Signal compression and reconstruction method and device

Abstract

According to the invention, feature extraction of signals is realized through a time sequence learning method to obtain a feature dictionary, and low-dimensional observation signals are obtained through compression; and then a Lagrange equation is constructed at a signal reconstruction end according to the low-dimensional observation signal and the feature dictionary, a sparse coefficient is calculated through the Lagrange equation, and a reconstruction signal is obtained according to the sparse coefficient and the feature dictionary, so that accurate feature extraction of a large amount of data is realized, and accurate reconstruction of the signal of the to-be-monitored object is realized. According to the invention, the signal of the to-be-monitored object is converted into the low-dimensional observation signal and the feature dictionary for transmission, and the data volume of the two data is smaller than that of the acquired signal of the to-be-monitored object, so that waste of transmission bandwidth is avoided, and hardware resources for transmission and storage are effectively saved.

Description

Technical Field

[0001] The present invention relates to the field of signal processing, and in particular to a signal compression and reconstruction method and device. Background Art

[0002] Mechanical monitoring is crucial for verifying product reliability. Full-cycle monitoring can fully record data throughout a product's entire lifecycle. However, long-term monitoring generates enormous amounts of data, which poses challenges in data transmission and storage. Therefore, a method for compressing and accurately reconstructing signals is urgently needed. Summary of the Invention

[0003] In view of the above problems, the present invention is proposed to provide a signal compression reconstruction method and apparatus that overcomes the above problems or at least partially solves the above problems.

[0004] In a first aspect, an embodiment of the present invention provides a signal compression method, comprising:

[0005] Collecting signals from the object to be monitored;

[0006] Representing the signal in a one-dimensional matrix form to obtain a one-dimensional matrix;

[0007] Converting the one-dimensional matrix into a two-dimensional matrix;

[0008] Using a method based on sequential time learning, learning the two-dimensional matrix to obtain a feature dictionary;

[0009] compressing the two-dimensional matrix to obtain a low-dimensional observation signal;

[0010] The feature dictionary and the low-dimensional observation signal are transmitted to a signal reconstruction end.

[0011] In one embodiment, the method based on sequential time learning is used to learn the two-dimensional matrix to obtain a feature dictionary, including:

[0012] Calculating a first sparse coefficient according to the two-dimensional matrix;

[0013] A feature dictionary is calculated according to the first sparse coefficient.

[0014] In one embodiment, calculating the first sparse coefficient according to the two-dimensional matrix includes:

[0015] updating an initial residual value according to the column vector of the first column of the two-dimensional matrix and the initial reconstructed signal;

[0016] Find the maximum absolute value of the inner product of the transpose of the preset sparse dictionary and the initial residual value, and determine its index position; update the first sparse coefficient according to the index position, the initial first sparse coefficient and the following formula for updating the sparse coefficient:

[0017] X i+1 (b) = X i (b)+τ*sign(re(b));

[0018] Where: X is the first sparse coefficient, τ is the step size, re is the residual value, and b is the index position;

[0019] Updating the reconstructed signal according to the first sparse coefficient and the preset sparse dictionary;

[0020] Update the residual value according to the column vector of the second column of the two-dimensional matrix and the updated reconstructed signal;

[0021] Finding the maximum absolute value of the inner product of the transpose of the preset sparse dictionary and the updated residual value, and determining its index position; updating the first sparse coefficient according to the formula for updating the sparse coefficient;

[0022] Updating the reconstructed signal according to the first sparse coefficient and the preset sparse dictionary;

[0023] The column vectors of the two-dimensional matrix are processed in sequence until the 1-norm of the first sparse coefficient is less than a preset threshold or the number of iterations reaches a preset sparse coefficient iteration number threshold, thereby obtaining the first sparse coefficient.

[0024] In one embodiment, calculating a feature dictionary based on the first sparse coefficient includes:

[0025] Input the first sparse coefficient, the column vector of the first column of the two-dimensional feature matrix, and the initial feature dictionary into the following formula to update the feature dictionary;

[0026]

[0027] Where: A is the feature dictionary, f is the column vector of the two-bit feature matrix, and X is the first sparse coefficient;

[0028] Input the first sparse coefficient, the column vector of the second column of the two-dimensional feature matrix, and the updated feature dictionary into the above formula to update the feature dictionary;

[0029] The column vectors of the two-dimensional matrix are processed in sequence until the number of iterations reaches a preset feature dictionary iteration number threshold or the difference between two updated feature dictionaries is less than a preset difference threshold.

