Research On Super-resolution Reconstruction Algorithm Of Remote Sensing Image Based On Dictionary Learning

The emergence of image super-resolution reconstruction technology has provided a new way to improve the spatial resolution of remote sensing image,which has become research focuses in remote sensing image processing.By means of the knowledge of machine learning and neural network,super-resolution reconstruction technique based on dictionary learning makes full use of prior information to reconstruct the high resolution image,which is more suitable for super-resolution reconstruction of large magnification.In this paper,the super-resolution reconstruction algorithm of single optical remote sensing image is taken as the research direction,and based on the theory of sparse representation and different dictionary updating methods.The super-resolution reconstruction algorithm based on dictionary learning is expounded and studied thoroughly.The main research contents and conclusions include:1)This paper describes the degradation model of remote sensing image and the sparse representation theory,and introduces the sparse representation of remote sensing image super-resolution reconstruction.This paper analyzes the construction of over-complete dictionary in super-resolution reconstruction model,and briefly introduces some algorithms of sparse matrix and dictionary updating.2)The natural image super-resolution reconstruction model based on the joint dictionary is applied to remote sensing image reconstruction.In this paper,three classic dictionary updating algorithms are analyzed,including Method of Optional Directions,K-SVD and Online Dictionary Learning.In the dictionary learning stage of the joint dictionary,the above dictionary update algorithms are adopted respectively.Experimental results has shown that for a large amount of information of the remote sensing image data,and higher magnification,super-resolution reconstruction algorithms based on online dictionary learning performs higher efficiency in the dictionary learning.3)The super-resolution reconstruction algorithm based on online dictionary learning is further studied.Aiming at the universality of remote sensing image reconstruction based on the method,this paper proposes a super-resolution algorithm based on Optimization of Online Dictionary Learning,and through the analysis of the key parameters of the algorithm,selects the optimal values.ZY-3 and the latest GF-4 image data are taken as the test image.Results have shown that the Optimization of Online Dictionary Learning algorithm can effectively improve the quality of the reconstructed image for different test images of different size.4)This paper proposes based on coupled dictionary learning algorithm of super-resolution reconstruction.In order to solve the problem of large amount of information in remote sensing images,the IDL language is used to realize the automatic selection of the training samples,which improves the efficiency of data preprocessing.Compared with the super-resolution reconstruction based on sparse representation algorithm(ScSR),better results have proved that algorithm proposed has higher spatial resolution capability.Through the above theoretical research,the improvement of related algorithms and the experimental analysis,the achievement of this paper have certain guiding significance for the further study of super-resolution reconstruction of remote sensing images based on learning.