Research On Spatio-temporal Fusion Algorithm Of Remote Sensing Image Based On Sparse Representation

The sensor carried by the remote sensing satellite is constrained by its own performance,so the single remote sensing image data collected cannot have both high spatial resolution and high temporal resolution.Remote sensing image space-time fusion technology is one of the most important methods to solve this problem.This technology combines the different advantages of several satellite sensors to obtain high temporal and spatial resolution image data.In terms of fusion data,due to the large time interval,the local area of the predicted time image is changed relative to the prior time image,resulting in the failure of the priori based on the prior time image to the high-low resolution images constructed in the changing area.In terms of sparse representation method,the method is applied to spatio-temporal fusion,and the reconstructed fused image needs to meet the assumption of linear consistency between the sparse coefficients of high and low resolutions.However,due to the large-scale difference between high-low resolution images,this hypothesis is difficult to be established.In this paper,a new spatio-temporal fusion method based on sparse representation is proposed to improve the accuracy of image fusion by focusing on the construction of the dictionary of varying regions and the transformation relationship of the sparse coefficients between high-low resolutions.The contribution of this paper is mainly in the following two aspects.（1）Aiming at the problem that it is difficult to accurately represent the changing region,a spatio-temporal fusion method based on cross-temporal dictionary learning is proposed to ensure the representation ability of the dictionary to the changing region.The existing spatio-temporal fusion algorithm uses a priori time image to establish a high-resolution dictionary,so the fusion accuracy is low in the area where the temporal image changes the land cover type.In this method,the similarity between MODIS time-series images is used to construct a high-ow resolution dictionary pair priori of cross-time-series to improve the representation ability of the dictionary in the changing region.In addition,the sensor bias mapping module is proposed to eliminate the influence of sensor bias on the fusion results.（2）Due to the large-scale gap between high-resolution and low-resolution images,the assumption that the sparse coefficients of high-resolution and low-resolution images have a linear relationship is not valid.Therefore,a space-time fusion method based on sparse coefficients of support vector regression is proposed in this paper.First of all,it is proved by experiments that the linear relationship between the assumed high and low sparse coefficients in dictionary learning is not tenable.Moreover,under the assumption of online relationship,the sparsity of the high-resolution sparse coefficients decreases during the conversion process,thus losing the sparsity,which seriously violates the idea of sparse representation and affect the results of sparse reconstruction.To solve these problems,the method proposed to establish the nonlinear relationship between the sparse coefficients of high-low resolution based on support vector regression,so as to improve the precision of the estimation of the sparse coefficients of high resolution,and thus improve the quality of the fusion image.