[0030] In one embodiment, compressing the two-dimensional matrix to obtain a low-dimensional observation signal includes:

[0031] The inner product of the two-dimensional matrix and a two-dimensional Gaussian random matrix is ​​used as the low-dimensional observation signal.

[0032] In a second aspect, an embodiment of the present invention provides a signal reconstruction method, comprising:

[0033] receiving the feature dictionary and the low-dimensional observation signal transmitted from a signal compression end;

[0034] Calculating a second sparse coefficient according to the feature dictionary and the low-dimensional observation signal;

[0035] A reconstructed signal is obtained according to the second sparse coefficient and the feature dictionary.

[0036] In one embodiment, calculating the sparse coefficient according to the feature dictionary and the low-dimensional observation signal includes:

[0037] Constructing a Lagrangian equation using the feature dictionary and the low-dimensional observation signal;

[0038] According to the Lagrange equation, the second sparse coefficient is calculated using an alternating direction multiplier method.

[0039] In one embodiment, obtaining a reconstructed signal according to the second sparse coefficient and the feature dictionary includes:

[0040] The inner product of the feature dictionary and the second sparse coefficient is used as the reconstructed signal.

[0041] In a third aspect, an embodiment of the present invention provides a signal compression device, including:

[0042] An acquisition module, used for acquiring signals from the object to be monitored;

[0043] A representation module, configured to represent the signal in a one-dimensional matrix form to obtain a one-dimensional matrix;

[0044] A conversion module, configured to convert the one-dimensional matrix into a two-dimensional matrix;

[0045] A learning module, configured to learn the two-dimensional matrix using a sequential time learning method to obtain a feature dictionary;

[0046] A compression module, used for compressing the two-dimensional matrix to obtain a low-dimensional observation signal;

[0047] A transmission module is used to transmit the feature dictionary and the low-dimensional observation signal to a reconstruction end.

[0048] In a fourth aspect, an embodiment of the present invention provides a signal reconstruction device, including:

[0049] A receiving module, configured to receive the feature dictionary and the low-dimensional observation signal transmitted from the learning compression end;

[0050] A sparse coefficient module, configured to calculate a second sparse coefficient based on the feature dictionary and the low-dimensional observation signal;

[0051] A reconstruction module is used to obtain a reconstructed signal according to the second sparse coefficient and the feature dictionary.

[0052] In a fourth aspect, an embodiment of the present invention provides a computing device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the program executed by the processor is the signal compression method described above.

[0053] In a fifth aspect, an embodiment of the present invention provides a computing device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the program executed by the processor implements a signal reconstruction method.

[0054] In a sixth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements a signal compression method when executed by a processor.

[0055] In a seventh aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements a signal reconstruction method when executed by a processor.

[0056] The beneficial effects of the above technical solutions provided by the embodiments of the present invention include at least:

[0057] The embodiment of the present invention uses a time-sequential learning method to extract signal features to obtain a feature dictionary, and obtains a low-dimensional observation signal through compression. Then, at the signal reconstruction end, a Lagrangian equation is constructed based on the low-dimensional observation signal and the feature dictionary, and a sparse coefficient is calculated using the Lagrangian equation. Then, a reconstructed signal is obtained based on the sparse coefficient and the feature dictionary. This achieves accurate feature extraction from a large amount of data, thereby achieving accurate reconstruction of the signal of the monitored object. By converting the signal of the monitored object into a low-dimensional observation signal and a feature dictionary for transmission, since the data volume of the above two data is smaller than the collected signal of the monitored object, waste of transmission bandwidth is avoided, and hardware resources for transmission and storage are effectively saved.

[0058] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purposes and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.

[0059] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0061] Figure 1 A flowchart of a signal compression method provided by an embodiment of the present invention;

[0062] Figure 2 A time domain diagram of a bearing fault signal with a length of 100,000 data points collected by a sensor provided in an embodiment of the present invention;

[0063] Figure 3 A time domain diagram of a feature dictionary provided by an embodiment of the present invention;

[0064] Figure 4 A time domain diagram of a low-dimensional observation signal provided by an embodiment of the present invention;

[0065] Figure 5 A flowchart of a signal reconstruction method provided by an embodiment of the present invention;

[0066] Figure 6 A time domain diagram of a reconstructed signal provided by an embodiment of the present invention;

