基于时空数据融合模型的TM影像云去除方法研究

针对已提出的各类云去除方法在实际应用中存在的局限性,将时空数据融合模型引入到云去除方法中。首先基于MODIS数据提供的时间维变化信息和辅助时相TM数据提供的空间信息,应用增强时空适应反射率融合模型(ESTARFM)得到了目标时相似TM合成数据;然后用TM合成数据替换掉目标时相TM影像中被云及其阴影覆盖区域的数据。在修复后的影像中替换区域与非云区域色调基本一致。通过非云区TM合成数据间接对替换云及其阴影区数据的精度进行定量评价。结果表明:相对于真实TM影像,非云区域合成数据各波段均值差异都在1%以内;各波段的相对误差分别为16.29%、12.92%、13.47%、12.87%、9.71%和11.84%,且各波段的相关系数均大于0.7;非云及其阴影区融合影像数据间接表明填补云及阴影区数据各波段的总体精度优于83%。因此,所提出的方法能够修复TM影像中被云及其阴影覆盖区域的数据,提高MODIS与TM数据的利用率。

Research on Cloud Removal from Landsat TM Image based on Spatial and Temporal Data Fusion Model

To solve the limitation of the existing models for cloud removal in practical application,in this paper,a new method was proposed based on spatial and temporal data fusion models.First,the data,like TM image at target time was composed by enhanced spatial and temporal adaptive reflectance fusion model (ESTARFM) based on temporal change of MODIS data and spatial information of auxiliary TM data;Then,the pixels in target TM image where were contaminated by clouds and shades which were replaced by the compose data.The result show that the color of the replaced area is consistent with the color of area uncontaminated by clouds and shade.Ultimately,the precision of the replaced data is verified indirectly based on the data of target TM image and composed image without cloud and its shade cover.Compared to actual image,the result showed that the relative difference of individual band of composed data is less than 1%;The mean relative error of each band are 16.29%,12.92%,13.47%,12.87%,9.71%,11.84%,respectively;All correlation coefficients are greater than 0.7;The accuracy of non|cloud and non|shade area fusion data indicates indirectly that the accuracy of each band of the data to fill the area,contaminated by cloud and shade,is better than 83%.Therefore,the method proposed in this paper which can repair the data contaminated by clouds and shades from TM image and improve MODIS and TM data utilization level.