Method for protecting privacy account based on big data joint training

Abstract

The invention discloses a method for protecting a privacy account based on big data joint training, and relates to a method for protecting the privacy account. The invention aims to solve the problem that a traditional privacy account protection means cannot really achieve the purpose of protecting a privacy account of a user. The method comprises the following steps: establishing a user account data management system; inputting user common login position information and password input speed information into a user account data management system, and training the input information; when the user logs in the account, the user account data management system compares the common login position information of the user with the password input speed information, and after the comparison succeeds, a verification image is displayed; performing sharpness processing on the verification image; after verification succeeds, the user is allowed to enter, otherwise, the abnormal condition transmitting module transmits abnormal condition information to the user; and after the user confirms the abnormal condition information, authorization information is transmitted to the abnormal condition user confirmation module. The invention belongs to the technical field of network account security.

Description

Technical Field

[0001] The invention relates to a method for protecting a privacy account, and belongs to the technical field of network account security. Background Art

[0002] With the continuous development of the Internet around the world, people's work and life have become inseparable from the Internet. More and more people have their own private accounts on the Internet, and some people even have multiple personal private accounts. The increase in private accounts has brought a series of security issues. Therefore, how to protect and manage users' private accounts has become an urgent problem for various Internet operators. At present, the commonly used means of protecting private accounts are to set up private questions and bind them to users' mobile phones. Once an abnormal situation occurs, the system will send a notification to the user as soon as possible to avoid causing greater losses to the user. However, this protection method is actually just a preventive mechanism and cannot truly achieve the purpose of protecting users' private accounts. Summary of the invention

[0003] The present invention aims to solve the problem that traditional privacy account protection means cannot truly achieve the purpose of protecting user privacy accounts, and further proposes a method for protecting privacy accounts based on big data joint training.

[0004] The technical solution adopted by the present invention to solve the above-mentioned problem is: the steps of the present invention include:

[0005] Step 1: Establish a user account data management system;

[0006] Step 2: Input the user's frequently used login location information and password input speed information into the user account data management system, and train the input information;

[0007] Step 3: When the user logs into the account, the user account data management system compares the user's common login location information and password input speed information, and after the comparison is successful, displays the verification image;

[0008] Step 4: Clarify the verification image;

[0009] Step 5: After successful verification, the user is allowed to enter; otherwise, the abnormal situation transmitting module transmits abnormal situation information to the user;

[0010] Step 6: After the user confirms the abnormal situation information, the user sends the authorization information to the abnormal situation user confirmation module. After the abnormal situation user confirmation module receives the user's authorization information, the user privacy account management module allows access, otherwise, the user privacy account management module is locked.

[0011] Furthermore, the structure of the management system in step 1 includes:

[0012] Data collection module, data analysis module, data storage module, OLAP analysis engine and OLTP read-write engine;

[0013] The data acquisition module is used to collect multimodal data and write the data into the data storage module;

[0014] The data analysis module is used to divide the collected data of the data collection module and write the divided data into the data storage module;

[0015] The data storage module is used to store the divided data;

[0016] The OLAP analysis engine is used to perform statistical queries on the stored data;

[0017] The OLTP read-write engine is used to call the stored data and perform addition, deletion, modification and query on the stored data in tuples;

[0018] The divided data includes relational data and file data;

[0019] The data storage module includes n column storage nodes, n row storage nodes and an LSM-Tree structure;

[0020] The n column storage nodes and the n row storage nodes are used to store relational data;

[0021] The LSM-Tree structure is used to store file-type data;

[0022] The LSM-Tree structure includes an LSM file system and a Blob file system.

[0023] Furthermore, the management method of the management system in step 1 is:

[0024] Step 101: Collect multimodal data based on the data acquisition module, and write the multimodal data into the data analysis module;

[0025] Step 102: dividing the collected data into relational data and file data based on the data analysis module;

[0026] Step 103: intelligently selecting row storage, column storage, or using both row storage and column storage at the same time according to user needs, and storing relational data based on n column storage nodes and n row storage nodes in the data storage module;

[0027] Storing file-type data based on the LSM-Tree structure;

[0028] Step 104: Perform statistical query on the stored relational data and file-type data based on the OLAP analysis engine, and call the stored relational data and file-type data based on the OLTP read-write engine and perform addition, deletion, modification and query in tuple units.

