The Variational Fusion Methods For Multisource Spatio-Temporal-Spectral Optical Remote Sensing Images

Remote sensing is one of the important ways to obtain the information of the surface features,and it has been playing an important role in a variety of relevant fileds.Specially,the spatial resolution,the spectral resolution,and the temporal resolution are important indictors for remote sensing applications.However,duo to the technical limitations on the design of the remote sensing hardwares and other factors,there is tradeoff between the spatial resolution,the spectral resolution,and the temporal resolution.In addition,the optical remote sensing images are inevitably influenced by clouds and the haze,etc.Fortunately,remote sensing image fusion is an effective way to integrate the complementary spatial,spectral and the temporal information of the multisource remote sensing images,to produce a knowledge of phenomena under investigation better than the knowledge achievable from individual data sets.Therefore,this paper has carried out the research of the multisource remote sensing image fusion based on the variational framework.The main research contents are as follows:(1)This paper proposed to compare of different fusion frameworks based on an idea of Meta analysis.The panchromatic(PAN)/multispectral(MS)fusion is one of the most fundamental fusion methods for remote sensing images,therefore,this paper firstly carried out the researches on PAN/MS fusion.To data,there have developed large amounts of PAN/MS fusion methods.On the whole,most of the existing methods can be classfied into four categories,i.e.,the component substituation(CS)based fusion methods,the multiresolution analysis(MRA)based fusion methods,the model-based optimization(MBO)based fusion methods,and the learning-based fusion(LBF)methods.However,most of the existing understanding on the performance of different frameworks mainly based on a small number of traditional methods or some special methods,therefore,they have inconsistent understandings on the performance of the above different categories of the PAN/MS fusion methods.Therefore,this paper novelly proposes to evaluate the performance of the above different categories of the PAN/MS fusion methods based on an idea of Meta analysis.Firstly,the researches on PAN/MS fusion between 2000?2016 are retrieved as many as possible.Then,the representative 63 researches which are selected from more than 1000 papers are refused to provide a more robust and convincing results.In addition,the development progress,the current status and the development trend for PAN/MS fusion are comprehensively introduced.(2)This paper proposed a variational PAN/MS fusion method based on moment matched gradient for remote sensing images.The traditional variational PAN/MS fusion methods construct the relationship model between the fused image and the PAN image,based on the linear combination of the spectral bands;however,they are limited by the spectral response function of the satellite sensors and the difference of spectral range of the bands.Therefore,this paper proposed a novel variational PAN/MS fusion method based on moment matched gradients.The relationship model between the fused image and the PAN image was constructed based on the gradient texture information,and the robustness of the relationship model was further enhanced based on moment matching.The proposed method extends the application ability of the traditional variational PAN/MS fusion methods for different kinds of satellite remote sensing datasets,and it can obtain the fused images with both high spatial and spectral resolutions.The proposed method has been verified by both the simulated and real experiments.(3)This paper proposed an integrated spatio-temporal-spectral fusion methods based on multisensor remote sensing images.The traditional fusion methods have been developed independently,and they are lack of the unified theory framework,therefore,most of the existing fusion methods can only be applied for the fusion of two of the three spatial,spectral,and the temporal resolutions.Therefore,the integrated spatio-temporal-spectral fusion methods are proposed and developed;however,most of the existing fusion methods perform poor for the fusion of more than two-sensor images.Therefore,this paper proposed an integrated spatio-temporal-spectral fusion methods based on multisensor remote sensing images.In the proposed method,the integrated fusion model was constructed by comprehensively consider the spatial,spectral,and the temporal relationships among multisource remote sensing images,in addition,several different fusion demands,including the multiview fusion,the spatio-spectral fusion,and the spatio-temporal fusion,are taken into account.The proposed method can not only effectly integrate the complementary spatial,spectral and temporal information of more than two-sensor images,it can also statisfy the demands for multiview fusion,the spatio-spectral fusion,and the spatio-temporal fusion.The proposed integrated fusion method has been comprehensively verified from different aspects based on IKONOS,QuickBird,Landsat ETM+,MODIS,HYDICE,and SPOT5 satellite images.(4)This paper proposed an fusion method for optical remote sensing images with cloud and fog contamination.Most of the image fusion methods only focus on the resolution degradation of the remote sensing images;however,they don't take into consideration of the possible cloud cover of the optical remote sensing images.Therefore,this paper proposed an fusion method for optical remote sensing images with cloud and fog contamination.The proposed integrated fusion model was constructed based on the multisource and multitemporal remote sensing images,and multiple degradation factors,including the thin clouds,the thick clouds,the cloud shadows,and the resolution degradation are comprehensivey considered.The proposed integrated fusion model mainly includes two steps:1)the integrated removal for thin cloud,haze,and the light cloud shadow;2)the integrated processing for thick clouds,the dark cloud shadows,and the resolution improvement.Finally,followed by the above two steps,the information reconstruction and the resolution enhancement can be simutaneously achieved.The proposed method has been verified by both the simulated and real experiments.