基于DTW距离的时序NDVI数据植被信息提取——以秦巴山区为例

Extraction of vegetation information using adding windows DTW distance with NDVI time series data

秦巴山区是我国重要的生态屏障,对该区的植被信息提取开展研究,可为区内生态服务功能及自然资源开发利用提供基础数据。通过加窗处理改进DTW距离相似性算法,结合临近度模糊分类方法对2005—2014年的MODIS NDVI时序数据进行植被信息提取。首先利用S-G滤波对MODIS NDVI时序数据进行重建;再利用2013年的采样数据构建各类植被的标准NDVI时序曲线,逐像元计算与标准NDVI时序曲线的加窗DTW距离,利用临近度模糊分类实现植被信息提取;最后验证提取精度。结果表明,算法具有较高的运行效率,可避免错误匹配,以较高的精度(总体精度83.8%,kappa系数0.77)实现长时间序列的植被信息提取。

Qinling-Bashan Mountain area is an ecological barrier.The vegetation information extraction research can provide base data for ecological service function and natural resources exploitation.DTW distance similarity measure is improved by adding windows,and vegetation information is extracted by proximity-fuzzy classification from MODIS NDVI in the years of 2005 to 2014.S-G filter method is first applied to weakening the noise and reconstructing the NDVI time series.Each type of vegetation standard NDVI time series curves is generated from the 2013 sample data.For each to-be-classified pixel,a quantitative similarity between its time-curve and standard time series curves is calculated using adding windows DTW,then vegetation information is extracted in research area.Finally,validate data is used to verify the extraction accuracy.The result shows that the algorithm has high efficient and can avoid mismatch,and can realize long time serial vegetation information extraction with high accuracy（overall accuracy：83.8%,Kappa coefficient：0.77）.