CN104598735A - Time-varying homotopy algorithm based on L1 homotopy recovery algorithm - Google Patents

Time-varying homotopy algorithm based on L1 homotopy recovery algorithm Download PDF

Info

Publication number
CN104598735A
CN104598735A CN201510034330.7A CN201510034330A CN104598735A CN 104598735 A CN104598735 A CN 104598735A CN 201510034330 A CN201510034330 A CN 201510034330A CN 104598735 A CN104598735 A CN 104598735A
Authority
CN
China
Prior art keywords
signal
homotopy
algorithm
time
value
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201510034330.7A
Other languages
Chinese (zh)
Inventor
李智
李健
杨谨
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sichuan University
Original Assignee
Sichuan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Sichuan University filed Critical Sichuan University
Priority to CN201510034330.7A priority Critical patent/CN104598735A/en
Publication of CN104598735A publication Critical patent/CN104598735A/en
Pending legal-status Critical Current

Links

Landscapes

  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

The invention relates to a time-varying homotopy algorithm based on an L1 homotopy recovery algorithm. The current compression perception research mainly aims at a static bounded signal, a blocking processing mode is usually adopted when a time-varying current signal is processed, but the problems of low efficiency, blocking effect, high delay and the like can be easily caused by adopting the mode. Aiming at the problems, the invention provides the time-varying homotopy algorithm for dynamic compression sensing by aiming at the time-varying current signal, compression sampling is performed by constructing an overlapping flow time window convection signal, and then an original signal is reconstructed in real time by using the L1 homotopy recovery algorithm. Experiments shows that the compression sampling can be quickly performed in real time by using the algorithm, the time-varying signal can be accurately reconstructed, the relative error of the reconstructed signal is controlled to be within a 10 to 2 range, and dynamic tracking compression sampling can be effectively performed on the time-varying signal.

