Data element differential privacy encryption and transmission method based on controllable compressed sensing
By using a data element differential privacy encryption method based on controllable compressed sensing, data compression and privacy protection are integrated, solving the problems of high resource consumption and poor flexibility in existing technologies. This method is suitable for resource-constrained devices and meets the privacy protection requirements of real-time monitoring systems.
Patent Information
- Application Number
- CN202511198973.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-26
- Publication Date
- 2025-11-28
AI Technical Summary
Existing technologies for data privacy protection suffer from problems such as separation of privacy protection and data compression, lack of flexibility in measurement matrix encryption, high hardware resource consumption, and the impact of differential privacy on data quality, making it difficult to meet the complex needs of real-time monitoring systems.
A differential privacy encryption method based on controllable compressed sensing is adopted. By encrypting the measurement matrix with sparsity processing and pseudo-random numbers, data compression and privacy protection are integrated. The encryption level can be precisely controlled by parameter adjustment, which is suitable for resource-constrained hardware devices.
It improves processing efficiency, reduces hardware resource consumption, ensures data quality, adapts to privacy protection needs in different scenarios, and broadens the scope of applications.
Smart Images

Figure CN121037059A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data transmission and data privacy protection technology, and in particular to a data element differential privacy encryption and transmission method based on controllable compressed sensing. Background Technology
[0002] With the acceleration of digitalization, image and video data elements are being used more and more widely across various fields. However, privacy and security issues are becoming increasingly prominent during data transmission and storage. Real-time monitoring systems in public places collect data containing a large amount of sensitive information, and leaks could pose a serious threat to personal privacy and public safety. For example, in scenarios such as smart security, smart homes, and industrial monitoring, improper acquisition or misuse of data could lead to issues such as the exposure of personal identities and the leakage of trade secrets, making privacy protection a critical issue that urgently needs to be addressed.
[0003] To address these challenges, researchers have proposed numerous methods to ensure the privacy and security of data elements, such as compressed sensing (CS) and differential privacy; however, existing technologies have the following shortcomings in data privacy encryption:
[0004] Firstly, privacy protection and compression are separated: Currently, most methods for solving privacy and security issues in consumer electronics devices treat privacy protection and data compression as separate steps. For example, while methods such as structure obfuscation and multi-level encryption protect data privacy to some extent, they are only implemented after the compression stage is completed. This makes it impossible to integrate compression and privacy protection, resulting in a complex and inefficient overall process.
[0005] Secondly, the measurement matrix encryption lacks flexibility: In technologies related to privacy protection using compressed sensing, pseudo-random number generators (PRNGs) are often used to generate projection matrices (measurement matrices). On the one hand, the number of PRNGs depends on the number of users, making it difficult to generate multiple projection matrices; on the other hand, there is a lack of effective control over the amount of error introduced into the projection matrix, making it difficult to precisely adjust the encryption level according to actual needs and failing to flexibly meet the privacy protection requirements of different scenarios.
[0006] Third, it consumes a lot of hardware resources: Although existing privacy protection technologies can achieve compression and data hiding, they are computationally complex and require a large amount of hardware resources. This is almost unsuitable for resource-constrained hardware devices, such as sensors based on resource-constrained hardware in real-time monitoring systems, which severely limits their application scope in real-world scenarios.
[0007] Fourth, differential privacy applications have limitations: Although differential privacy methods have been introduced into privacy protection during data transmission, such as in network training models and federated learning, when applied to data privacy computation, the simple method of differential privacy, which protects privacy by injecting noise, can significantly impact data availability, leading to a decline in data quality and failing to meet the data quality requirements of real-time monitoring systems.
