Research On The Key Technologies Of Remote Sensing Image Processing Based On Sparse Representation And Dictionary Learning

With the rapid development of information technology and earth observation technology,remote sensing technology plays a more and more important role in the social life and economic construction. At present, remote sensing technology has been widely used in environmental protection, geological prospecting, land and resources survey, urban planning and monitoring, forestry and agricultural census, military interpretation, etc. The development of artificial intelligence and high resolution remote sensing technology, especially the growth of spatial resolution and spectral resolution, lead the growth of data by orders of magnitude. This has made a higher request to the data transmission and processing. Remote sensing image processing technology has changed significantly in theory, technology and application, and the traditional remote sensing image processing methods have been difficult to meet the needs of the current remote sensing application.With the rapid development of remote sensing technology, the research of theory and method in the field of machine vision is also in rapid progress. In recent years, the sparse representation and compressed sensing theory have been used in image processing, network engineering, medical and military fields and have been a huge success. In this paper, according to the characteristics of the remote sensing image and its application background, the traditional remote sensing image processing technology combined with the theory of computer vision, and the sparse representation and dictionary learning theory and typical applications were studied. In the sparse representation and dictionary learning, its basic principle is to use overcomplete dictionary of redundant basis replace the orthogonal basis, and the selection of dictionary should contain the decomposition signal information as much as possible. Sparse representation can decrease the high price needed to build sensors and reduce the cost of transfer between the sensor and ground receivers, and the computing cost of the sensor is transferred to the ground. The proposed approach can tap the potential applications of remote sensing images, and provid the reference for all kinds of application.The main achievements and innovations are as follows:1 Through studying the related theory of sparse representation and dictionary learning and the achievement of application, the principle of remote sensing image processing based on sparse representation and dictionary learning is analyzed. Through nullspace property, restricted isometry property and boundary constraint, how to construct sensing matrix is analyzed. Through the example, the guarantee of sparse recovery is illustrated.2 The principle of signal sparse representation is reviewed through the mimimum of1? and sparse recovery algorithm, and focuses on the principle and method of dictionary learning algorithm. The dictionary learning model and algorithm of multispectral remote sensing image and hyperspectral remote sensing image is constructed according to the structure characteristics of multispectral and hyperspectral remote sensing images.3 Combined with the research achievements in the field of computer vision, three new algorithms of remote sensing image denoising are presented. Respectively are remote sensing image denoising based on generalized gaussian distribution and locally adaptive, remote sensing image denoising based on sparse representation and dictionary learning,and the multispectral remote sensing image denoising based on clustery and grouped sparse representation and dictionary learning. Simulation results show that the better denoising effect can be obtained as compared to other recent image denoising methods.4 Based on some previous work,two new algorithms of remote sensing image super-resolution reconstruction based on the sparse representation are presented. Through blocking the remote sensing image, then the K-SVD algorithm is used to dictionary learning through analysing the high resolution remote sensing image library or remote sensing image itself,and obtain the dictionary which can sparsely represent the high resolution remote sensing image. Through feature extraction, dimension reduction from independent component analysis and reconstruction, high resolution remote sensing image is implementation. Simulation results show that the method improves higher peak-signal to noise ratio and the efficiency the algorithm.5 Structured sparse representation and dictionary learning for hyperspectral image classification is presented. Hyperspectral sample is excavated deeply incorporates both spectral and spatial characteristics, and two new technology of structured sparse representation and dictionary learning for hyperspectral image classification, the dictionary learning based on clustering are presented. Using linear SVM as classifier, completed the classification of hyperspectral remote sensing images and the effectiveness of the algorithm is verified by experiment. Through the study of the clustering of remote sensing images, the technology of clustered sparse representation and dictionary learning for hyperspectral image classification is presented, which incorporates both spectral and spatial characteristics of a sample clustered to obtain a dictionary of each pixels. The pixels have a common sparsity pattern in identical clustered groups. The sparse coefficients are obtained by the dictionary learning approach, and the sparse representation features of the remote sensing images are obtained.The theory of sparse representation and dictionary learning has very good prospects and development in the field of remote sensing image processing. This article made some exploration on the sparse representation and dictionary learning theory, and the application in the field of remote sensing image processing. Due to the sparse representation and dictionary learning theory emerged in recent years, research is just emerging in the field of remote sensing image processing, and remote sensing image processing based on sparse representation and dictionary learning research still has great potential.