽[1] 裴韬[2] 周成虎[2] 

机构地区：[1]武汉科技大学汽车与交通工程学院,湖北武汉430081 [2]中国科学院地理科学与资源研究所资源与环境信息系统国家重点实验室,北京100101

出　　处：《测绘科学技术学报》2014年第4期372-376,382,共6页Journal of Geomatics Science and Technology

基　　金：国家重点基础研究发展计划项目(2006CB701305);湖北省自然科学基金项目(2011CDB235)

摘　　要：利用小波变换的多尺度分辨功能,对时间序列数据进行多次Haar小波变换,将时间序列分解为尺度分量和细节分量;通过保留尺度分量,提取时间序列的趋势信息,并结合统计特征量计算,将时间序列的趋势信息与几种统计特征量组合在一起,构成SOM神经网络的输入向量,对时间序列进行聚类分析。通过在几类模拟时间序列数据上进行实验分析,取得了较好的实验效果。并将此方法应用到基于MODIS遥感影像的林地植被提取中,获得了较高的提取精度。The functionality of the multi-resolution of wavelet transform is utilized and a multiple wavelet transform is performed on time series data in this paper. By wavelet transform, the time series can be decomposed into scale components and detail components. By retaining the scale components and removing the detail components, the trend information of time series is extracted. And then, the statistical characteristics data are calculated from the time series. In order to perform clustering analysis on the time series, the trend data and the statistical characteristics of time series are combined to form the input vector of SOM neural network. After that, a series of experiments for clustering analysis is conducted on four sets of simulated time series and one set of MODIS remote sensing vegetation data. The results of clustering analysis show that the proposed method is effective.

关 键 词：时间序列 HAAR小波变换 统计特征量提取 MODIS遥感影像 NDVI数据 SOM神经网络 

分 类 号：P237[天文地球—摄影测量与遥感]