Description

Time-varying homotopy algorithm based on L1 homotopy recovery algorithm
Technical Field
The invention belongs to the field of signal processing, and particularly relates to a compressed sensing reconstruction algorithm for a dynamic flow signal, which can be used for reconstructing medical time-varying signals such as electrocardiosignals.
Background
Electrocardiosignals are one of the bioelectric signals which are researched by people at the earliest and applied to clinical medicine, and along with the development of an electrocardio remote monitoring (ECG Telemonitor) system, the data volume of the electrocardiosignals acquired by the system is very large, which brings huge pressure to data transmission bandwidth and data storage. Therefore, a completely new theoretical framework for data acquisition and reconstruction is needed to solve the above problems of the conventional method.
In the modern signal processing field, the bandwidth frequency of the signal is higher and higher, and the traditional shannon-nyquist sampling theorem requires the sampling rate of the signal to reach at least twice of the highest frequency, which means that the traditional sampling theorem can not meet the practical requirement. Compressed Sensing (CS) is a completely new theory proposed by cans, Tao, Donoho, etc. in the field of signal processing in 06 years, and it utilizes the sparsity or compressibility of signals to convert the traditional shannon-nyquist sampling into random measurement sampling, thereby breaking through the limit of the shannon-nyquist sampling theorem, and this new theory has a wide application prospect in the field of signal processing.
Until now, compressed sensing has mostly been applied to static systems, often to process signals of a limited determined dimension. One assumes that the unknown signal is of finite length and uses a specific set of linear measurements to accomplish the reconstruction of the signal by solving an L1 norm minimization problem. In real life, signals such as sound, images and radio signals which are frequently contacted by people are flow signals which change in real time, and the most remarkable characteristic is that the signal dimension is not fixed, and complete signals cannot be acquired at one time, so that the traditional compressed sensing signal processing model is obviously not suitable for a real-time system. In order to solve the problem, N.Vaswani et al, Kalman Filtered Compressed sensing, 15th IEEE International Conference on Image Processing, 2008 ICIP 2008 IEEE 2008, 2008: 893-. The main idea is as follows: a long stream signal is divided into a plurality of relatively independent finite-dimension signals, and each block is reconstructed by compressing and sampling independently by using a traditional finite-dimension signal processing method. However, the defects of this block processing mode are also obvious, for example, the blocking effect caused by ignoring the correlation between blocks is obvious, the problem of high input and output delay caused by obtaining a long-stream signal block in advance is also brought, and the method cannot be applied to a system with high real-time tracking requirements, which greatly restricts the development of compressed sensing in a dynamic system.
Disclosure of Invention
In order to solve the problems of obvious blocking effect, high input and output delay and the like in the conventional blocking processing mode, the invention provides a time-varying homotopy algorithm based on an L1 homotopy recovery algorithm.
The invention adopts the following technical scheme:
and firstly, selecting a proper time window length to sample the flow signal. Selecting a Bernoulli matrix as a measurement matrix to perform compression sampling on signals under the same time window, and performing real-time recovery by using the existing L1 homotopy algorithm;
and secondly, clipping the obtained estimated value, and taking only the previous N values as a final estimated value. But the unused values are shifted up by N units and filled in later with N0 s. Using the signal value as an initial value of the next time;
and thirdly, sliding N units of the time window, correspondingly modifying the measurement matrix, and repeating the process to realize real-time tracking and sampling of the convection signal.
Drawings
FIG. 1 is a flow chart of the time-varying homotopy algorithm based on the L1 homotopy recovery algorithm of the present invention
FIG. 2 is a diagram illustrating the reconstruction of a 16384-long time-varying cardiac signal according to the present invention
FIG. 3 is a graph of the absolute error of reconstruction of a time-varying electrocardiosignal according to the invention
FIG. 4 is a graph of the relative error of the reconstruction of time varying cardiac signals of the present invention.
Detailed Description
The invention will be further described with reference to the accompanying drawings.
Referring to fig. 1, the time-varying homotopy algorithm based on the L1 homotopy recovery algorithm of the present invention includes the following steps:
step one, initialization: initializing the external iteration number I = 0;
step two, before the time-varying signalFast wavelet transform is carried out on the signal value X and the signal value X is projected to a perception matrix to obtain a signal value XDimension measurement vector Y:
wherein X isA is oneThe Bernoulli perception matrix of, wherein<<
Thirdly, reconstructing the original signal by utilizing an L1 homotopy recovery algorithm to obtain a reconstructed signalAnd taking the first N signal values to store as a final recovery estimation value;
step four, updating the perception matrix, namely after the perception matrix of the iterationAfter walkingWith columns as front of new sensing matrixBefore goingColumns, and the next diagonal regenerates a new setThe remaining elements are filled with 0. Where R is the down-sampling rate of the system, i.e.
Step five, updating the initial value of the L1 homotopy algorithm, namely reconstructing the iteration of this time to obtain a signalMove up N units and the next N values are filled with zeros;
step six, updating the iteration times: i = I + 1. And returning to the step two until the stream signal is finished.
Further, the reconstructing the original signal X by using the L1 homotopy algorithm described in step three includes the following steps:
(3-1) initialization: the number of iterations i =0 and,
(3-2) calculating homotopy path direction:
whereinIs composed ofSupport set of
(3-3) finding the maximum length of the optimal solution move without changing the homotopy path direction (i.e. the support set does not change), i.e. the maximum length of the optimal solution moveIncreased lengthComprises the following steps:
wherein,
(3-4) judgmentIf the value is greater than 1, if the value is greater than or equal to 1, jumping out of the loop to obtain a result:
if less than 1, updatingAnd
(3-5) judgmentWhether or not it is true, whenWhen, it means that x is inThe non-zero factor of (A) is set to 0, i.e. the support set of (X) is removed(ii) a When in useIt means that it needs to be added in the support concentration of xThe corresponding symbol vector is also changed;
(3-6) the number of iterations i = i +1, go to step (3-2).
The advantages of the present invention are further illustrated by the following simulation.
The experiment takes the measurement matrix window size asUpdating the signal value in one iterationAnd (4) respectively. The down-sampling rate of the system isI.e. on average, 128 hearts can be accurately reconstructed using 32 measurementsThe value of the electrical signal.
We set the starting values of the recovery algorithm as follows: when the ith iteration is finished, obtaining an estimated value of a section of signal, and setting the estimated value asAt i +1 iterations, the window of the signal is shifted down by R values and willAs the initial value of the homotopy algorithm of the next iteration L1 is very close to the final solution of the (i + 1) th iteration, the update of the optimal solution can be completed in a very short time.
All the recovered signal values are directly given in figure 2, and as can be seen from figure 4, the time-varying homotopy algorithm provided in the paper is utilized to compress and sample in real time and reconstruct a long section of electrocardiosignals, and the relative error of the recovered signals obtained by real-time tracking is always controlled at 10-2Within the range, the feasibility of the time-varying homotopy algorithm is proved from simulation experiments.
In conclusion, the time-varying electrocardiosignal compression sampling method can well compress, sample and reconstruct the time-varying electrocardiosignal.

Claims (4)

