环境星归一化植被指数时间序列滤波算法比较

摘要

由于环境卫星成像条件和卫星平台的限制，针对环境卫星归一化植被指数（HJ NDVI）时间序列中存在较多噪音的现象，该文比较分析了 Savitzky-Golay（S-G）滤波法、非对称高斯函数拟合法（AG）、双逻辑曲线拟合法（DL）和时间序列谐波分析法（Hants）4种滤波算法。建立了南京市典型植被类型区域的 HJ NDVI 时间序列，对4种滤波方法进行实验。对比纯像元样点、样区的滤波结果以及滤波后5类典型植被的分类精度，评价4种滤波方法的滤波效果，并利用 MODIS NDVI 时序数据验证结果可靠性。结果表明：滑动窗口大小为5的 S-G 滤波的滤波效果最佳。该研究结论为基于 HJ NDVI 时间序列的应用研究滤波方法选择提供参考。%HJ-1A/B NDVI (HJ NDVI)time-series data possess relatively high spatio-temporal resolution.However,due to the restrictions of HJ satellite's imaging conditions and limits of the satellite platform,HJ NDVI time series are polluted by noises.Finding the most optimal filtering algorithm can be the key to solve this problem.This study introduced four kinds of filters including the asymmetric Gaussian fitting method (AG),the double logistic curve fitting method (DL),the Savitzky-Golay (SG)filtering method and the harmonic analysis of NDVI time-series (Hants).HJ NDVI time series of typical vegetation region in Nanjing city are established,and filtering experiments of the four filters are performed on them.Filtering results of the pure pixel sample points and the sample zones as well as the five typical vegetation's classification accuracy of the filtered data are compared to evaluate the performance of the four filters and finally MODIS NDVI tine-series data are included to test the reliability of the four filters.The results show that the S-G filter with a sliding window edge length of 5 has the best performance.Conclusions of this study can provide reference for filter selection of HJ NDVI time series based research.