[0008] In summary, existing privacy protection methods, whether based on encryption, authentication, or other means, only address specific aspects and fail to form a unified and efficient solution. These methods often compromise on other aspects, failing to simultaneously address multiple key factors such as privacy protection strength, data processing efficiency, hardware resource consumption, and data quality, thus making it difficult to meet the increasingly complex needs of consumer electronics and real-time monitoring systems. Summary of the Invention
[0009] To overcome the problem that existing data privacy protection methods cannot simultaneously balance privacy protection strength, data processing efficiency, hardware resource consumption, and data quality, this invention proposes a data element differential privacy encryption method based on controllable compressed sensing. This method can efficiently compress data, precisely control the degree of privacy protection, and is applicable to resource-constrained hardware devices.
[0010] This invention proposes a differential privacy encryption method for data elements based on controllable compressed sensing, comprising the following steps:
[0011] S1. Perform sparsification on the original data to obtain the sparse coefficient vector X;
[0012] S2, Component Measurement Matrix Φ M×N Linear projection is performed on the signal to obtain the encoded signal Y; M is the dimension of Y, N is the dimension of X, and M is much smaller than N;
[0013] S3. Divide X into data blocks x of size B×B. i To improve compression efficiency, a block compression sensing measurement matrix Φ is generated corresponding to the data block. B ; Obtain the output vector y of each data block after M compressed sensing operations. i ;Φ B For n B ×B 2 The matrix, Indicates rounding down;
[0014] S4. Using pseudo-random numbers R to measure the block compressed sensing matrix Φ B Encrypt some columns to generate each Φ B Corresponding encrypted measurement matrix
[0015] S5. For each small block x i Encryption measurement matrix Encryption is performed to obtain the compressed signal y'i.
[0016] Preferably, the measurement matrix Φ in step S2 M×N It satisfies the restricted isometric property.
[0017] Preferably, in step S4, the matrix Φ is first... B The columns are divided into Q groups. A portion of columns from each group is XORed with the 0 and 1 sequence [R] of the pseudo-random number R. Then, the matrices from the XORed groups are concatenated to form the encrypted measurement matrix.
[0018] Preferably, the sparse coefficient vector X in step S1 is obtained by the following steps:
[0019] First, we set the optimization objective of basis pursuit as follows: Given that the original signal and the sparse coefficient vector X satisfy a dictionary matrix mapping, we want to make X... -Minimum norm;
[0020] The basis pursuit problem is then transformed into a standard linear programming problem, and a linear programming algorithm is used to solve for the sparse coefficient vector X.
[0021] Repeat the above steps to iteratively update the sparse coefficient vector X until the convergence condition is met.
[0022] Preferably, during the iteration of the sparse coefficient vector X, the objective function is:
[0023]
[0024] Where f(X) is the objective function, and the optimization objective is to minimize the objective function; λ is the equilibrium parameter; Let denote the square of the 2-norm, and D be a dictionary matrix;
[0025] The gradient is:
[0026]
[0027] in, The gradient is represented by `sign(.)`, which is the sign function; the superscript `T` indicates matrix transpose.
[0028] The data encryption transmission method proposed in this invention first encrypts the data using the differential privacy encryption method based on controllable compressed sensing, then transmits the data, and finally encrypts the measurement matrix after transmission. The original data is reconstructed using linear estimation algorithms.
[0029] Preferably, a decoding matrix is used. The compressed signal y'i is restored to generate the database x. i Reconstruction signal
[0030]
[0031] Among them, R xx It is x i The autocorrelation matrix; assuming the existence of a linear observation model y i =Φ B x i +e, where Φ B Let e be the measurement matrix, and τ be zero-mean Gaussian white noise. 2 The variance of the noise is represented by I, and the identity matrix is represented by y. i For data block x i The output vector after M compressed sensing operations.
[0032] Preferred:
[0033]
[0034] Where ρ is the set correlation coefficient, ρ∈(0,1); G is an integer G=B 2 .
[0035] The present invention proposes a data element differential privacy encryption system based on controllable compressed sensing, comprising a memory and a processor. The memory stores a computer program, and the processor is connected to the memory. The processor is used to execute the computer program to implement the data element differential privacy encryption method based on controllable compressed sensing.