[0067] Figure 7 The present invention provides an overall flowchart of a signal compression and reconstruction method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0068] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0069] In order to solve the problem that the transmission and storage of the aforementioned signal consumes a lot of bandwidth and storage hardware, and it is difficult to effectively extract the main features of a large amount of data, an embodiment of the present invention provides a signal compression method, the flow chart of which is as follows: Figure 1 Shown, including:

[0070] S11, collecting signals of the object to be monitored;

[0071] S12, expressing the signal in a one-dimensional matrix form to obtain a one-dimensional matrix;

[0072] S13, converting the one-dimensional matrix into a two-dimensional matrix;

[0073] S14, using a method based on sequential time learning, learning the two-dimensional matrix to obtain a feature dictionary;

[0074] S15, compressing the two-dimensional matrix to obtain a low-dimensional observation signal;

[0075] S16. Transmit the feature dictionary and the low-dimensional observation signal to the signal reconstruction end.

[0076] In S11, the object to be monitored may be, for example, a bearing, and a sensor may be used to collect the signal of the object to be monitored. For example, the bearing fault signal collected by the sensor is a signal with a length of 100,000 data points, and its time domain diagram may be, for example, as shown in FIG. Figure 2 shown; in Figure 2 In the example, since the signal is a discrete signal, the horizontal axis represents the time interval caused by the data point or sampling frequency. The aforementioned signal can be any mechanical signal or signal with a large amount of data to be transmitted.

[0077] Specifically, in the aforementioned step S13 , a one-dimensional matrix is ​​converted into a two-dimensional matrix by, for example, dividing the one-dimensional matrix into segments, and each segment is used as a column vector of each column of the two-dimensional matrix.

[0078] Taking the bearing fault signal collected by the sensor as an example, the length of the bearing fault signal to be monitored is 100,000, which is represented by a one-dimensional matrix. The one-dimensional matrix is ​​then folded into a two-dimensional matrix of 100*1000, where 100 represents the number of column vectors of the two-dimensional matrix, and 1000 represents the number of row vectors of the two-dimensional matrix. The 100*1000 two-dimensional matrix represents the one-dimensional matrix folded into 1000 groups of bearing fault signals with a length of 100.

[0079] In the aforementioned step S14, learning the two-dimensional matrix and obtaining the feature dictionary includes the following two steps:

[0080] Step 1: Calculate the first sparse coefficient according to the two-dimensional matrix;

[0081] Step 2: Calculate the feature dictionary based on the first sparse coefficient.

[0082] In one embodiment, the aforementioned step 1 can be performed as follows:

[0083] Update the initial residual value according to the column vector of the first column of the two-dimensional matrix and the initial reconstructed signal;

[0084] Find the maximum absolute value of the inner product between the transpose of the preset sparse dictionary and the initial residual value, and determine its index position; this step is based on the vector inner product, and selects the atom with the largest absolute value of the inner product as the direction for solution. The maximum value indicates that the sparse dictionary atom has the highest correlation with the signal, so this atom is selected as the gradient ascent direction;

[0085] Update the first sparse coefficient according to the index position, the initial first sparse coefficient and the following formula for updating the sparse coefficient:

[0086] X i+1 (b) = X i (b)+τ*sign(re(b));

[0087] Where: X is the first sparse coefficient; τ is the step size, which can be set to 0.01, for example; re is the residual value; b is the index position;

[0088] The sign function is a symbol function. When x>0, y=1; when x<0, y=-1; when x=0, y=0.

[0089] Updating the reconstructed signal according to the first sparse coefficient and a preset sparse dictionary;

[0090] Update the residual value according to the column vector of the second column of the two-dimensional matrix and the updated reconstructed signal;

[0091] Find the maximum absolute value of the inner product of the transpose of the preset sparse dictionary and the updated residual value, and determine its index position; update the first sparse coefficient according to the formula for updating the sparse coefficient;

[0092] The reconstructed signal is updated according to the first sparse coefficient and the preset sparse dictionary; for example, the reconstructed signal can be updated with reference to the following formula: E = A * X i+1 ;

[0093] The column vectors of the aforementioned two-dimensional matrix are sequentially substituted into the above iterative process until the first sparse coefficient is less than a preset threshold or the number of iterations reaches a preset sparse coefficient iteration number threshold, thereby obtaining the first sparse coefficient.

[0094] The aforementioned initial residual value, initial reconstructed signal, and initial first sparse coefficient can be set as a vector whose elements are all 0, and the initial residual value, initial reconstructed signal, and initial first sparse coefficient are vectors that are the same as the column vector; and the aforementioned sparse dictionary can be set as a random matrix, for example.