[0029] Furthermore, the step of storing relational data based on n column storage nodes and n row storage nodes in step 103 includes:

[0030] Step 1031: Arrange the relational data and file data into a table format, assign a unique table ID in the cluster to each relational table, and assign a unique row / column ID in the table to each row of data in the relational data table;

[0031] Step 1032: Use table ID + row / column ID as the key of each tuple, and the content of the tuple as the corresponding value;

[0032] Step 1033: For the indexes of relational data tables and file-type data tables, use table ID + index ID + index value as the key of each row of index data, and the data row / column ID pointed to by the index as the value;

[0033] Step 1034: For the meta information of the relational data table and the file-type data table, a specific prefix + table ID is used as the key, and the structure definition of the relational data table is used as the value to store;

[0034] Step 1035: Use block storage technology to process relational data and file data in blocks, divide the entire key-value pair space into segments, each segment is a series of continuous keys, each segment is used as a block, and limit the data stored in each block to be less than a preset threshold;

[0035] Step 1036: Use a distributed read / write engine to store relational data and file data in n row / column storage nodes in blocks, and manage them in backup groups.

[0036] Furthermore, the step of storing file-type data based on the LSM-Tree structure in step 103 includes:

[0037] Step a: Assign a unique ID as the key to each file in the file-type data, and the file itself as the value;

[0038] Step b: At the flush and compaction time of the LSM-Tree, the value of the file memory exceeding the preset threshold is separated from the LSM-tree and stored in the Blob file object;

[0039] Step c, the LSM file system uses SST as the basic unit for storage. Each key in SST is a file ID, and the value is a Blob object pointer of the file.

[0040] Step d: Each Blob object is stored using a block backup strategy, with each block size being 100MB. The Blob file header block stores the file's meta information and a pointer to each block group of the file.

[0041] Furthermore, the step of clearing the verification image in step 4 includes:

[0042] Step 401, point spread function analysis of the hypersurface unit;

[0043] Step 402, image degradation, i.e. image restoration modeling and image preprocessing: performing imaging result analysis based on the point spread function modeling result of the hypersurface unit, and performing simple preprocessing on the acquired original image;

[0044] Step 403, obtaining a denoiser; selecting a denoiser according to the optical system model and the image degradation model obtained in step 402;

[0045] Step 404: construct a regularization term based on the denoiser obtained in step 403; use the output metric of the denoiser as part of the regularization by defining a penalty function related to the denoiser output;

[0046] Step 405, constructing an objective function by combining the data fidelity term and the regularization term constructed in step 404;

[0047] Step 406: using an optimization algorithm to solve the objective function constructed in step 405 to obtain a denoised hypersurface imaging result; iteratively updating an estimated value of the denoised image, and using a denoiser for processing in each iteration;

[0048] Step 407: Use the image preprocessed in step 402 as the input of the deep learning network, use the denoised image as the output of the deep learning network, perform self-supervised training based on the penalty function in the objective function constructed in step 405 until convergence, and obtain a trained deep learning network;

[0049] Step 408: input the original image to be tested into the trained deep learning network, and the trained deep learning network outputs the denoised image.

[0050] Furthermore, the specific process of step 401 is as follows:

[0051] When the optical system is linearly degraded, that is, the point spread function space is unchanged, without considering the noise, the imaging process is expressed by the following formula:

[0052]

[0053] In formula (1), I(x, y) represents the degraded image, S(u, v) represents the original image, and P(xu, yv) represents the spatially invariant point spread function;

[0054] The image degradation process of the spatially varying point spread function with the field of view is expressed as:

[0055]

[0056] In formula (2), P(u,v,xu,yv) represents the spatially varying point spread function.

[0057] Furthermore, the specific process of step 402 is as follows:

[0058] The image degradation process can be regarded as a linear space-invariant degradation system, and its process can be described as: the clear image f(x, y) is convolved with the corresponding blur kernel h(x, y) to generate the degraded image g(x, y) under the influence of noise;

[0059] g(x,y)=h(x,y)*f(x,y)+n(x,y) (3),

[0060] In formula (3), * represents the convolution operator;

[0061] When the point spread function is spatially invariant, the blurred image is expressed as:

[0062]

[0063] The blurred image is represented as:

[0064]

[0065] Furthermore, the regularization term expression in step 404 is:

[0066]

[0067] In formula (6), R(y) represents the regularization term, D σ (y) represents the denoiser, y represents the preprocessed image, and the superscript T represents the transposition;

[0068] Step 5 The objective function is expressed as:

[0069]

[0070] In formula (7), x represents the target image, y represents the preprocessed image, and l(y, x) represents the data fidelity term, that is, to ensure y≈x. The bit represents the penalty degree for R(y), and the weights between the data fidelity term and the regularization term are adjusted according to actual needs to balance the relationship between the two. λR(y) represents the penalty function.

