融合HJ-1 CCD和MODIS生成高分辨率影像方法对比

高时空分辨率遥感影像的反演可有效解决南方云雨地区的数据缺失问题。本文以广西典型丘陵山地为试验区，利用时空自适应反射率融合模型（Spatial and Temporal Adaptive Reflectance Fusion Model，STRAFM）和增强型时空自适应反射率融合模型（Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model，ESTRAFM）两种融合算法，选取小范围的国产环境减灾卫星（HJ-1 CCD）和中分辨率成像光谱仪（Moderate-resolution Imaging Spectroradiometer，MODIS）数据，比较分析两种融合算法所生成的高空间分辨率影像的优劣。与真实HJ-1 CCD的红、近红外（Near-infrared，NIR）波段影像相比，STRAFM和ESTRAFM预测影像在空间分布上均具有较好的一致性，R值均为极显著相关，差分图像98.94%以上像元反射率差值小于0.1，平均绝对差值（Average Absolute Difference，AAD）、平均差值（Average Difference，AD）、标准差（Standard Deviation，SD）均较小，融合效果好。与STRAFM相比较，ESTRAFM融合算法对真实HJ-1CCD影像的细节捕捉能力更强，高低反射率区域没有明显缩小或放大现象，破碎地物边界更清晰，不存在斑块。ESTRAFM预测影像与真实HJ-1 CCD红、近红外波段影像的相关性均高于STRAFM，相关系数（Pearson Correlation Coefficient，R）分别为0.930， 0.885。ESTRAFM预测影像与真实HJ-1 CCD影像差异小于STRAFM，其差分影像的AD、AAD、SD分别为-0.005、0.013、0.017。

Comparison of different fusion methods for generating high spatial-temporal resolution data using HJ-1 CCD and MODIS data

Inversion of high temporal-spatial resolution remote sensing data can effectively solve the problem of missing data due to the cloudy weather in the southern China. In this paper, taking Guangxi typical hilly area as the research area, using Chinese HJ-1 CCD and MODIS(Moderate-resolution Imaging Spectroradiometer) data, two fusion models include STRAFM(Moderate-resolution Imaging Spectroradiometer) and ESTRAFM(Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model) for generating high spatial-temporal resolution data are compared. The results show that, compared to the actual HJ-1 CCD images, there is a good consistency in spatial distribution for prediction red and NIR band images by STRAFM and ESTRAFM. Correlation coefficients between real images and prediction images are high and significant. The reflectance values of the difference image are less than 0.1 for more than 99.94% pixels. The absolute difference(AD), average absolute difference(AAD) and standard deviation(SD) are all very small. In short, the fusion effect is good for both STRAFM and ESTRAFM. However, compared with STRAFM, ESTRAFM fusion algorithm can capture more details on real HJ-1 CCD image. There is no significant reduction or magnification for high or low reflectance regions in prediction images by ESTRAFM. Besides, boundary of broken object is more clearly. Moreover, no blocks exist on the ESTRAFM prediction images. For the ESTRAFM predicted red and NIR band images, R is 0.930 and 0.885 respectively, smaller than those of the STRAFM. The difference between the real images and ESTRAFM predicted images are also smaller than that of the STRAFM, with its AD, AAD and SD are -0.005,0.013 and 0.017 respectively.