Image Fusion Guided By Prior Knowledge In Multispectral Remote Sensing

Multispectral?MS?image is one of the most common data in remote sensing data and plays a critical role in vegetation monitoring and other fields.Due to the limitation of technology and efficiency,MS cameras are generally equipped with MS sensors and panchromatic?PAN?sensors with both spectral resolution and spatial resolution advantages.In many application fields,data combination of multiple sensors provides more comprehensive information.Therefore,the research on MS remote sensing image fusion technology can make full use of the information,which has great practical value.At present,many researchers worldwide have done a lot of work to explain,solve the problem of image fusion,and put some methods into practice,but these methods do not fully utilize the characteristics of MS images,the scope of application is not wide.Based on the previous research work and the characteristics of MS images,a multispectral image fusion model guided by prior knowledge is proposed to improve the performance of MS images in fusion.The specific work is as follows:?1?The background of MS remote sensing images is introduced in chapter 1,then the research progress of MS image fusion methods is reviewed in chapter 2.In chapter3,a MS image fusion method employing adaptive spectral-spatial gradient sparse regularization is proposed.Integrating the prior knowledge of spectral consistency,gradient sparsity and gradient information transfer,this method use ₁l norm to achieve spatial structure consistency,and designs an adaptive weight matrix to adaptively select different weights for different regions.This choice helps to keep the spatial information similarity between the MS image and PAN image.In order to ensure convergence,this method uses alternating direction multiplier method to obtain fusion image iteratively.Experiments on Parrot Sequoia and other datasets have proved that this method can effectively reduce the spectral distortion and spatial information blurring,reach the state-of-the-art in many fusion method.?2?Normalized Difference Vegetation Index?NDVI?images are generated by MS image,which can precisely mirror the status of surface vegetation cover.Motivated by the features of NDVI and dictionary learning,a fusion method for NDVI images is proposed in chapter 4.This method has the following characteristics:1.The index similar to the NDVI form is constructed by using PAN image and MS image.Due to the addition of PAN image,the corresponding image of the index contains more spatial texture information,which can be transferred to NDVI image to improve the spatial resolution of NDVI image.2.The prior knowledge of sparse representation and scale invariance is used to construct the mapping relationship of high-low NDVI image block dictionaries,so that the process of spatial information injection can be completed by constructing an online over-complete dictionary without the help of other image contents.Experiments on GF-2 dataset show that the NDVI image obtained by this method after spatial information injection has more details than the original NDVI image and the NDVI image obtained from the fused MS image.?3?After obtaining the fusion image,how to use and take full advantage of it is also worth discussing.Compared with the original MS image,the fusion image possesses more spatial textures,that is,clearer details.Based on the two fusion methods proposed in chapter 3 and chapter 4,chapter 5 discusses the feasibility of fusion image in practical application by using multiple datasets.In the experiment of NDVI image frequency distribution,the NDVI image acquired by the method proposed in chapter 3not only improves the spatial resolution of the image,but also is close to the real NDVI distribution.In the experiment of temporal change detection of multi-source NDVI images,the fusion method proposed in chapter 4 is used to obtain NDVI images,which proves that using different types of satellite MS data in the same region can effectively improve the temporal resolution of the changed images.In addition,the spatial information injected into NDVI images can improve the spatial resolution.In the process of change detection,the fused NDVI images have more edges and details,and the detection accuracy is also higher.