[0071] Furthermore, the specific process of step 406 is as follows:

[0072] Step 40601: Construct the augmented Lagrangian function:

[0073]

[0074] Formula (8), f(x) represents the intermediate variable, is defined as, p(y│x represents the prior probability of finding x given y, g(x) represents the intermediate variable, p(x) represents the probability of x, f(x)+λg(v) is the original optimization objective, v represents the target image; under the condition that x = v is satisfied, (x - v) is the equality constraint condition, u is the Lagrange multiplier related to the constraint (x - v) = 0, and 蟻 represents the penalty coefficient;

[0075] Step 40602: Set the maximum number of iterations K;

[0076] Step 40603: The number of iterations k = 1;

[0077] Step 40604: To solve the minimum value problem, convert the augmented Lagrangian function into the following sub-problem:

[0078]

[0079] u (k+1) = u (k) + (x (k+1) - v (k+1) )(11),

[0080] ρ k+1 = γ k ρ k (12),

[0081] In formulas (9), (10), (11) and (12), x (k+1) represents the target image of the (k + 1)-th iteration, x (k) represents the target image of the k-th iteration, ρ k represents the penalty coefficient of the k-th iteration, ρ k+1 represents the penalty coefficient of the (k + 1)-th iteration, v (k) represents the target image of the k-th iteration, v (k+1)represents the target image of the k+1th iteration, u (k) represents the Lagrange multiplier associated with the constraint (xv) = 0 at the kth iteration, u (k+1) represents the Lagrange multiplier associated with the constraint (xv) = 0 at the k+1th iteration, D σk represents the denoiser at the kth iteration, γ k Represents ρ k The adjustment parameter, D σk is the denoising algorithm, is a parameter that controls the strength of the denoiser, and improves ant to ρ k+1 =γ k ρ k , γ k ≥1 is the adjustment parameter of Ant.

[0082] Step 40605: Set the number of iterations k = k + 1, and repeat step 405 until the maximum number of iterations K is reached to obtain x K That is, the target image x in the objective function satisfies corresponding;

[0083] The specific process of step 407 is:

[0084] The image preprocessed in step 402 is used as the input of the deep learning network, the target image x obtained in step 406 is used as the output of the deep learning network, and self-supervised training is performed based on the penalty function in the objective function constructed in step 405 until convergence to obtain a trained deep learning network.

[0085] The beneficial effects of the present invention are:

[0086] 1. The present invention uses a method of comparing location information and password input speed simultaneously to ensure the security of user privacy account information and avoid account theft or privacy information leakage; compared with traditional privacy account protection methods, the present invention not only has a reminder and prevention mechanism, but also can lock user privacy account information in real time to prevent it from being stolen and leaked by outsiders; the user-side positioning information module of the present invention can collect the location information of the logon user in real time. Once the logon user is a criminal, his location information will be collected and saved, providing information assistance for the subsequent capture of the criminal;

[0087] 2. The present invention utilizes the denoising regularization method to constrain the complexity of the model by introducing a regularization term, thereby achieving effective noise suppression. Compared with the traditional denoising method, this method can better retain the original features and structural information of the image while removing noise;

[0088] 3. The present invention uses denoising to constrain the solution space of the final data, which is similar to the effect of the regularization term using prior knowledge to constrain the solution space. Therefore, the denoiser is embedded in the optimization framework, and the denoising algorithm is used to define the regularization term, so as to achieve the regularization effect while realizing denoising. While obtaining high-quality data, the generalization ability of the model is enhanced, the denoising accuracy is improved, and the error between the denoised image and the real image is reduced;

[0089] 4. The present invention realizes efficient distributed storage of relational data and file-type data through a key-value-based storage method. For relational data, the present invention stores data in units of rows, ensuring the consistency and high availability of multiple copies of data. At the same time, the present invention also designs a storage mode that supports efficient distributed query processing, meets ACID constraints, and provides users with persistence and read-write services. For file-type data, the present invention separates the value and stores it separately, reducing the write amplification phenomenon, saving I / O bandwidth and CPU resources, and providing efficient read-write performance;

