Research On Theory And Methods Of Compressed Sensing Remote Sensing Video Imaging

A video imaging system provides high-resolution and real-time remote sensing(RS) data. However, the system also faces the problems of large amounts of data and RS transmission difficulties. Another common problem is completing video data compression and transmission while protecting image quality.The emergence of compressed sensing(CS) theory provides a radical idea to solve the aforementioned problem. CS is an efficient and fast-growing signal recovery framework. The basic principle of CS theory is that when the image of interest is significantly sparse or highly compressible, relatively few well-selected observations are sufficient to reconstruct the most significant non-zero components. CS provides a theory for remote sensing video compression. Thus, a CS-based RS imaging system is investigated in this study.Measurement and reconstruction models of CS-based RS video imaging are established. A CS-based RS imaging framework is also proposed. CS-based RS video imaging single-frame and differential measurements, as well as their corresponding reconstruction models, are proposed based on the correlation among video imaging sequences. The advantages and disadvantages of the two measurement and reconstruction models are analyzed through numerical simulations. Results show that the differential measurement reconstruction model improves the utilization efficiency of data, increases data compression ratio, and improves the quality of the reconstructed video.The aforementioned theory works on the following aspects.First, online sparse representation of RS video images is researched. The basic theory of online sparse representation of videos is introduced. Feedforward and feedback online sparse representation frameworks of videos are proposed. The advantages and disadvantages of the two online sparse representation frameworks are theoretically analyzed and compared. Based on the comparison results, the feedback online sparse representation framework is more suitable for the sparse representation of RS video images. Three types of training sample selection methods, namely, iterative block, the block matching algorithm, and the edge information method, are presented. The advantages and disadvantages of these sample selection methods are compared. The results show that the edge information method is the most ideal approach for sample selection. Two dictionary training methods, namely, the method of optimal direction(MOD) and K-singular value decomposition(K-SVD), are introduced. Global-principal component analysis(global-PCA) and independent component analysis(ICA) are then conducted. The results of the numerical simulation analysis and the comparison between the advantages and disadvantages of the four types of dictionary training methods show that ICA is more effective than MOD and K-SVD. Moreover, ICA reconstruction time is less than those of the other two techniques. Meanwhile, global-PCA has no significant advantage over the other two methods.Second, RS video image reconstruction techniques based on the method of Bregman are presented. The Bregman iterative method is introduced to solve single-frame and differential reconstruction models of CS-based RS video imaging. The classical Bregman iterative, linear Bregman iterative, and linear acceleration Bregman iteration methods are used to solve two reconstruction models under the L1-regularized condition. Six reconstruction algorithms are then proposed. The numerical simulation results show that the linear Bregman iteration method is easy to implement and exhibits good performance in solving the two models. The linear Bregman weighted iterative algorithms based on Lp-regularization are proposed under the Lp-regularization condition. The numerical simulation results show that Lp-regularization is sparser and exhibits better reconstruction performance than L1-regularization.Finally, three types of compression measurement system for an RS video compressed measurement system are proposed, namely,(1) phase modulation and subframe superimposition,(2) correlation estimation, and(3) dynamic estimation. The advantages and disadvantages of each compression measurement system are compared and analyzed through numerical simulations. The simulation results show that the first system(phase modulation and subframe superimposition) is helpful in improving imaging quality and increasing field of view. Meanwhile, correlation estimation decreases the video compression ratio of RS transmission data. Lastly, dynamic estimation is suitable for highly dynamic imaging scenes, and the video reconstructed from this system is continuous.The single pixel that functions as the contrast experiment is also investigated. A CS-based RS linear array imaging experiment using circuit compressive sampling is proposed. The experimental results show that the proposed imaging system exhibits higher efficiency than the single-pixel imaging system. The proposed imaging system has high practical value in RS satellite linear array imaging.