Research On Structured Data Processing Technology Based On Bayesian Theory

With the rise of mobile internet and the advent of the era of big data,there is an increasing demand for signal processing,which brings huge challenges to signal acquisition,transmission and storage.On the one hand,with the deployment of a large number of low-power sensors,the existing acquisition method which needs sampling first,then encoding and decoding no longer meets the existing needs.There is an urgent need for acquisition and recovery algorithms for signal data collected in low-power sensor-based information collection scenarios(such as,medical monitoring scenarios in wireless body area networks(WBANs),farmland environmental monitoring scenarios in wireless sensor networks(WSNs),hyperspectral imaging in hyperspectral remote sensing(HSR),etc.).On the other hand,the existing signal processing technology puts forward higher requirements for the research on compressed sensing and affine rank minimization based on sparse and low-rank structure,including higher compression ratio,faster speed,and higher recovery or reconstruction accuracy.Based on the feature of low-rank and sparse structure of data collected in WBANs and WSNs,this thesis not only proposes a general recovery algorithm of low-rank and joint sparse(L&S),but also proposes a specific L&S fast recovery algorithm for multi-leads L&S data.In addition,for 3D(3-Dimensional,3D)hyperspectral images(HSI)in HSR,an L&S data recovery algorithm that combines with the collection methods is proposed,as well as two 2-Dimensional(2D)L&S data recovery algorithms for recovery HSI data are presented.The contributions are summarized as follow:Firstly,the low-rank and sparse structures of medical data such as ECG and EEG in WBANs are analyzed,and the problem is modeled with MMV based on compressed sensing technology.Then we turn the MMV problem into a block single measurement vector problem,and illustrate the L&S signal covariance matrix structure for signal recovery.Based on this structure,an L&S signal recovery algorithm based on block sparse Bayesian learning is proposed,which improves the signal reconstruction performance.The process of problem solving is disassembled into two steps.In the first step,we solve the initial values of all hyperparameters.In the second step,we use the obtained initial values of hyperparameters to obtain the best reconstruction parameters through the upper limit approximation,then reconstruct the source signal.By using synthetic data simulation and actual signal simulation experiments,it is proved that the algorithm proposed in this thesis has better performance than existing algorithms,and can be applied to all L&S structure data recovery.Secondly,aiming at solving the problem of recovering L&S structure data with better performance,this thesis analyzes the characteristics of multi-lead data in the sparse low-rank domain and its corresponding principal component distribution.Based on the structure of the L&S signal covariance matrix,the relationship between the indicator factor of the covariance matrix and the signal principal component distribution is found.Moreover,a fast recovery algorithm for multi-lead L&S signal based on Sparse Bayesian Learning(SBL)is proposed.Simulation experiments on synthetic data and real data verify the effectiveness and feasibility of the proposed fast algorithm.Finally,this thesis extends the algorithm for 2D data to solve the problem of 3D data acquisition and transmission.The existing push-broom acquisition methods of HSI data are analyzed,and low-rank sparse analysis and principal component analysis are performed on the corresponding push-broom sliced data.HSI 3D data is decomposed into 2D data from two perspectives,and two L&S data recovery algorithms are proposed.Then,an L&S signal reconstruction algorithm based on Bayesian theory similar to distributed compressed sensing is proposed.We first use the CS algorithm to compress and sample the sliced data collected by push-broom on the satellite,and then reconstruct all the received data at the ground station.The simulation results show the effectiveness of the proposed algorithm.On the basis of greatly improving the compression ratio,the reconstruction performance is improved.