Research On Remote Sensing Image Fusion Based On Spatial-spectral Information Preservation

The management of geographic information relies on remote sensing data generated by satellite sensors,thereby to monitor the utilization of natural resources in multiple dimensions.In order to accurately reflect the difference in the utilization of natural resources,the high-spatial resolution multispectral(HRMS)images are required in the management of geographic information.Due to the limitations of existing satellite sensing technology,remote sensing satellites cannot directly collect HRMS images.They usually carry two types of sensors,which are used to generate low-spatial-resolution multispectral(MS)images and high-spatial-resolution panchromatic(PAN)image.The two kinds of the images have complementary features.With remote sensing image fusion technology,the comprehensive,accurate and reliable HRMS images can be obtained and the hardware constraints can be settled.High-quality remote sensing fusion images should retain the spatial information on PAN images,while keeping the spectral information on MS images.In recent years,remote sensing image fusion technology has achieved rapid development,but there are still some challenges.For instance,there are problems of inaccurate detail injection leading to spatial distortion of the fused image,ignoring the structure information on MS image that is different from the PAN image and resulting in spatial and spectral distortion,difficulty in balancing the spectrum and spatial information,and in balancing the effectiveness and efficiency of the algorithm,causing spectral distortion of the fusion image when the PAN image is lowly correlated with the MS image.This paper analyzes the key factors affecting the fusion quality and researches the corresponding key technologies,thereby to achieve the following goals:(1)constructing an accurate fusion model and optimizing the parameters of the model,(2)obtaining details highly relevant with the MS image and reducing spectral distortion,(3)retaining the dual fidelity of spatial and spectral information,(4)achieving high efficiency of the algorithm.The main innovations of the paper are as follows.(1)Aiming at the problem that the multi-scale analysis is prone to produce spatial distortion,the paper proposes an adaptive injection model.First,a multi-scale Gaussian filter is defined by simulating the characteristics of the MS sensor,and the PAN image is convolved with the filter to extract details,so as to obtain the details that are highly related to the MS image.Then,the spectral information and the detailed information are comprehensively considered,and an adaptive injection coefficient is designed to achieve accurate detail injection.In addition,a new edge preserving weight matrix is proposed to better maintain the edge information on the fused image and achieve the dual fidelity to spatial and spectral information.Finally,the optimized injection coefficients are multiplied by the details and injected into the up-sampled MS image to obtain the final fusion result.The performance of the proposed method has been analyzed,and a large number of experiments have been carried out on multiple satellite databases.Compared with some advanced remote sensing image fusion methods,the results show that the proposed method can perform better in both subjective and objective evaluations.(2)In order to maintain high spatial resolution while reducing spectral distortion and improving algorithm efficiency,the paper proposes a fusion model based on fuzzy logic and saliency measure.In this model,a fuzzy rule based on global saliency measures is designed to fuse the detail information on the MS the PAN images,so as to obtain the detail information highly related to the MS image.In addition,in order to better maintain the edges of the fused image,this paper fuses the edges of the PAN and MS images according to the local saliency measurement.The fusion details and fusion edges obtained above are applied to the MS image together to obtain the final HRMS image.A series of experimental analysis on remote sensing images of 4 satellite datasets have verified the effectiveness of the method.Compared with some of the latest remote sensing image fusion methods,the proposed method shows good performance in both subjective and objective evaluations.(3)Aiming at the problem that traditional fusion methods cannot balance spectral and spatial information well,the paper proposes a remote sensing image fusion model based on conditional random fields(CRFs).By designing the state feature function of CRFs,the model keeps the predicted HRMS image consistent with the up-sampled MS image after filtering,so as to maintain the spectral information.In order to obtain a suitable blur function,in the state feature function,a new filter acquisition method is designed to construct an accurate degradation model.In addition,by defining the transfer feature function of CRFs,the transition of HRMS pixels can follow the gradient of the PAN image to ensure the spatial clarity of the fused image.Finally,the alternating direction multiplier method is used to solve the augmented Lagrangian function of the model to obtain the final fusion result.The paper analyzes the performance of the proposed method through a large number of simulation and real experiments.Compared with the existing fusion methods,the proposed method can achieve better fusion results with high computational efficiency(4)Aiming at the problems of spectral distortion caused by low correlation between PAN and MS images,and spatial distortion caused by inaccurate extraction of the required details of each channel of MS image,a fusion method based on joint guided detail extraction is proposed.First,a new PAN image is constructed through a variational model guided by the intensity component of the MS image,which improves the correlation between the PAN image and the MS image and reduces spectral distortion.Secondly,in order to obtain accurate detail information on the new PAN image,the adaptive coefficients are obtained through a regression model guided by the reduced-scale MS image to extract the details of the PAN image.Finally,the extracted details are injected into the up-sampled MS image to obtain a fused image.A large number of simulation and real experiments were performed,and the experimental results were compared with the existing fusion methods.The results show that the proposed method can efficiently obtain the HRMS image of high spatial and spectral fidelity.The proposed remote sensing image fusion methods effectively improve the spatial and spectral quality of HRMS images,and the computational cost is relatively low.They can provide real-time and accurate data support for monitoring tasks such as land use and resource census in geographic information management.