[0036] The present invention proposes a storage medium storing a computer program, which, when executed, is used to implement the data element differential privacy encryption method based on controllable compression sensing.
[0037] The advantages of this invention are:
[0038] (1) This invention proposes a data element differential privacy encryption method based on controllable compressed sensing. It performs differential privacy encryption based on compressed sensing, deeply integrating the core principle of compressed sensing with the differential privacy protection mechanism. Data compression and encryption are completed synchronously in a single stage of data processing, abandoning the traditional cumbersome mode, greatly improving processing efficiency, and providing an efficient privacy protection solution for related systems.
[0039] (2) This invention integrates compression and privacy protection. Without adding extra complex steps, it utilizes the controllable encryption of the measurement matrix to complete data compression and encryption in a single stage. This changes the traditional method's separation of privacy protection and compression functions, simplifies the processing flow, and improves processing efficiency. Furthermore, this invention can precisely control the encryption level of the measurement matrix and flexibly adjust the privacy protection level according to different scenario requirements, balancing privacy and data availability.
[0040] (3) Controllable privacy protection is achieved by using parameter p. While ensuring the privacy protection effect, the consumption of hardware resources is reduced as much as possible, and it is compatible with resource-constrained devices. Compared with complex encryption mechanisms, it reduces the computational overhead, can run effectively on resource-constrained hardware devices, broadens the application scope, and provides a brand-new solution for solving the privacy and security problems of real-time monitoring systems.
[0041] (4) A unique measurement matrix encryption method was designed. By performing an XOR operation between the column elements of the measurement matrix and the bits of the controllable key stream generated by the logical AND operation of the key stream R and the privacy protection parameter p, the encryption level of the measurement matrix can be precisely controlled. This method overcomes the lack of flexibility in traditional measurement matrix encryption methods. It can flexibly adjust the encryption strength according to the privacy protection requirements of different scenarios, ensuring that while meeting privacy protection requirements, the impact on data availability is minimized, and the measurement matrix encryption is precisely controlled.
[0042] (5) Balancing privacy protection and data quality: This invention achieves a controllable measurement matrix encryption method through parameter settings, minimizing the impact on data quality under different privacy protection levels, and effectively balancing the relationship between privacy protection and data quality while protecting data privacy; ensuring that in application scenarios such as real-time monitoring systems, even after privacy protection processing, the data can still maintain high availability and meet actual usage needs. Attached Figure Description
[0043] Figure 1 Here is a flowchart of a data element differential privacy encryption method based on controllable compressed sensing proposed in this invention;
[0044] Figure 2 for Figure 1 The step-by-step flowchart of step S4. Detailed Implementation
[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0046] like Figure 1 As shown in the figure, the differential privacy encryption method for data elements based on controllable compressed sensing proposed in this embodiment includes the following steps S1-S6.
[0047] S1. Perform sparsification on the original data s to obtain the sparse coefficient vector X.
[0048] Specifically, in this step, the basis pursuit algorithm is used to achieve sparsification of the original data. The basis pursuit problem is transformed into a standard linear programming (LP) problem, and the sparse coefficient vector X is initialized as a random vector. The sparse coefficients X are iteratively updated until the convergence condition is met.
[0049] Basis Pursuit (BP) is a method that... - An algorithm that uses norm minimization to solve for sparse representations. Its goal is to represent a signal s as a sparse coefficient vector X under a dictionary D, i.e.:
[0050] s = DX (1)
[0051] Where X is a sparse vector, meaning most of its elements are zero. The core idea of basis pursuit is to... - Norm minimization to approximate -norm, thus finding sparse solutions.
