Multispectral And Panchromatic Image Fusion Method Based On Spatial Information Enhancement

With the development of image processing technology,remote sensing images have been widely used in many fields.However,due to the limitations of hardware and other factors,a single sensor cannot obtain more detailed ground information,such as remote sensing images with both high spatial resolution and spectral resolution.In order to overcome the inherent shortcomings of a single remote sensing image,image fusion technology was introduced into the field of remote sensing image processing.At present,there are more and more diversified methods of remote sensing image fusion,each method has its own characteristics,and at the same time,good fusion results have been achieved.The fusion methods based on component replacement have high fusion efficiency and simple operation,but there is spectral distortion in the fusion result.The method based on multi-resolution analysis can better retain the spectral information,but the high and low frequency decomposition in the fusion process makes the fusion result lose some spatial details and texture information.The method based on the variational model can get a good fusion effect,but the process of solving the energy function will take a lot of time,making the fusion efficiency very low.The fusion method based on deep learning has great advantages in processing large amounts of data,but it is difficult to be widely applied due to the lack of reference images in remote sensing images.Aiming at the problems of current fusion methods,the main research contents of thesis are as follows:1.This thesis propose a fusion method combining the variational method and the method based on component substitution.In view of the high time complexity of the existing variational method,combining it with the method based on component substitution has complementary advantages.The traditional IHS method based on component substitution directly replaces the intensity components of a multispectral image with a panchromatic image,which will cause spectral distortion.This paper proposes to use the variation method to calculate a more accurate new intensity component from the original intensity component and the panchromatic image,used to replace the original image intensity component.Experimental results show that this method not only avoids the spectral distortion of the IHS method,but also improves the fusion efficiency of the variational method.2.A fusion method based on super resolution to reconstruct network structure is proposed.Different from most deep learning-based methods,this paper uses deep superresolution reconstruction network VDSR to learn the spatial structure information of the panchromatic image and the original intensity component to reconstruct a new intensity component.In order to avoid information redundancy,the new intensity component is subtracted from the original intensity component to get the high-frequency intensity component,and then inject a certain model gain into the high-frequency intensity component and add it to the multispectral image,so as to obtain a final fused image.The experimental results show that the fusion results can well protect the multi-spectral spectral information,and also improve its spatial structure information.