[0095] The sparse coefficients obtained according to the above method can be, for example:

[0096]

[0097] Where: f is the column vector of the two-dimensional matrix, f∈R N*1 ; A is a sparse dictionary, A∈R N*N ; X is the sparse coefficient, X∈R N *1 ; T is the sparse parameter, which is the preset threshold and can be set to 1;

[0098] Among them, ||f-AX|| F is the Frobnius norm of f-AX, which calculates the square root of the sum of the squares of the absolute values ​​of the elements of f-AX; ||X||1 is the 1-norm of X, also known as the sum norm, which calculates the maximum sum of the absolute values ​​of the elements of the X vector.

[0099] In one embodiment, the aforementioned step 2 may be performed as follows:

[0100] Input the first sparse coefficient, the column vector of the first column of the two-dimensional feature matrix, and the initial feature dictionary into the following formula to update the feature dictionary; the aforementioned initial feature dictionary can be set to a random matrix;

[0101]

[0102] Where: A is the feature dictionary, f is the column vector of the two-bit feature matrix, and X is the first sparse coefficient; ρ=1 / XX T (i,i), and XX T (i,i) is XX T The maximum absolute value of the diagonal;

[0103] Inputting the first sparse coefficient, the column vector of the second column of the two-dimensional feature matrix, and the updated feature dictionary into the formula for updating the feature dictionary to update the feature dictionary;

[0104] The column vectors of the aforementioned two-dimensional matrix are sequentially substituted into the above iterative process until the number of iterations reaches a preset feature dictionary iteration threshold or the difference between the two updated feature dictionaries is less than a preset difference threshold.

[0105] The expression of the final feature dictionary is as follows:

[0106]

[0107] Where A is the feature dictionary and f is the column vector of the two-dimensional matrix.

[0108] Taking the bearing fault signal collected by the above sensor as an example, the time domain diagram of the feature dictionary A obtained by sequential time learning can be as follows: Figure 3 shown.

[0109] In the aforementioned step S15, the low-dimensional observation signal can be obtained, for example, in the following manner:

[0110] The inner product of the two-dimensional matrix and the two-dimensional Gaussian random matrix is ​​used as the low-dimensional observation signal.

[0111] Taking the bearing fault signal as an example, the time domain diagram of the compressed low-dimensional observation signal can be shown as follows: Figure 4 shown.

[0112] Corresponding to the aforementioned signal compression method, the embodiment of the present invention further provides a signal reconstruction method, the flow chart of which is as follows: Figure 5 Shown, including:

[0113] S51, receiving the feature dictionary and the low-dimensional observation signal transmitted from the signal compression end;

[0114] S52, calculating a second sparse coefficient according to the feature dictionary and the low-dimensional observation signal;

[0115] S53: Obtain a reconstructed signal according to the second sparse coefficient and the feature dictionary.

[0116] In the aforementioned step S52, the second sparse coefficient is calculated, for example, in the following manner:

[0117] Construct the Lagrangian equation using the feature dictionary and low-dimensional observation signals;

[0118] According to the Lagrange equation, the second sparse coefficient is calculated using the Alternating Direction Method of Multipliers (ADMM).

[0119] The augmented Lagrange equation constructed by it has the following mathematical expression:

[0120]

[0121] By alternating the solution of P E and P U The saddle point of , and the gradient of the Lagrange multiplier terms U1 and U2 is increased to solve the maximum value. The final update formula is shown in the following formula:

[0122]

[0123] in:

[0124] C is an auxiliary variable, ρ and γ are augmented Lagrangian coefficients, PE is the minimum value of the objective function, PU is the minimum value of the objective function, U1 is Lagrangian coefficient 1, and U2 is Lagrangian coefficient 2; these are all iterative auxiliary parameters that only participate in iteration and not in output; X is the second sparse coefficient, b is the low-dimensional observation signal, and ψ = ΦA.

[0125] The solution to X is the sparse coefficient, and its solution is:

[0126] min||C||1

[0127] s.tb=ψX

[0128] X=C

[0129] Where ψ = ΦA, Φ is a Gaussian matrix.

[0130] In the above step S52, since the sparse coefficients calculated based on the feature dictionary and the low-dimensional observation signal are not exactly the same as the original sparse coefficients, they are distinguished by the first sparse coefficients and the second sparse coefficients; and the data volume of the original sparse coefficients is equal to the signal itself, which wastes bandwidth and storage resources when transmitted; the present invention utilizes the feature dictionary and the low-dimensional observation signal, and the length of these two sets of data and the sum of the data points are much smaller than the characteristics of the signal itself. Therefore, compared with directly transmitting the sparse coefficients, transmitting the feature dictionary and the low-dimensional observation signal uses less bandwidth and storage resources; thereby achieving the purpose of saving bandwidth and storage resources.

