ESTARFM相似像元选取方法的改进研究

Study of the Improved Similar Pixel Selection Method on ESTARFM

ESTARFM(Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model)是一种经典的基于权重滤波的时空融合算法,它在众多领域得到广泛应用。相似像元选取是其一个重要步骤,ESTARFM模型中相似像元选取过程受搜索框大小和分类数影响,当前的研究中搜索框大小的设定较为统一,而分类数大小设定缺乏统一性。为降低ESTARFM算法中分类数对算法性能的影响,将STNLFFM(A Spatial and Temporal Nonlocal Filter-Based Data Fusion Method)中相似像元选取方法与ESTARFM模型相结合,提出改进的ESTARFM_NL模型。研究设计了两组不同时相变化条件下的数据进行对比分析。结果表明:ESTARFM_NL与ESTARFM融合结果相对误差直方图总体分布趋近一致,同时利用平均相对误差和相关系数对融合结果进行评价,发现两种算法之间精度差异较小,表明两种算法融合精度相当;对比两种算法运算效率,发现ESTARFM_NL运行时间能够得到大幅缩减。因此,ESTARFM_NL为大区域或长时间序列遥感数据的时空融合提供了一种可选择的融合方案。

The ESTARFM(Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model)is a classic spatiotemporal filter-based algorithm,which is used in the many fields.The similar pixel selection process in the ESTARFM model is affected by the size of the size of search window and the number of classifications.In the current study,the size of the search windows is more uniform,and the number of classifications lacks uniformity.In order to reduce the influence of the number of classifications in the ESTARFM algorithm on the performance of the algorithm.The similar pixel selection method in the STNLFFM(A Spatial and Temporal Nonlocal Filter-Based Data Fusion Method)combined with the ESTARFM model to propose the ESTARFM_NL model.The study designed two sets of data under different conditions of phase change for comparative analysis.The results show that the overall distribution of the relative error histogram of ESTARFM_NL and ES‐TARFM is tight and consistent.When the fusion results are evaluated by the average relative error and correlation coefficient,the difference between the two algorithms is considerable,indicating that the fusion accuracy of the two algorithms is equivalent.Comparing the efficiency of the two algorithms,we found that the ES‐TARFM_NL running time can be greatly reduced.Therefore,ESTARFM_NL provides an alternative fusion scheme for large-area or long-term sequence remote sensing data with large data volume.