1. The time-varying homotopy algorithm based on the L1 homotopy recovery algorithm is characterized by comprising the following steps:
the method comprises the following steps: obtaining a reconstructed signal by using an L1 homotopy algorithmThen, only the first N signal values are taken and stored as the final signal recovery value;
step two: last timeIteratively reconstructed signalMoving up by N units, and filling the following N values with zeros as initial values of an L1 homotopy recovery algorithm used in the next iteration;
step three: considering signal correlation under a flow window of two iterations, the lower triangle content of a perception matrix in the last iteration is kept to be transferred to an upper triangle, and only one perception matrix is updatedThe submatrix of (a) is placed at the lower triangular position of the iteration sensing matrix.
2. With respect to the estimates of each final extraction in the flow overlap window as claimed in claim 1, we take the following:
assuming a signal under a flow window of a certain lengthUpdate per unit timeA signal value, a measured valueIs updated in a unit timeThe number of the measured values is,represents the down-sampling rate of the system; that we recover after each iterationFirst N out of the signal values ofThe value is used as the final recovery value, and the rest are not reserved and only used as the initial value of the recovery algorithm of the next iteration.
3. The determination of the initial value of the recovery algorithm when iterating down as claimed in claim 1, we take the following: when the ith iteration is finished, obtaining an estimated value of a section of signal, and setting the estimated value asAt i +1 iterations, the window of the signal is shifted down by R values and willThe zero padding is used as the initial value of the L1 homotopy algorithm of the next iteration, so that under the condition that a future signal value is unknown, no extra energy is added to the signal value as an estimation value, and the initial value is close to the final solution of the (i + 1) th iteration, so that the convergence speed of the recovery algorithm can be effectively improved.
4. The update problem of the perceptual matrix as claimed in claim 1, wherein we fully consider the correlation between two adjacent iterations, and apply the correlation to the perceptual matrix of the current iterationAfter walkingWith columns as front of new sensing matrixBefore goingColumns, and the next diagonal regenerates a new setThe remaining elements are filled with 0, where R is the down-sampling rate of the system, i.e.
CN201510034330.7A 2015-01-23 2015-01-23 Time-varying homotopy algorithm based on L1 homotopy recovery algorithm Pending CN104598735A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201510034330.7A CN104598735A (en) 2015-01-23 2015-01-23 Time-varying homotopy algorithm based on L1 homotopy recovery algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201510034330.7A CN104598735A (en) 2015-01-23 2015-01-23 Time-varying homotopy algorithm based on L1 homotopy recovery algorithm

Publications (1)

Publication Number Publication Date
CN104598735A true CN104598735A (en) 2015-05-06

Family

ID=53124514

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201510034330.7A Pending CN104598735A (en) 2015-01-23 2015-01-23 Time-varying homotopy algorithm based on L1 homotopy recovery algorithm

Country Status (1)

Country Link
CN (1) CN104598735A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112534842A (en) * 2018-08-07 2021-03-19 昕诺飞控股有限公司 Compressive sensing system and method using edge nodes of a distributed computing network

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112534842A (en) * 2018-08-07 2021-03-19 昕诺飞控股有限公司 Compressive sensing system and method using edge nodes of a distributed computing network
CN112534842B (en) * 2018-08-07 2024-05-28 昕诺飞控股有限公司 Compressive sensing system and method using edge nodes of a distributed computing network

Similar Documents

Publication Publication Date Title
CN103400402B (en) Based on the sparse compressed sensing MRI image rebuilding method of low-rank structure
WO2021093393A1 (en) Video compressed sensing and reconstruction method and apparatus based on deep neural network
CN103840838B (en) Method for Bayes compressed sensing signal recovery based on self-adaptive measurement matrix
CN103929649B (en) Reconstruction method for video compressed sensing
WO2018068629A1 (en) Signal sampling and recovery method and apparatus applicable to ovxdm system and ovxdm system
CN105827250A (en) Electric-energy quality data compression and reconstruction method based on self-adaptive dictionary learning
CN107527371B (en) Approximating smoothness L in compressed sensing0Design and construction method of norm image reconstruction algorithm
CN103873170B (en) A kind of frequency spectrum detecting method of compressed sensing under blind sparse condition
CN107481293B (en) Differential image compressed sensing reconstruction method based on multi-hypothesis weighting and intelligent terminal
CN108419083B (en) Image multilevel wavelet full subband compressed sensing coding method
CN102915562A (en) Compressed sensing-based multi-view target tracking and 3D target reconstruction system and method
CN103237204A (en) Video signal collection and reconfiguration system based on high-dimension compressed sensing
CN104779960A (en) A signal reconstruction method based on block compressed sensing
CN104574450A (en) Image reconstruction method based on compressed sensing
CN105050105A (en) High-energy-efficiency low-information-density data collecting method based on compressed sensing
CN105590296A (en) Dual-dictionary learning-based single-frame image super-resolution reconstruction method
CN101901493B (en) Method and system for multi-view image combined reconstruction based on compression sampling
CN113284202B (en) Image compression sensing method of scalable network based on content self-adaption
CN104598735A (en) Time-varying homotopy algorithm based on L1 homotopy recovery algorithm
CN106301384A (en) A kind of signal reconfiguring method based on splits&#39; positions perception
KR20210075826A (en) Method for Image Compressed Sensing based on Deep Learning via Learnable Spatial-Spectral transformation
CN106056554B (en) A kind of magnetic resonance fast imaging method of gradient field convolution sparse coding
CN105447894A (en) Compressed sensing (CS) reconstruction algorithm on the basis of Quasi Newton formula
CN107154061A (en) The regularization coding/decoding method that a kind of splits&#39; positions are perceived
CN103985100A (en) Partitioned compression sensing method based on self-adaptive observation combinational optimization

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20150506