[0090] 5. The present invention realizes efficient distributed storage of relational data and file-type data through a key-value-based storage method. For relational data, the present invention stores data in units of rows, ensuring the consistency and high availability of multiple copies of data. At the same time, the present invention also designs a storage mode that supports efficient distributed query processing, meets ACID constraints, and provides users with persistence and read-write services. For file-type data, the present invention separates the value and stores it separately, which reduces the write amplification phenomenon, saves I / O bandwidth and CPU resources, and provides efficient read-write performance. This efficient distributed storage and query processing capability will greatly improve the performance and availability of data management systems, and provide users with faster and more stable data services;

[0091] 6. The present invention uses multiple copies to ensure data consistency and robustness. Whether it is relational data or file data, it can be backed up and stored in the system through multiple copies to ensure data security and reliability. This guarantee of data consistency and reliability will greatly reduce the risk of data management system, avoid the possibility of data loss or damage, and protect the user's data assets;

[0092] 7. The technical solution of the present invention provides efficient data processing and analysis capabilities through storage expansion technology based on column storage structure. Storing in columns can solve the problem of slow data analysis in row storage, while also providing the performance advantages of high-speed updates and batch reading. This efficient data processing and analysis capability will greatly enhance the value of data utilization by users and accelerate the development of data-driven scientific research and engineering practice;

[0093] 8. The technical solution of the present invention provides users with a more flexible and intelligent data storage and management method. Users can write multiple key-value pairs at one time without worrying about whether these key-value pairs are on the same data slice, which improves the flexibility of data writing. At the same time, the present invention also provides a compatibility interface to the database, so that the database can intelligently choose to use row storage or column storage, and select the most appropriate storage method according to actual needs. This flexibility and intelligence will greatly simplify the data management process and improve the efficiency and quality of data management;

[0094] 9. The technical solution of the present invention has good scalability and adaptability. Whether facing the ever-increasing amount of data or the changes in data processing requirements in the future, the system of the present invention can be flexibly adjusted and expanded to maintain efficient and stable operation. This scalability and adaptability will provide users with long-term and reliable data management solutions and provide a solid foundation for business development;

[0095] 10. The technical solution of the present invention covers key technical research in relational data storage, file-based data storage, column storage structure, etc., and has high technical innovation and industry leadership. Through in-depth research and innovative practice in the field of data management, the present invention will bring new technological breakthroughs and development directions to the industry, and promote the continuous progress and improvement of the entire data management field. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] Figure 1 It is a structural diagram of the user account data management system;

[0097] Figure 2 It is the LSM-Tree structure diagram of the user account data management system that supports file storage;

[0098] Figure 3 It is a flow chart of a management method of a user account data management system;

[0099] Figure 4 It is a mapping flow chart of relational data to sword-finger structure of user account data management system;

[0100] Figure 5 It is a flow chart of the data block storage and grouping scheme of the user account data management system;

[0101] Figure 6 It is a flowchart of the file data block division scheme of the user account data management system;

[0102] Figure 7 It is a storage flow chart of the column structure and distributed storage system of the user account data management system;

[0103] Figure 8is a schematic diagram of the image degradation and restoration model; DETAILED DESCRIPTION

[0104] Specific implementation method 1: Figures 1 to 8 As shown, a method for protecting privacy accounts based on big data joint training, step 1, establishing a user account data management system;

[0105] Step 2: Input the user's frequently used login location information and password input speed information into the user account data management system, and train the input information;

[0106] Step 3: When the user logs into the account, the user account data management system compares the user's common login location information and password input speed information, and after the comparison is successful, displays the verification image;

[0107] Step 4: Clarify the verification image. The specific steps include:

[0108] Step 401, point spread function analysis of the hypersurface unit;

[0109] When the optical system is linearly degraded, that is, the point spread function space is unchanged, without considering the noise, the imaging process is expressed by the following formula:

[0110]

[0111] In formula (1), I(x, y) represents the degraded image, S(u, v) represents the original image, and P(xu, yv) represents the spatially invariant point spread function;

[0112] The image degradation process of the spatially varying point spread function with the field of view is expressed as:

[0113]

[0114] In formula (2), P(u,v,xu,yv) represents the spatially varying point spread function;

[0115] Step 402, image degradation, i.e. image restoration modeling and image preprocessing: performing imaging result analysis based on the point spread function modeling result of the hypersurface unit, and performing simple preprocessing on the acquired original image;

[0116] The image degradation process can be regarded as a linear space-invariant degradation system, and its process can be described as: the clear image f(x, y) is convolved with the corresponding blur kernel h(x, y) to generate the degraded image g(x, y) under the influence of noise;

