摘要
时间序列影像能够反映植被的物候信息,有助于大幅度提高植被聚类精度,尤其对于单时相影像上生长特性相似的树种效果尤其明显。动态时间规整算法(DTW)能够解决不等长时间序列的匹配问题,且能够抵抗噪声造成的时间序列中出现的异常值,从而取得更好的相似特征匹配效果。基于6时间序列GF-1影像数据,分别采用DTW和欧式距离(ED)进行时间序列的相似性度量,然后将DTW距离和ED距离运用到K-Means算法中,完成对图像树种的聚类。结果表明:基于DTW-K Means的时间序列遥感影像分类方法能够适用于树种分类,总精度为92.21%,Kappa系数为0.90,均高于ED-K Means方法。
Phenologieal information of vegetation reflected by time serious remote sensing images contrilmtes to greatly improve the vegetation clustering accuracy, especially for species with the similar growth characteristics in the single phase image. Dynamic time warping algorithm （DTW） is able to solve the matching problem of unequal time series, and can resist outliers in time series caused by the cloud or other noise, so as to achieve better similarity matching effect. Based on six GF-1 time series images, DTW and ED （Euclidean distance） , respectively, were used to measure the similarity of time series, and then the two distances were used in K-Means algorithm to complete the tree species clustering. Time series remote sensing image classification method based DTW-K Means can he applied in tree species classification with the accuracy of 92.21%, and Kappa eoetficient of 0.90, which were higher than those by the ED-K Means method.
出处
《东北林业大学学报》
CAS
CSCD
北大核心
2017年第5期56-61,共6页
Journal of Northeast Forestry University
基金
黑龙江省博士后基金资助(LBH-Z10279)
中央高校基本科研业务费专项资金项目(2572014CB20)
哈尔滨市应用技术研究与开发项目(2015RQQXJ071)
关键词
动态时间规整算法
时间序列
树种分类
欧式距离
Dynamic time warping
Time series
Tree species classification
Euclidean distance