[0052] The specific steps of step S1 are as follows:
[0053] S11. Set the optimization objective of basis pursuit: Under the premise of satisfying formula (1), make the sparse coefficient vector X such that... -Minimum norm; the formula is as follows:
[0054]
[0055] in: It is an observation signal. It is a dictionary matrix. It is a sparse coefficient vector; N>n, where N is the dimension of the sparse vector X, and n is the dimension of the observed signal s; ||X||1 is the sparse coefficient vector of X. -norm;
[0056] S12. Transform the basis pursuit problem into a standard linear programming (LP) problem. The specific form is as follows:
[0057] X = uv (3)
[0058]
[0059] Where u and v are auxiliary variables, representing the non-negative and non-positive parts of X respectively, formula (3) achieves the accurate decomposition of the sparse coefficient vector X; here, by introducing auxiliary variables u and v, the non-smooth - Norm transforms into a smooth linear programming problem; the superscript T indicates matrix transpose, and "1" indicates a column vector consisting entirely of 1s. Where O is the number of dimensions of X, X i Let u represent the value of the i-th dimension of X. i and u i Let u and v be the values of the i-th dimension, respectively;
[0060] S13. Select the linear programming (LP) algorithm to solve formulas (2)-(4), initialize the sparse coefficient vector X as a zero vector or a random vector, and update the sparse coefficient vector X iteratively until the convergence condition is met.
[0061] The gradient descent iterative process is as follows:
[0062] Objective function:
[0063]
[0064] Where f(X) is the objective function, and the optimization objective is to minimize the objective function; λ is a balancing parameter used to achieve a balance between sparsity constraints and data fitting accuracy; This represents the square of the 2-norm.
[0065] The gradient is:
[0066]
[0067] in, Let be the gradient, sign(s) be the sign function, D be the dictionary matrix, and the superscript T denotes the matrix transpose.
[0068] Update sparsity coefficient:
[0069]
[0070] Where k and k+1 represent the number of iterations, and η is the step size parameter; X (k+1) Let X be X after k+1 iterations. (k) Represents X after k iterations;
[0071] In this embodiment, the projection gradient method is used, which requires projecting X onto the feasible region (i.e., the set of conditions that satisfy s = DX) after each iteration.
[0072] Use the following conditions to determine whether the iteration has converged:
[0073] ||X (k+1) -X(k) ||2<ε (8)
[0074] Where ε is the preset convergence threshold.
[0075] If the convergence condition is met, the iteration stops; otherwise, the iteration continues. Finally, the converged X is output as the sparse coefficient vector X of the original data, or simply signal X.
[0076] S2, through the measurement matrix Φ M×N Linear projection of signal X yields a CS (compressed sensing) encoded signal Y containing M data values; M < <N;
[0077] Y M×1 =Φ M×N X N×1 (9)
[0078] Among them, Y M×1 This means that Y contains M data points, and X... N×1 This means that X contains N data points.
[0079] Thus, in step S2, the signal Y is obtained from the sparse coefficient vector X using compressed sensing at a rate lower than that of the Nyquist sampling rate, which is beneficial to improving the efficiency of subsequent data processing.
[0080] The accuracy of compressed sensing largely depends on the selection of a suitable measurement matrix Φ. M×N In order to accurately recover X from Y; measurement matrix Φ M×N The Restricted Isometry Property (RIP) must be satisfied to recover the best possible estimate of X from the coded signal Y. The RIP formula is expressed as follows:
[0081]
[0082] Here, δ represents the setting parameter of the isometric property (RIP).
[0083] Bernoulli random matrices have been proven to satisfy the RIP condition, making them suitable for CS compression and reconstruction. According to available theoretical results, the value of δ is between (0,1). Therefore, Bernoulli random matrices can be specifically used as the measurement matrix Φ for compressed sensing. M×N Formula (11) gives the structure of the Bernoulli random matrix.
[0084] Φ M×N ={Φ j,c |1≤j≤M,1≤c≤N}
[0085]
[0086] Where, Φ j,c Represents the measurement matrix ΦM×N The value in row j and column c is randomly selected as 1 or -1.