[0117] g(x,y)=h(x,y)*f(x,y)+n(x,y) (3),

[0118] In formula (3), * represents the convolution operator;

[0119] When the point spread function is spatially invariant, the blurred image is represented as:

[0120]

[0121] The blurred image is represented as:

[0122]

[0123] Step 403, obtaining a denoiser; selecting a denoiser according to the optical system model and the image degradation model obtained in step 402;

[0124] Step 404: construct a regularization term based on the denoiser obtained in step 403; use the output metric of the denoiser as part of the regularization by defining a penalty function related to the denoiser output;

[0125] The regularization term expression is:

[0126]

[0127] In formula (6), R(y) represents the regularization term, D σ (y) represents the denoiser, y represents the preprocessed image, and the superscript T represents the transposition;

[0128] Step 5 The objective function is expressed as:

[0129]

[0130] In formula (7), x represents the target image, y represents the preprocessed image, l(y,x) represents the data fidelity term, that is, ensuring y≈x, λ represents the degree of penalty for R(y), and the weight between the data fidelity term and the regularization term is adjusted according to actual needs to balance the relationship between the two, and λR(y) represents the penalty function; Step 405, combining the data fidelity term and the regularization term constructed in step 404 to construct the objective function;

[0131] Step 406: using an optimization algorithm to solve the objective function constructed in step 405 to obtain a denoised hypersurface imaging result; iteratively updating an estimated value of the denoised image, and using a denoiser for processing in each iteration;

[0132] Step 40601, construct the Lagrangian augmented function:

[0133]

[0134] Formula (8), f(x) represents the intermediate variable, is defined as. \(p(y|x)\) represents the prior probability of finding \(x\) given \(y\). \(g(x)\) represents an intermediate variable, \(p(x)\) represents the probability of \(x\), \(f(x)+\lambda g(v)\) is the original optimization objective, and \(v\) represents the target image. Under the condition that \(x = v\), \((x - v)\) is the equality constraint condition, \(u\) is the Lagrange multiplier related to the constraint \((x - v)=0\), and \(\lambda\) represents the penalty coefficient.

[0135] Step 40602: Set the maximum number of iterations \(K\).

[0136] Step 40603: Let the iteration count \(k = 1\).

[0137] Step 40604: To solve the minimum value problem, transform the augmented Lagrangian function into the following sub-problem:

[0138]

[0139] u (k+1) = u (k) +(x (k+1) - v (k+1) )(11),

[0140] ρ k+1 = γ k ρ k (12),

[0141] In formulas (9), (10), (11), and (12), \(x\) (k+1) represents the target image of the \((k + 1)\)-th iteration, \(x\) (k) represents the target image of the \(k\)-th iteration, \(\rho\) k represents the penalty coefficient of the \(k\)-th iteration, \(\rho\) k+1 represents the penalty coefficient of the \((k + 1)\)-th iteration, \(v\) (k) represents the target image of the \(k\)-th iteration, \(v\) (k+1) represents the target image of the \((k + 1)\)-th iteration, \(u\) (k) represents the Lagrange multiplier related to the constraint \((x - v)=0\) of the \(k\)-th iteration, \(u\) (k+1) represents the Lagrange multiplier related to the constraint \((x - v)=0\) of the \((k + 1)\)-th iteration, \(D\) σk represents the denoiser of the \(k\)-th iteration, \(\gamma\) k represents the adjustment parameter of \(\rho\) k and \(D\) σk is a denoising algorithm. is a parameter that controls the strength of the denoiser. \(\lambda\) is improved to \(\rho\) k+1 = γ k ρ k and \(\gamma\) k ≥ 1 is the adjustment parameter of \(\lambda\).

[0142] Step 40605: Set the number of iterations k = k + 1, and repeat step 405 until the maximum number of iterations K is reached to obtain x K That is, the target image x in the objective function satisfies corresponding;

[0143] The specific process of step 407 is:

[0144] The image preprocessed in step 402 is used as the input of the deep learning network, the target image x obtained in step 406 is used as the output of the deep learning network, and self-supervised training is performed based on the penalty function in the objective function constructed in step 405 until convergence, thereby obtaining a trained deep learning network;

[0145] Step 407: Use the image preprocessed in step 402 as the input of the deep learning network, use the denoised image as the output of the deep learning network, perform self-supervised training based on the penalty function in the objective function constructed in step 405 until convergence, and obtain a trained deep learning network;

[0146] Step 408: input the original image to be tested into the trained deep learning network, and the trained deep learning network outputs a denoised image;

[0147] Step 5: After successful verification, the user is allowed to enter; otherwise, the abnormal situation transmitting module transmits abnormal situation information to the user;

[0148] Step 6: After the user confirms the abnormal situation information, the user sends the authorization information to the abnormal situation user confirmation module. After the abnormal situation user confirmation module receives the user's authorization information, the user privacy account management module allows access, otherwise, the user privacy account management module is locked.

