Research On Fusion Model Of Multispectral And Panchromatic Remote Sensing Images Under Hybrid Multi-scale Framework

Multispectral and panchromatic remote sensing images are important components of multi-source detection systems in high-performance aerospace integrated monitoring,environmental monitoring,and precision agriculture.By fusing the rich spectral information of multispectral image with the fine spatial information of panchromatic image,a fused image that simultaneously considers spectral and spatial resolution can be obtained,which can achieve more comprehensive description for land and effectively improve the interpretation accuracy of remote sensing images.Currently,the hybrid algorithm that combine color space transform with multiresolution analysis method is a mainstream algorithm for multispectral and panchromatic remote sensing image fusion.It can leverage the complementary advantages of both methods,effectively balance the spectral fidelity and spatial information distortion of fused images.However,there are two major issues with such the hybrid algorithms:Firstly,most hybrid algorithms rely on a priori knowledge to determine the two methods and simply combine them,without deeply analyzing the specific impact of different color space transform methods and multiresolution analysis methods on fusion performance.This results in a lack of specificity in the hybrid algorithms,which in turn can leading to biases in fusion performance and decrease in the quality or efficiency of the fusion process.Secondly,for different fusion requirements such as land classification or large-scale remote sensing image fusion,how to design a hybrid framework and adjust and optimize the fusion model to make the fusion purpose more targeted,has become a key issue in multispectral and panchromatic remote sensing image fusion.To address the above issues,this dissertation constructs a hybrid multiscale framework for guiding fusion and separately selects methods with strong specificity for spectral and spatial information in each module of the framework.Based on the requirements for high-quality and high-efficiency fusion,a corresponding hybrid multiscale fusion model is constructed.The main research contents of this dissertation include the following aspects:(1)Analysis of multispectral and panchromatic remote sensing image fusion requirements and determination of quality evaluation methods:Combining the imaging principles of multispectral and panchromatic remote sensing images,the characteristics of these two types of images are analyzed.Based on the complementary information analysis of multispectral and panchromatic remote sensing images,the fusion objectives and different fusion requirements for different tasks in practical applications are analyzed.The subjective and objective performance evaluation of fusion image and the impact of fusion efficiency on remote sensing image fusion are analyzed.From the perspective of measuring the spectral fidelity and spatial information preservation of the fusion image,the fusion quality evaluation indicators based on spectral and spatial characteristics are analyzed,and representative indicators are selected to provide an effective basis for remote sensing image fusion evaluation.(2)Construction of a hybrid multiscale fusion framework and research on module performance:Taking the hybrid algorithms of color space transform and multiresolution analysis as the starting point,the three main modules of color space transform,multiscale transform,and detail extraction transform are determined in the hybrid multiscale framework.The specific roles of each module in the fusion process are clarified,and the performance of each module is analyzed.The experimental results show that the fusion image obtained by YUV color space transform shows the best fusion performance in spectral fidelity,and multi-scale transform with non-subsampling characteristics shows the best fusion performance in spatial information preservation.(3)Research on hybrid multiscale model for high-quality fusion:Under the hybrid multiscale framework,a fusion model using Do Gs,shearlet transform and multiscale morphological transform in the YUV color space is proposed.First,the multispectral image is transformed in the YUV color space.Then,a multiscale decomposition is performed using Do Gs transform with a scale-increasing Gaussian filter that has non-subsampling characteristics.Next,shearlet transform with multi-directional characteristics are used to extract edge details from the high-frequency image.Meanwhile,a dual-channel multiscale morphological top-hat and bottom-hat transform based on theàtrous algorithm is used to extract light and dark information from the low-frequency image.Finally,the improved Laplacian energy high-frequency fusion rule and window energy maximum low-frequency fusion rule are used to fuse each decomposition layer,and the final fused image is obtained through inverse transform.By selecting a variety of typical and advanced experimental comparison objects and using both simulated and real data for fusion experiments,the results show that the hybrid multiscale fusion model demonstrates excellent fusion performance in terms of both subjective visual effects and objective evaluation performance in terms of spectral fidelity and spatial information preservation.In particular,the ERGAS and Q_E objective evaluation indicators prove that the proposed algorithm greatly reduces the spectral and spatial information distortion of the fused image.(4)Research on hybrid multiscale model for high-efficiency fusion:Under the hybrid multiscale framework,a fusion model using support value transform,multi-directional morphological edges,and multiscale morphological transform in the YUV color space is proposed.First,a support value transform based on least squares support vector machines is used for multiscale decomposition to obtain high-frequency support degree images with significant features and low-frequency smooth background images.Then,an efficient multi-directional morphological edge transform is constructed to extract edge details in different directions from the high-frequency support degree images.Meanwhile,the dual-channel multiscale morphological top-hat and bottom-hat transform is still used to extract brightness and darkness details from the low-frequency background images.Finally,the high-frequency fusion rule that selects the maximum gradient within the window and the low-frequency fusion rule of selecting lighter in the light-information and selecting darker in the dark-information are used for fusion,and the final fused result is obtained through inverse transform.Through comprehensive fusion comparison experiments on simulated and real data,the proposed hybrid multiscale fusion model is proven to greatly improve fusion efficiency without significantly reducing fusion quality.Especially the size of the fused image is 1024×1024,the improvement of fusion efficiency is very significant.