[0087] In practice, a Bernoulli random matrix can be generated using physically non-cloning functions (PUFs). The storage resources required for the Bernoulli random matrix are very small, so only an adder is needed to implement the CS encoder in order to implement the encoding process of formula (9).
[0088] S3. Divide vector X into data blocks x of size B×B. i To improve compression efficiency, a block compression sensing measurement matrix Φ is generated corresponding to the data block. B ; Obtain the output vector of each data block after M compression sensing operations;
[0089] The formula is expressed as:
[0090] y i =Φ B x i (12)
[0091] Where, x i For the i-th B×B data block, y i For x i The output vector after M compressed sensing operations; Φ B It is an n B ×B 2 The matrix, This indicates rounding down to the nearest integer.
[0092] x i The dimension is B×B, that is, B 2 ×1; According to formula (12), y i The dimension is n B ×1, all y i Combining them together yields Y with dimension M, which has dimension M×1.
[0093] Thus, the measurement matrix Φ M×N It can be represented as a block diagonal matrix, with the following form:
[0094]
[0095] In this step, block compressed sensing is achieved by dividing the vector X into blocks, transforming the storage of the M×N measurement matrix into n... B ×B 2 Bernoulli random matrix Φ B The storage is more efficient in terms of memory usage. Furthermore, from the perspective of multi-rate signal processing, block compressed sensing can be implemented using a random two-dimensional filter bank, where each finite impulse response filter H... g(z1,z2)(0≤g≤n B -1) has a scope of B×B. A trade-off needs to be made when choosing the block dimension B: a smaller B value requires less storage memory and is faster to implement; while a larger B value results in better reconstruction performance. In practice, B is typically chosen between 8 and 64.
[0096] S4. Using pseudo-random numbers R to measure the block compressed sensing matrix Φ B Encrypt some columns to generate each Φ B Corresponding encrypted measurement matrix
[0097] In practice, the pseudo-random number R is converted into n B A sequence of 0s and 1s of ×1, then with Φ B Perform an XOR operation on the selected columns to achieve the desired result for n. B A sequence of 0s and 1s of ×1, then with Φ B Perform an XOR operation on the columns to achieve the measurement matrix Φ B Encryption reduces information leakage in compressed sensing.
[0098] make and Φ B Columns; Φ B The dimension is n B ×B 2 A column has n dimensions. B ×1, i.e., Φ B There should be B. 2 Each column.
[0099] The encryption column is randomly selected according to the ratio p% and encrypted to obtain Φ. B Encryption measurement matrix p takes values between 0 and 100.
[0100] Reference Figure 2 In practice, step S4 can be divided into the following sub-steps:
[0101] S41, Matrix Φ B The columns are divided into Q groups. The first A columns of each group are XORed with the 0 and 1 sequence [R] of the pseudo-random number R; A = p% × B 2 / Q;
[0102] In practice, A can be rounded up or down, and the values of A in each group can be adjusted to ensure that the sum of the values of A in each group is Φ. B The nearest integer to p% of the column number is sufficient.
[0103] S42. Concatenate the matrices after the XOR operation to form an encrypted measurement matrix.
[0104] In practice, Q can be set to 10 to facilitate matrix column grouping. XORing the grouped columns ensures a uniform distribution of the encrypted columns, thereby improving encryption quality.
[0105] S5. For each small block x i Encryption measurement matrix Encryption is performed to obtain y'i; thus, we get:
[0106]
[0107] X N×1 x divided into B×B i , recorded as Therefore, the above formula can be written as:
[0108]
[0109] in, Encryption measurement matrix for X The encryption result; n B Much smaller than B×B.
[0110] This embodiment also proposes a data encryption transmission method. First, data encryption is performed using steps S1-S5 as described above. Then, the data is transmitted. After transmission is complete, the following method is used... The original data is reconstructed using the minimum mean square error (MMSE) linear estimation algorithm. This ensures that the data sender uses block compression to store simple structured data, while also guaranteeing that the receiver quickly obtains a good initial solution.