[0149] Specific implementation method 2: Figures 1 to 7 As shown, the structure of the management system described in step 1 includes:

[0150] Data collection module, data analysis module, data storage module, OLAP analysis engine and OLTP read-write engine;

[0151] The data acquisition module is used to collect multimodal data and write the data into the data storage module;

[0152] The data analysis module is used to divide the collected data of the data collection module and write the divided data into the data storage module;

[0153] The data storage module is used to store the divided data;

[0154] The OLAP analysis engine is used to perform statistical queries on the stored data;

[0155] The OLTP read-write engine is used to call the stored data and perform addition, deletion, modification and query on the stored data in tuples;

[0156] The divided data includes relational data and file data;

[0157] The data storage module includes n column storage nodes, n row storage nodes and an LSM-Tree structure;

[0158] The n column storage nodes and the n row storage nodes are used to store relational data;

[0159] The LSM-Tree structure is used to store file-type data;

[0160] The LSM-Tree structure includes an LSM file system and a Blob file system.

[0161] Specific implementation method three: Figures 1 to 7 As shown, the management method of the management system described in step 1 is:

[0162] Step 101: Collect multimodal data based on the data acquisition module, and write the multimodal data into the data analysis module;

[0163] Step 102: dividing the collected data into relational data and file data based on the data analysis module;

[0164] Step 103: intelligently selecting row storage, column storage, or using both row storage and column storage at the same time according to user needs, and storing relational data based on n column storage nodes and n row storage nodes in the data storage module;

[0165] The steps of storing relational data based on n column storage nodes and n row storage nodes include:

[0166] Step 1031: Arrange the relational data and file data into a table format, assign a unique table ID in the cluster to each relational table, and assign a unique row / column ID in the table to each row of data in the relational data table;

[0167] Step 1032: Use table ID + row / column ID as the key of each tuple, and the content of the tuple as the corresponding value;

[0168] Step 1033: For the indexes of relational data tables and file-type data tables, use table ID + index ID + index value as the key of each row of index data, and the data row / column ID pointed to by the index as the value;

[0169] Step 1034: For the meta information of the relational data table and the file-type data table, a specific prefix + table ID is used as the key, and the structure definition of the relational data table is used as the value to store;

[0170] Step 1035: Use block storage technology to process relational data and file data in blocks, divide the entire key-value pair space into segments, each segment is a series of continuous keys, each segment is used as a block, and limit the data stored in each block to be less than a preset threshold;

[0171] Step 1036: Use a distributed read / write engine to store relational data and file data in n row / column storage nodes in units of blocks, and manage them in units of backup groups;

[0172] The steps to store file-type data based on the LSM-Tree structure include:

[0173] Step a: Assign a unique ID as the key to each file in the file-type data, and the file itself as the value;

[0174] Step b: At the flush and compaction time of the LSM-Tree, the value of the file memory that exceeds the preset threshold is separated from the LSM-tree and stored in the Blob file object;

[0175] Step c, the LSM file system uses SST as the basic unit for storage. Each key in SST is a file ID, and the value is a Blob object pointer of the file.

[0176] Step d: Each Blob object is stored using a block backup strategy. The size of each block is 100MB. The Blob file header block stores the file's meta information and a pointer to each block group of the file.

[0177] Step 104: Perform statistical query on the stored relational data and file-type data based on the OLAP analysis engine, and call the stored relational data and file-type data based on the OLTP read-write engine and perform addition, deletion, modification and query in tuple units.

[0178] The above is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this profession can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent replacement and improvement made to the above embodiments without departing from the content of the technical solution of the present invention, based on the technical essence of the present invention, within the spirit and principles of the present invention, still fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for protecting private accounts based on big data joint training, characterized in that: The specific steps include: Step 1: Establish a user account data management system; Step 2: Input the user's frequently used login location information and password input speed information into the user account data management system, and train the input information; Step 3: When the user logs into the account, the user account data management system compares the user's common login location information and password input speed information, and after the comparison is successful, displays the verification image; Step 4: Clarify the verification image; Step 5: After successful verification, the user is allowed to enter; otherwise, the abnormal situation transmitting module transmits abnormal situation information to the user; Step 6: After the user confirms the abnormal situation information, the user sends the authorization information to the abnormal situation user confirmation module. After the abnormal situation user confirmation module receives the user's authorization information, the user privacy account management module allows access, otherwise, the user privacy account management module is locked.

