Study Of Fusion Models And Algorithms For Multispectral Images

As the development of satellite remote sensing technology,the remote sensing images with higher resolution can be captured.And then,the research in satellite remote sensing image processing bacame more and more popular.Fusion of the multispectral images is one of the most important area in the research field of remote sensing image processing.This fusion technique is so promising,and could be applied to various fields,including medical images,geological exploration,municipal administration,prediction of natural disasters and protection of the forest and the ocean,et al.Considering the background and the development of the fusion technique with multispectral images,this thesis is described in following four parts:The background of fusion of multispectral images and its corresponding research status will be introduced in the first chapter.As similar as the development of computer science,the fusion technique was born with this special age.With the help of the compute power of the modern computer,the data of multispectral images could be processed efficiently,and then they would be applied to many different areas.The germ of fusion in multispectral images could be deried from the end of last century,and became well-known in the early time of this century.During the last decades,a lot of great work have been exhibited to us.In the second part,the concept of the fusion of multispectral images will be introduced first,and then the main kinds of the methods,which extrate the feature of the image data to construct the new images with high-resolution both in spatial and spectral spaces,will be illustrated.At the end of this section,some measurement methods with corresponding quality-index will be introduced.In the third chapter,we proposed a novel method based on variational model.After introducing this variational model we proposed,the corresponding optimal algorithm will be give next.Apart from these,in the section of algorithm,we also give some brief introduce to some technique,including soft-thresholding and fast fourier fransformation techniques.The FISTA and ADMM frameworks will be used in the designed algorithm.In the part of experienments,the comparation of some state-of-art methods will be given with a brief analysis.The last one is a part of conclusion,giving the construction of this thesis and summarising the main ideas in our work,including both advantages and disadvantages.In the last,a research proposal of our next work will be given.