Research On Remote Sensing Image Fusion Based On Injection Model And Its Application

The resolution of remote sensing images directly affects the comprehensiveness and accuracy of land resources information.With the development of remote sensing technology,remote sensing images have been applied more and more widely in land resources,and land resources management has higher requirements on the resolution of remote sensing images.In practical applications,due to the technical limitations of satellite sensors,the commercial optical satellites such as IKONOS,Quick Bird,and World View-2,cannot directly collect high-resolution multispectral(HRMS)images.However,they provide panchromatic(PAN)and multispectral(MS)sensors simultaneously.The PAN sensor produces a high spatial resolution but low spectral resolution image,while the MS sensor produces a high spectral resolution but low spatial resolution image.To obtain a HRMS image,the best solution is to develop an efficient remote sensing images fusion technique,which can fuse the complementary information from the PAN and MS images.Remote sensing image fusion aims at preserving the original spectral characteristics of the MS image and adding spatial characteristics of the PAN image to the fused image.Therefore,in order to meet the growing needs of the land and resources information management,remote sensing image fusion technology has attracted much attention in the research of the information management technology of the land and resources and has become the current research trend.Normally,remote sensing image fusion algorithms can be divided into three classes in different fusion level: 1)pixel-level image fusion,2)feature-level image fusion and 3)decision-level image fusion.Among them,pixel-level remote sensing image fusion is the most commonly used fusion at present.It fuses the information of each pixel in the source images one by one,and it can retain as much important information as possible in the source image.Along with the development of the image fusion algorithms,a general remote sensing image fusion scheme based on injection model is formed in the field of pixel-level remote sensing image fusion.The scheme assumes that the lost spatial information of low-spatialresolution multispectral(LRMS)images can be compensated by the spatial information of high-spatial-resolution PAN images.The pansharpening of the MS images by the injection model can be completed by exploring the spatial information of PAN images to extracte its high frequency information and inject that into the LRMS images.The injection model mainly concludes three steps: 1)the extraction of the injection details,2)the process of the object for accepting the details;and 3)the process of the detail injection.Two problems consist in the exsited algorithm based on the injection model,such as 1)the problem of the low relevant between MS and PAN images.2)the problem of the over injection for details.This paper focuses on the key technologies of the injection model to carry out research.The study works mainly conclude three aspects: 1)the improvement of the injection details,2)the improvement of the object for accepting the details;and 3)the improvement of the injection gains.The main research results are summarized as follows:(1)A remote sensing image fusion algorithm based on refining detail injection is proposed,and its application in national resource management is described.Remote sensing image fusion has a potential spectral distortion problem due to the global/local spectral and spatial correlations between PAN and MS images.To overcome this problem,A remote sensing image fusion algorithm based on refining details injection is presented.In the proposed method,the spatial details are first extracted from the MS and PAN images through à trous wavelet transform and multiscale guided filter.Different from the traditional detail injection scheme,the extracted details are then sparsely represented to produce the primary joint details by dictionary learning from the sub-images themselves.To obtain the refined joint details information,we subsequently design an adaptive weight factor considering the correlation and difference between the previous joint details and PAN image details.Finally,the refined joint details are injected into the MS image using modulation coefficient to achieve the fused image.The proposed method has been tested on Quick Bird,IKONOS and World View-2 datasets and compared to several state-of-theart fusion methods in both subjective and objective evaluations.The experimental results indicate that the proposed method is effective and robust to images from various satellites sensors.The algorithm can effectively eliminate the information acquisition error,and its fusion image can meet the needs of the management for land and resources.(2)A remote sensing image fusion algorithm based on compensation detail injection is proposed,and its application in national resource management is described.Remote sensing image fusion based on the detail injection scheme consists of two steps: spatial details extraction and injection.The quality of the extracted spatial details plays an important role in the success of a detail injection scheme.In this paper,a compensationdetail based injection fusion model is presented from a new perspective of compensatory learning.In contrast to the traditional method,the two categories of detail,namely,the PAN details and the compensation details,are considered to compensate for the spatial and spectral differences between LRMS and HRMS images.To obtain the compensation details,robust sparse representation was employed to calculate the difference between the PAN and MS images during the fusion.The compensation details combined with the PAN details extracted by a multiscale-guided filter are then injected into the upsampled LRMS image to achieve a fused image.Extensive experiments were undertaken on several image datasets,and the results demonstrate the effectiveness of the proposed CDI method.The algorithm can solve the problem of low classification accuracy in the management for land and resources.(3)An improved injection model fusion algorithm based on multispectral images is proposed,and its application in national resource management is described.The pansharpening process builds a high-resolution(HR)multispectral image by injecting the details extracted from the PAN image into the LRMS image.This process may result in nonignorable spectral distortion in the fused image because of the incongruent geometric structure and potential mismatch between the extracted details and LRMS image.To solve this problem,we propose a learning-based injection algorithm to improve the quality of the LRMS image by learning the spatial information of the degraded PAN(DPAN)image.In the proposed method,LR/HR dictionary pairs were firstly constructed from the DPAN image and the detail subimage of the PAN image,respectively.The sparse coefficient of the LRMS image was produced by sparse representation(SR)with the atoms of the obtained LR dictionary.Furthermore,the difference information between the intensity component of LRMS image and DPAN image can be restored by sharing the obtained sparse coefficient with the HR dictionary.The difference information was then injected into the LRMS image to obtain the improved LRMS(ILRMS)image.As a result,the extracted details from the PAN image are highly related to the ILRMS image.Finally,the fused image can be realized through injecting the details of PAN image into the ILRMS image.The fusion results from IKONOS and World View-2 datasets show that the proposed method is effective,robust and even superior to the other state-of-the-art fusion methods in terms of both full-reference and no-reference metrics as well as a visual inspection.The algorithm can provide comprehensive and accurate information to the land and resources management department.(4)A remote sensing image fusion algorithm based on spectral-intensity modulation is proposed,and its application in national resource management is described.The pansharpening algorithm often faces imbalance between spatial sharpness and spectral preservation,resulting in spectral and intensity inhomogeneities in the fused image.In this paper,to overcome this problem,we present a robust pansharpening method for multi-band images with adaptive spectral-intensity modulation.In the method,we propose an adaptive spectral modulation coefficient(ASMC)and an adaptive intensity modulation coefficient(AIMC)to modulate the spectral and spatial information in the fused image,respectively.Among these coefficients,ASMC is constructed based on two aspects: 1)the details extracted from the PAN and MS images;and 2)the spectral relationship between each MS band.AIMC is calculated by assessing the correlation and standard deviation between the PAN image and each MS band.Finally,we propose a mathematically linear model to combine ASMC and AIMC to achieve the fused image.Various remote sensing satellite images were used in the evaluations.Experimental results indicate that the proposed method achieves outstanding performance in balancing spatial and spectral information and outperforms several state-of-the-art fusion methods in terms of both full-reference and no-reference metrics,and on visual inspection.The problem of incomplete and inaccurate data information and low automatic extraction level of HRMS image in national resource management can be solved by the algorithm.