摘要:
时空融合能够生成具有足够时间和空间分辨率的图像序列。然而,当前的研究趋向于使用尽可能多的时相数据、复杂的非线性模型来提高预测的准确性,极少的研究将重点放在图像本身的分析,即充分利用影像包含的如趋势和纹理等内在特征。为此,文章基于二维奇异谱分析(2D singular spectrum analysis,2DSSA)技术,提出了一种2DSSA时空融合模型(2DSSA spatial-temporal fusion model,2DSSA-STFM),通过将已有影像分解为趋势分量和细节分量,分别对目标时刻影像的主要空间趋势和空间细节进行预测。首先,建立高空间分辨率数据趋势项与低空间分辨率数据的线性关系,计算得到目标时刻影像的趋势成分; 然后,建立2个时相下低分辨率细节分量和高分辨率细节分量的线性关系,得到目标时刻影像的细节成分; 最后,将计算得到的趋势和细节成分进行合成,即为目标预测影像。在2组中分辨率Landsat7 ETM+和MODIS影像上对提出的2DSSA-STFM进行了实验,结果表明,提出的模型在实验误差方面要优于传统的时空融合模型。
Abstract:
Spatio-temporal fusion can generate image sequences with sufficiently high temporal and spatial resolution. However, current studies tend to improve prediction accuracy using as much spatio-temporal data as possible and complex non-linear models, while few of them focus on analyzing images themselves by making full use of their intrinsic features, such as trends and textures. This study proposed a 2DSSA spatio-temporal fusion model (2DSSA-STFM) based on 2D singular spectrum analysis (2DSSA). In this model, the major spatial trends and details of the existing images at the target moment can be predicted by decomposing the images into trend and detail components. Firstly, the linear relationship between the trend of high-spatial-resolution data and low-spatial-resolution data was built to calculate the trend components of the images at the target moment. Then, the linear relationship between the low-resolution and the high-resolution detail components in two time phases was established to determine the detail components of the images at the target moment. Finally, the calculated trend and detail components were combined to form the target prediction images. The 2DSSA-STFM was applied to two sets of medium-resolution Landsat7 ETM+ and MODIS images, yielding smaller experimental errors than conventional spatio-temporal fusion models.
下载: