Research On Several Key Technologies Of Satellite Cloud Image Super-resolution

Satellite cloud images can show the characteristics and evolution of cloud systems from various aspects.It is an important tool to monitor weather changes and make meteorological d isaster forecasts.At present,meteorological satellites mainly observe the atmosphere to obtain satellite images of different bands through the infrared,visible light sensort of different channels.Due to the limitation of the imaging principle and the interference from internal and external factors in the satellite data reception process,the spatial resolution of the satellite cloud image can not meet the needs of meteorological monitoring,and the resolution of different channel satellite images is also different.In particular,the resolution of infrared cloud image is usually low,and the gray level is not clear,making it difficult to synthesize the information of multiple channel images.How to reconstruct the high-resolution infrared cloud image becomes an urgent need in the application of satellite cloud,based on the existing observation equipment and observation data.This paper has carried on the key technology research of satellite cloud image super-resolution,based on the characteristics of the infrared cloud image and the requirements of the meteorological service for the cloud image processing.The main work is as follows:(1)Researching on the degradation process of image,and the image degradation model,introducing the sparse representation theory to solve the super-resolution problem.This paper focuses on the theory of sparse representation and the optimization algorithm of sparse representation problem,including greedy method and relaxation method.This paper introduces the construction principle and method of learning dictionary,and studies the mainstream super-resolution algorithm based on dictionary learning process.(2)A super-resolution method of the infrared cloud image based on structural group sparse representation is proposed.This method is different from the traditional sparse representation model which using a single image block as the basic unit of sparse reconstruction,but constructs the image block structural group as the basic unit of sparse representation,establishes the sparse representation model of the structural group,and makes full use of the similarity information.In the dictionary learning phase,the Gaussian mixture model is used to cluster the sample blocks,and the principal component analysis method is used to study each class,and a compact classification dictionary is obtained.During the reconstruction process,the most suitable dictionary is selected for each structural group,and the sparse coefficient is solved by the iterative contraction algorithm.Finally,the effect of image reconstruction is verified by simulation experiment.(3)A super-resolution method of the infrared cloud image based on V1LT-decomposition is proposed.Firstly,considering the different characteristics of the morphological components in the infrared cloud image,the low resolution infrared cloud image is decomposed into cartoon parts and texture parts by using the separation model.Using targeted method processes the different morphological components.The SAI algorithm is used to interpolate the cartoon part,and the texture part is enhanced by the image enhancement method of NSCT transform.Finally,the processed cartoon part and texture part are combined to get the high resolution cloud image.The simulation results show that the proposed algorithm can improve the visual effect and image quality,and effectively maintain the texture information of the cloud image.