Modeling And Simulations On CS-based Remote Sensing Video Imaging System

High-resolution video imaging satellite provides us real-time and nicety remote sensing data on the Earth.However,its application is also severely restricted by varieties of problems such as huge amounts of processing data and difficult remote sensing transmission.As a newly developed sampling theory,Compressed Sensing(CS)asserts that,when dealing with signals which are sparse or approximately sparse in a known basis,one can reconstruct the signals accurately or high precisely by solving sparse optimization problems using a small number of linear measurements obtained at a rate that is far below the classical Nyquist rate.This emerging sampling theory provides a new solution for relieving the pressure on the sampling devices of video imaging satellite.Aiming at some key problems which are more likely to appear in compressive sensed remote sensing video imaging system,the following aspects have been studied in this paper:Firstly,some corresponding measurement system models are established from the level of optical imaging principles.After a retrospect of the imaging principles of optical devices,we pointed out the conflicts between acquisition and compression existed in traditional optical imaging system which is based on Nyquist sampling law.Two representative measurement system models are also established according to the physically implemented location of measurement matrices.Secondly,various compressive sensed remote sensing video reconstruction models are researched,including single frame model,inter-frame differential model,residual model and residual distributed model.By introducing distributed compressed video sensing frame,the dictionary learning based distributed reconstruction model for compressive sensed remote sensing video imaging is also founded.Thirdly,systemic research on typical dictionary learning methods and sample selection methods has been developed.Lots of attentions are focused on one side information block matching sample generation method and two dictionary learning methods,namely,K-SVD and PCA.Based on independent component analysis theory,a global PICA dictionary learning method for video sparse representation is proposed.Finally,a systemic retrospect of high-precision CS convex optimization reconstruction algorithms has been carried out.Two variable splitting based convex optimization reconstruction algorithms,GPSR and LCGP,are emphatically discussed.Based on the KKT condition in convex optimization theory,the efficient subset variable updating based UWSV-SR algorithm is then proposed.By programming MATLAB simulation codes,several simulation experiments are conducted to verify those aforementioned models,dictionaries and reconstruction algorithms.Experimental signals involve one-dimensional sparse signals,two-dimensional digital images and three-dimensional video signals.The simulation results fully testify the proposed models and algorithms.