2. According to claim 1, a method for protecting privacy accounts based on big data joint training is characterized in that: The structure of the management system described in step 1 includes: Data collection module, data analysis module, data storage module, OLAP analysis engine and OLTP read-write engine; The data acquisition module is used to collect multimodal data and write the data into the data storage module; The data analysis module is used to divide the collected data of the data collection module and write the divided data into the data storage module; The data storage module is used to store the divided data; The OLAP analysis engine is used to perform statistical queries on the stored data; The OLTP read-write engine is used to call the stored data and perform addition, deletion, modification and query on the stored data in tuples; The divided data includes relational data and file data; The data storage module includes n column storage nodes, n row storage nodes and an LSM-Tree structure; The n column storage nodes and the n row storage nodes are used to store relational data; The LSM-Tree structure is used to store file-type data; The LSM-Tree structure includes an LSM file system and a Blob file system.

3. A method for protecting privacy accounts based on big data joint training according to claim 1 or 2, characterized in that: The management method of the management system in step 1 is: Step 101: Collect multimodal data based on the data acquisition module, and write the multimodal data into the data analysis module; Step 102: dividing the collected data into relational data and file data based on the data analysis module; Step 103: intelligently selecting row storage, column storage, or using both row storage and column storage at the same time according to user needs, and storing relational data based on n column storage nodes and n row storage nodes in the data storage module; Storing file-type data based on the LSM-Tree structure; Step 104: Perform statistical query on the stored relational data and file-type data based on the OLAP analysis engine, and call the stored relational data and file-type data based on the OLTP read-write engine and perform addition, deletion, modification and query in tuple units.

4. The method for protecting privacy accounts based on big data joint training according to claim 3 is characterized in that: The step of storing relational data based on n column storage nodes and n row storage nodes in step 103 includes: Step 1031: Arrange the relational data and file data into a table format, assign a unique table ID in the cluster to each relational table, and assign a unique row / column ID in the table to each row of data in the relational data table; Step 1032: Use table ID + row / column ID as the key of each tuple, and the content of the tuple as the corresponding value; Step 1033: For the indexes of relational data tables and file-type data tables, use table ID + index ID + index value as the key of each row of index data, and the data row / column ID pointed to by the index as the value; Step 1034: For the meta information of the relational data table and the file-type data table, a specific prefix + table ID is used as the key, and the structure definition of the relational data table is used as the value to store; Step 1035: Use block storage technology to process relational data and file data in blocks, divide the entire key-value pair space into segments, each segment is a series of continuous keys, each segment is used as a block, and limit the data stored in each block to be less than a preset threshold; Step 1036: Use a distributed read / write engine to store relational data and file data in n row / column storage nodes in blocks, and manage them in backup groups.

5. The method for protecting privacy accounts based on big data joint training according to claim 3, characterized in that: The steps of storing file-type data based on the LSM-Tree structure in step 103 include: Step a: Assign a unique ID as the key to each file in the file-type data, and the file itself as the value; Step b: At the flush and compaction time of the LSM-Tree, the value of the file memory exceeding the preset threshold is separated from the LSM-tree and stored in the Blob file object; Step c, the LSM file system uses SST as the basic unit for storage. Each key in SST is a file ID, and the value is a Blob object pointer of the file. Step d: Each Blob object is stored using a block backup strategy, with each block size being 100MB. The Blob file header block stores the file's meta information and a pointer to each block group of the file.

6. The method for protecting privacy accounts based on big data joint training according to claim 1, characterized in that: The step of clearing the verification image in step 4 includes: Step 401, point spread function analysis of the hypersurface unit; Step 402: image degradation, i.e., image restoration modeling and image preprocessing; analyzing the imaging results according to the point spread function modeling results of the hypersurface unit, and performing simple preprocessing on the acquired original image; Step 403, obtaining a denoiser; selecting a denoiser according to the optical system model and the image degradation model obtained in step 402; Step 404: construct a regularization term based on the denoiser obtained in step 403; use the output metric of the denoiser as part of the regularization by defining a penalty function related to the denoiser output; Step 405, constructing an objective function by combining the data fidelity term and the regularization term constructed in step 404; Step 406: using an optimization algorithm to solve the objective function constructed in step 405 to obtain a denoised hypersurface imaging result; iteratively updating an estimated value of the denoised image, and using a denoiser for processing in each iteration; Step 407: Use the image preprocessed in step 402 as the input of the deep learning network, use the denoised image as the output of the deep learning network, perform self-supervised training based on the penalty function in the objective function constructed in step 405 until convergence, and obtain a trained deep learning network; Step 408: input the original image to be tested into the trained deep learning network, and the trained deep learning network outputs the denoised image.