[0131] The above step S53 can be implemented, for example, by the following method:

[0132] The inner product of the feature dictionary and the second sparse coefficient is used as the reconstructed signal. The time domain diagram of the reconstructed signal is as follows Figure 6 shown.

[0133] In one embodiment, the flowchart of the present invention compressing and reconstructing the entirety is as follows: Figure 7 As shown, in Figure 7 In the proposed method, the collected signals are respectively learned through feature matrix to obtain feature dictionary, and low-dimensional observation signal is obtained through signal compression; and the feature dictionary and low-dimensional observation signal are transmitted from the signal compression end to the signal reconstruction end; and then the reconstructed signal is obtained according to the reconstruction algorithm.

[0134] Based on the same inventive concept, an embodiment of the present invention further provides a signal compression device, comprising:

[0135] An acquisition module, used for acquiring signals from the object to be monitored;

[0136] A representation module is used to represent the signal in a one-dimensional matrix form to obtain a one-dimensional matrix;

[0137] A conversion module, used for converting a one-dimensional matrix into a two-dimensional matrix;

[0138] A learning module is used to learn a two-dimensional matrix using a sequential time learning method to obtain a feature dictionary;

[0139] Compression module, used to compress the two-dimensional matrix to obtain low-dimensional observation signals;

[0140] The transmission module is used to transmit the feature dictionary and low-dimensional observation signals to the reconstruction end.

[0141] Based on the same inventive concept, an embodiment of the present invention further provides a signal reconstruction device, comprising:

[0142] A receiving module, configured to receive a feature dictionary and a low-dimensional observation signal transmitted from a learning compression end;

[0143] A sparse coefficient module, configured to calculate a second sparse coefficient based on a feature dictionary and a low-dimensional observation signal;

[0144] The reconstruction module is used to obtain a reconstructed signal according to the second sparse coefficient and the feature dictionary.

[0145] Based on the same inventive concept, an embodiment of the present invention further provides a computing device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements a signal compression method when executing the program.

[0146] Based on the same inventive concept, an embodiment of the present invention further provides a computing device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements a signal reconstruction method when executing the program.

[0147] Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, a signal compression method is implemented.

[0148] Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, a signal reconstruction method is implemented.

[0149] Since the principles of the problems solved by these devices are similar to the aforementioned signal compression method and signal reconstruction method, the implementation of the devices can refer to the implementation of the aforementioned methods, and the repeated parts will not be repeated.

[0150] The present invention uses a time-sequential learning method to extract signal features to obtain a feature dictionary, and then compresses the low-dimensional observation signal. Then, at the signal reconstruction end, a Lagrangian equation is constructed based on the low-dimensional observation signal and the feature dictionary, and a sparse coefficient is calculated using the Lagrangian equation. The reconstructed signal is then obtained based on the sparse coefficient and the feature dictionary. This allows for accurate feature extraction from a large amount of data, thereby achieving accurate reconstruction of the signal of the monitored object. By converting the signal of the monitored object into a low-dimensional observation signal and a feature dictionary for transmission, since the amount of these two data is smaller than the collected signal of the monitored object, waste of transmission bandwidth is avoided, effectively saving hardware resources for transmission and storage.

[0151] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0152] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0153] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0154] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1a process or multiple processes and / or boxes Figure 1 The present invention is a block or multiple blocks that specify the functions of the steps. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such modifications and variations fall within the scope of the claims and their equivalents, the present invention is intended to include such modifications and variations.

Claims

1. A signal compression method, characterized in that: include: Collecting signals from the object to be monitored; Representing the signal in a one-dimensional matrix form to obtain a one-dimensional matrix; Converting the one-dimensional matrix into a two-dimensional matrix; Using a method based on sequential time learning, learning the two-dimensional matrix to obtain a feature dictionary; compressing the two-dimensional matrix to obtain a low-dimensional observation signal; The feature dictionary and the low-dimensional observation signal are transmitted to a signal reconstruction end.

2. The method according to claim 1, wherein The method based on sequential time learning is used to learn the two-dimensional matrix to obtain a feature dictionary, including: Calculating a first sparse coefficient according to the two-dimensional matrix; A feature dictionary is calculated according to the first sparse coefficient.

