Research On Remote Sensing Image Fusion Algorithm Based On Component Replacement And Multi-resolution Analysis

Remote sensing image fusion is an effective tool for producing synthetic images with high spatial and spectral resolution and rich surface information.Multispectral(MS),panchromatic(PAN),and synthetic aperture radar(SAR)images collected by the different types of sensors convey different information about the area under earth observation,which can be used for different applications.Remote sensing image fusion improves the accuracy in decision making of remote sensing applications by combining different information into one image through relevant algorithms.Therefore,it has been a popular direction for academic research.Nowadays,remote sensing data shows the development trend of more wavebands and higher resolutions,and remote sensing applications are more diversified,which puts forward higher requirements on remote sensing image fusion algorithms.However,the performance of traditional fusion algorithms is unsatisfactory.To improve this situation,this paper takes advantage of the hybrid method that can absorb the advantages of multiple algorithms,and proposes three improved algorithms under the framework of component replacement method combined with multi-resolution analysis method for different types of remote sensing data.(1)The improved algorithm of multi-scale guided filtering combined with fractional order differentiation is proposed to address the shortcomings of the general method of component replacement that cannot effectively extract and enhance spatial information.The method extracts the high-frequency components of MS and PAN images with the bootstrap filter,estimates the approximate components of the high-frequency components of PAN images with the adaptive IHS method(AIHS),and constructs a multi-resolution analysis tool with the bootstrap filter to extract the spatial information in a multi-scale manner.Finally,the spatial information is enhanced with a fractional-order differential mask and injected into the MS images to obtain the fusion results.(2)To address the shortcomings of traditional methods for fusing new remote sensing images such as World View-2/3,the improved algorithm of HCS combined with à trous wavelet is proposed by taking advantage of the multi-band processing capability of hyperspherical color space transform(HCS)and the outstanding performance of à trous wavelet transform in image fusion.The method uses modulation transfer function(MTF)filter and normalized difference vegetation index(NDVI)to reasonably spatially enhance PAN images,to mitigate the possible spatial degradation caused by à trous wavelets.(3)In order to address the shortcomings of the traditional methods for MS and SAR image fusion,we propose an improved algorithm of HCS combined with the nonsubsampling Shearlet transform(NSST)that can capture the detailed features of SAR images at multiple levels,and the high-frequency coefficients are traded off using parametric adaptive pulse coupled neural network(PA-PCNN).Comparative experiments of multiple algorithms with multiple sets of data were conducted for each of the above three improved algorithms.The results of subjective evaluation and objective evaluation with ERGAS,RMSE,UIQI,CC,IE,and STD metrics show that all three algorithms can obtain better fusion results in their respective fusion experiments and are feasible.