7. The method for protecting privacy accounts based on big data joint training according to claim 6, characterized in that: The specific process of step 401 is: When the optical system is linearly degraded, that is, the point spread function space is unchanged, without considering the noise, the imaging process is expressed by the following formula: In formula (1), I(x, y) represents the degraded image, S(u, v) represents the original image, and P(xu, yv) represents the spatially invariant point spread function; The image degradation process of the spatially varying point spread function with the field of view is expressed as: In formula (2), P(u, v, x--u, y--v) represents the spatially varying point spread function.

8. The method for protecting privacy accounts based on big data joint training according to claim 6, characterized in that: The specific process of step 402 is: The image degradation process can be regarded as a linear space-invariant degradation system, and its process can be described as: the clear image f(x, y) is convolved with the corresponding blur kernel h(x, y) to generate the degraded image g(x, y) under the influence of noise; g(x,y)=h(x,y)*f(x,y)+n(x,y) (3), In formula (3), * represents the convolution operator; When the point spread function is spatially invariant, the blurred image is represented as: The blurred image is represented as:

9. The method for protecting privacy accounts based on big data joint training according to claim 6, characterized in that: The regularization term expression in step 404 is: In formula (6), R(y) represents the regularization term, D σ (y) represents the denoiser, y represents the preprocessed image, and the superscript T represents the transposition; Step 5 The objective function is expressed as: In formula (7), x represents the target image, y represents the preprocessed image, l(y,x) represents the data fidelity term, that is, ensuring that y≈x, λ represents the degree of penalty for R(y), and the weight between the data fidelity term and the regularization term is adjusted according to actual needs to balance the relationship between the two. λR(y) represents the penalty function.

10. The method for protecting privacy accounts based on big data joint training according to claim 6, characterized in that: The specific process of step 406 is: Step 40601, construct the Lagrangian augmented function: Formula (8), f(x) represents the intermediate variable, It is defined as, p(y|x) represents the prior probability of finding x given y, g(x) represents the intermediate variable, p(x) represents the probability of x, f(x)+λg(v) is the original optimization target, and v represents the target image; under the condition of x=v, (x--v) is the equality constraint, u is the Lagrange multiplier associated with the constraint (x--v)=0, and ant represents the penalty coefficient; Step 40602, setting the maximum number of iterations K; Step 40603, number of iterations k=1; Step 40604: To solve the minimum problem, convert the Lagrangian augmented function into the following sub-problem u (k+1) =u (k) +(x (k+1) -v (k+1) ) (11), r k+1 =c k r k (12), In formulas (9), (10), (11) and (12), x (k+1) represents the target image of the k+1th iteration, x (k) represents the target image of the kth iteration, ρ k represents the penalty coefficient for the kth iteration, ρ k+1 represents the penalty coefficient for the k+1th iteration, v (k) represents the target image of the kth iteration, v (k+1) represents the target image of the k+1th iteration, u (k) represents the Lagrange multiplier associated with the constraint (xv) = 0 at the kth iteration, u (k+1) represents the Lagrange multiplier associated with the constraint (xv) = 0 at the k+1th iteration, D σk represents the denoiser at the kth iteration, γ k Represents ρ k The adjustment parameter, D σk is the denoising algorithm, is a parameter that controls the strength of the denoiser, and improves ant to ρ k+1 =γ k ρ k , γ k ≥1 is the tuning parameter of Ant. Step 40605, let the number of iterations k = k + 1, repeat step 405 until the maximum number of iterations K is reached. The obtained xK is the target image x in the objective function, satisfying corresponding; The specific process of step 407 is: The image preprocessed in step 402 is used as the input of the deep learning network, the target image x obtained in step 406 is used as the output of the deep learning network, and self-supervised training is performed based on the penalty function in the objective function constructed in step 405 until convergence to obtain a trained deep learning network.

Images