3. The method according to claim 2, wherein: Calculating a first sparse coefficient according to the two-dimensional matrix includes: updating an initial residual value according to the column vector of the first column of the two-dimensional matrix and the initial reconstructed signal; Find the maximum absolute value of the inner product of the transpose of the preset sparse dictionary and the initial residual value, and determine its index position; update the first sparse coefficient according to the index position, the initial first sparse coefficient and the following formula for updating the sparse coefficient: X i+1 (b)=X i (b)+τ*sign(re(b)); Where: X is the first sparse coefficient, τ is the step size, re is the residual value, and b is the index position; Updating the reconstructed signal according to the first sparse coefficient and the preset sparse dictionary; Update the residual value according to the column vector of the second column of the two-dimensional matrix and the updated reconstructed signal; Finding the maximum absolute value of the inner product of the transpose of the preset sparse dictionary and the updated residual value, and determining its index position; updating the first sparse coefficient according to the formula for updating the sparse coefficient; Updating the reconstructed signal according to the first sparse coefficient and the preset sparse dictionary; The column vectors of the two-dimensional matrix are processed in sequence until the 1-norm of the first sparse coefficient is less than a preset threshold or the number of iterations reaches a preset sparse coefficient iteration number threshold, thereby obtaining the first sparse coefficient.

4. The method according to claim 2, wherein The step of calculating a feature dictionary according to the first sparse coefficient includes: Input the first sparse coefficient, the column vector of the first column of the two-dimensional feature matrix, and the initial feature dictionary into the following formula to update the feature dictionary; Where: A is the feature dictionary, f is the column vector of the two-bit feature matrix, and X is the first sparse coefficient; Input the first sparse coefficient, the column vector of the second column of the two-dimensional feature matrix, and the updated feature dictionary into the above formula to update the feature dictionary; The column vectors of the two-dimensional matrix are processed in sequence until the number of iterations reaches a preset feature dictionary iteration number threshold or the difference between two updated feature dictionaries is less than a preset difference threshold.

5. The method according to claim 1, wherein The compressing the two-dimensional matrix to obtain a low-dimensional observation signal includes: The inner product of the two-dimensional matrix and a two-dimensional Gaussian random matrix is ​​used as the low-dimensional observation signal.

6. A signal reconstruction method, characterized in that: include: receiving the feature dictionary and the low-dimensional observation signal transmitted from a signal compression end; Calculating a second sparse coefficient according to the feature dictionary and the low-dimensional observation signal; A reconstructed signal is obtained according to the second sparse coefficient and the feature dictionary.

7. The method according to claim 6, wherein The calculating of sparse coefficients according to the feature dictionary and the low-dimensional observation signal includes: Constructing a Lagrangian equation using the feature dictionary and the low-dimensional observation signal; According to the Lagrange equation, the second sparse coefficient is calculated using an alternating direction multiplier method.

8. The method according to claim 6, wherein Obtaining a reconstructed signal according to the second sparse coefficient and the feature dictionary includes: The inner product of the feature dictionary and the second sparse coefficient is used as the reconstructed signal.

9. A signal compression device, characterized in that: include: An acquisition module, used for acquiring signals from the object to be monitored; A representation module, configured to represent the signal in a one-dimensional matrix form to obtain a one-dimensional matrix; A conversion module, configured to convert the one-dimensional matrix into a two-dimensional matrix; A learning module, configured to learn the two-dimensional matrix using a sequential time learning method to obtain a feature dictionary; A compression module, used for compressing the two-dimensional matrix to obtain a low-dimensional observation signal; A transmission module is used to transmit the feature dictionary and the low-dimensional observation signal to a reconstruction end.

10. A signal reconstruction device, characterized in that: include: A receiving module, configured to receive the feature dictionary and the low-dimensional observation signal transmitted from the learning compression end; A sparse coefficient module, configured to calculate a second sparse coefficient based on the feature dictionary and the low-dimensional observation signal; A reconstruction module is used to obtain a reconstructed signal according to the second sparse coefficient and the feature dictionary.

11. A computing device, characterized in that include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the program executed by the processor implements the signal compression method according to any one of claims 1 to 5.

12. A computing device, characterized in that include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the program executed by the processor implements the signal reconstruction method according to any one of claims 6 to 8.

13. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the signal compression method according to any one of claims 1 to 5 is implemented.

14. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the signal reconstruction method according to any one of claims 6 to 8 is implemented.

Images