[0111] During data encryption, vector X is split into smaller blocks x. i After processing, each small block x is obtained. i The corresponding compressed signal is y'i; therefore, the signal reconstruction process is to restore the signal y'i into small blocks x. i That is, according to x i Get small piece x i Reconstruction signal
[0112] In this embodiment, the data reconstruction process includes the following steps:
[0113] S61. The data reconstruction objective is determined as follows:
[0114]
[0115] in, Here is the decoding matrix; E represents the expectation;
[0116] Assume there exists a linear observation model yi =Φ B x i +e, where Φ B Here is the measurement matrix, e is additive noise, and it is assumed that the noise e is zero-mean Gaussian white noise with a covariance matrix τ. 2 I, where τ 2 Let I represent the variance of the noise, and let I represent the identity matrix.
[0117] S62, Derivation Matrix
[0118]
[0119] Expanding the above equation, we get:
[0120]
[0121] Since E[e]=0 and E[ee] T ]=τ 2 I, and since the sparse coefficient signals X and e are independent, the above equation can be simplified to:
[0122]
[0123] To find the one that minimizes the mean square error Regarding mean square error Find the partial derivative and set it to zero. tr represents the trace of the matrix, which is the sum of all elements along the main diagonal of the square matrix (from the top left corner to the bottom right corner).
[0124] Based on the relevant knowledge of matrix differentiation, and through a series of derivations, we can obtain:
[0125]
[0126] Among them, R xx =E[x i x i T ] is data block x i The autocorrelation matrix.
[0127] S63. The autocorrelation matrix R is generated by simulating the first-order autoregressive model AR(1). xx Then substitute into formula (21) to get Thus, the signal reconstruction formula is obtained.
[0128] For natural images, the gray values of adjacent pixels are usually strongly correlated, and this correlation weakens as the spatial distance increases, which is consistent with the statistical characteristics of the AR(1) model.
[0129] Assume the i-th data block x iThe pixel value at any position h within the inner edge is x i If (h) is a pixel at position l, then its correlation with the pixel value at position l is determined by the distance d = |hl|. Assume the correlation coefficient is ρ, representing the correlation strength between adjacent pixels (d = 1), and ρ ∈ (0, 1), ensuring that the correlation decays exponentially with distance. The value of ρ can be obtained by analyzing the autocorrelation coefficient and partial autocorrelation coefficient of the samples based on the characteristics of the data.
[0130] Transfer the B×B data block x i Vectorized into length G = B 2 For a one-dimensional signal, the autocorrelation matrix R xx Let R be a symmetric matrix of size G×G, where R xx (h,l) represents the autocorrelation values of the pixels at positions k and l, i.e.:
[0131] R xx (h,l)=E[x i (h)x i (l)] (22)
[0132] For any positions h and l, define the distance d = |hl|, then the autocorrelation value is: R xx (h,l)=ρ d Since spatial correlation is independent of positional order, R xx It is a symmetric matrix, i.e., R xx (h,l)=R xx (l,h). Its matrix structure can be represented as:
[0133]
[0134] Substituting into formula (22), the reconstruction matrix can be calculated. The signal can then be reconstructed using the reconstruction matrix. For the complete Φ M×N The cost of computing MMSE solutions is too high. In block compressed sensing, due to Φ... B Much smaller in size, The computational cost will be significantly reduced. The MMSE solution refers to the solution directly calculated using the minimum mean square error (MMSE) linear estimation algorithm for the complete matrix.
[0135] Of course, those skilled in the art will recognize that the present invention is not limited to the details of the exemplary embodiments described above, but also includes the same or similar structures that can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
[0136] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.
[0137] The technologies, shapes, and structures not described in detail in this invention are all known technologies.
Claims
1. A differential privacy encryption method for data elements based on controllable compressed sensing, characterized in that, Includes the following steps: S1. Perform sparsification on the original data to obtain the sparse coefficient vector X; S2, Component Measurement Matrix Φ M×N The signal is linearly projected to obtain the encoded signal Y; M is the dimension of Y, N is the dimension of X, and M is much smaller than N; S3. Divide X into data blocks x of size B×B. i To improve compression efficiency, a block compression sensing measurement matrix Φ is generated corresponding to the data block. B ; Obtain the output vector y of each data block after M compressed sensing operations. i ;Φ B For n B ×B 2 The matrix, Indicates rounding down; S4. Using pseudo-random numbers R to measure the block compressed sensing matrix Φ B Encrypt some columns to generate each Φ B Corresponding encrypted measurement matrix S5. For each small block x i Encryption measurement matrix Encryption is performed to obtain the compressed signal y'i.
2. The data element differential privacy encryption method based on controllable compressed sensing as described in claim 1, characterized in that, The measurement matrix Φ in step S2 M×N It satisfies the restricted isometric property.
3. The data element differential privacy encryption method based on controllable compressed sensing as described in claim 1, characterized in that, In step S4, firstly, matrix Φ B The columns are divided into Q groups. A portion of columns from each group is XORed with the 0 and 1 sequence [R] of the pseudo-random number R. Then, the matrices from the XORed groups are concatenated to form the encrypted measurement matrix.
4. The data element differential privacy encryption method based on controllable compressed sensing as described in claim 1, characterized in that, The sparse coefficient vector X in step S1 is obtained by following these steps: First, the optimization objective of basis pursuit is set as follows: under the premise that the original signal and the sparse coefficient vector X satisfy the dictionary matrix mapping, the l1-norm of X is minimized. The basis pursuit problem is then transformed into a standard linear programming problem, and a linear programming algorithm is used to solve for the sparse coefficient vector X. Repeat the above steps to iteratively update the sparse coefficient vector X until the convergence condition is met.
5. The data element differential privacy encryption method based on controllable compressed sensing as described in claim 4, characterized in that, During the iteration of the sparse coefficient vector X, the objective function is: Where f(X) is the objective function, and the optimization objective is to minimize the objective function; λ is the equilibrium parameter; Let denote the square of the 2-norm, and D be a dictionary matrix; The gradient is: in, The gradient is represented by `sign(.)`, which is the sign function; the superscript `T` indicates matrix transpose.
6. A data encryption transmission method employing the data element differential privacy encryption method based on controllable compressed sensing as described in any one of claims 1-5, characterized in that, First, the data is encrypted using the differential privacy encryption method for data elements based on controllable compressed sensing, then the data is transmitted. After transmission, an encrypted measurement matrix is then used. The original data is reconstructed using linear estimation algorithms.
7. The data encryption transmission method as described in claim 6, characterized in that, Using a decoding matrix The compressed signal y'i is restored to generate the database x. i Reconstruction signal Among them, R xx It is x i The autocorrelation matrix; assuming the existence of a linear observation model y i =Φ B x i +e, where Φ B Let e be the measurement matrix, and τ be zero-mean Gaussian white noise. 2 The variance of the noise is represented by I, and the identity matrix is represented by y. i For data block x i The output vector after M compressed sensing operations.
8. The data encryption transmission method as described in claim 6, characterized in that: Where ρ is the set correlation coefficient, ρ∈(0,1); G is an integer G=B 2 .
9. A data element differential privacy encryption system based on controllable compressed sensing, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, the processor is connected to the memory, and the processor is used to execute the computer program to implement the data element differential privacy encryption method based on controllable compressed sensing as described in any one of claims 1-5.
10. A storage medium, characterized in that, The system contains a computer program that, when executed, implements the data element differential privacy encryption method based on controlled compressed sensing as described in any one of claims 1-5.
Citation Information
Cited By
Encryption compression transmission method and system for data of intelligent fusion terminal
CN121217487A
