Private data security intelligent transmission method based on binary sequence segmentation original network

Abstract

The invention relates to the technical field of data security, and discloses a privacy data security intelligent transmission method based on a binary sequence segmentation original network, comprising the following steps: S1, performing binary sequence segmentation on a deep neural network in a spatial dimension; s2, performing feature transmission; s3, bullet fish densification; and S4, performing secure storage. According to the secure and intelligent private data transmission method for segmenting the original network based on the binary sequence, the original network is segmented through the binary sequence, the client only transmits the feature vector instead of the original private data, and the sensitive information exposure risk is reduced from the source; the bullet fish algorithm enables transmission data to exist in a ciphertext form, and the possibility of reverse decoding is minimized in cooperation with antagonistic reconstruction learning, so that triple privacy barriers of data segmentation, dynamic encryption and model protection are formed; hash verification before cloud storage ensures data integrity, tampering and leakage are avoided, violent attacks and reverse engineering are effectively resisted, and the privacy compliance requirement of a high-sensitivity scene is met.

Description

Technical Field

[0001] The present invention relates to the field of data security technology, and in particular to a method for secure intelligent transmission of private data based on binary sequence segmentation of an original network. Background Art

[0002] In distributed deep learning scenarios, segmentation learning (SL) avoids sharing original data by collaboratively training models between the client and the server. However, it has two major drawbacks: first, the client needs to bear a high computing load, and second, the transmission of intermediate features still poses a risk of privacy leakage. Existing encryption technologies, such as fully homomorphic encryption, can protect data, but the computing and transmission overhead is too high, making it difficult to adapt to resource-constrained consumer electronics such as smartphones and wearable devices. Therefore, a secure and intelligent transmission method for private data based on binary sequential segmentation of the original network is proposed. Summary of the Invention

[0003] (1) Technical problems solved

[0004] In response to the shortcomings of the existing technology, the present invention provides a method for secure and intelligent transmission of private data based on binary sequence segmentation of the original network. It has the advantages of reducing computing resource consumption while protecting private data through a multi-layer security mechanism, and solves the two major defects of the existing technology: one is that the client needs to bear a high computing load, and the other is that there is still a risk of privacy leakage in the transmission of intermediate features; existing encryption technologies, such as fully homomorphic encryption, can protect data, but the computing and transmission overhead is too large, making it difficult to adapt to resource-constrained consumer electronic products such as smartphones and wearable devices.

[0005] (2) Technical solution

[0006] To achieve the aforementioned goal of reducing computing resource consumption while protecting private data through a multi-layered security mechanism, the present invention provides the following technical solution: a method for securely and intelligently transmitting private data based on binary sequence segmentation of the original network, comprising the following steps:

[0007] S1. Perform binary sequential segmentation on the deep neural network in the spatial dimension, determine the segmentation boundary by calculating the vector average magnitude absolute difference d(A, B) and vector sum angle θ(A, B) of sets A and B, and form a client sub-network and a server sub-network.

[0008] S2. Feature transmission: The client performs cyclic tiling processing on the private data, extracts the feature vector, and then encrypts and transmits it to the server.

[0009] S3, Bulletfish encryption: The server uses the Bulletfish algorithm to encrypt the feature vector through prime number selection, key generation and hash function combination. The ciphertext formula is:

[0010]

[0011] Among them, Y(Q n ) indicates DPQ n The encrypted text of

[0012] S4. Secure storage: The server stores the encrypted data in the cloud and performs integrity verification through hash value comparison.

[0013] Preferably, the binary sequence segmentation satisfies the assertion condition:

[0014] θ(F,R)>θ(R1,R2)

[0015] d(F,R)>d(R1,R2)

[0016] Among them, F is the original feature set, and R, R1, and R2 are the feature subsets after segmentation.

[0017] Preferably, the output segmentation scheme of the client sub-network is n Input partitioning scheme I with server terminal network (n+1) Aligned in the spatial dimension, satisfying:

[0018] lc(I i )=O i s―q b

[0019] uc(I i )=(O i ―1)s+k―q b

[0020] Where lc / uc is the input index boundary, q b is the starting fill value.

[0021] Preferably, the BulletFish algorithm includes:

[0022] Select prime numbers z and y and calculate K = z × x;

[0023] Then, an integer g is selected that satisfies the conditions that it is greater than 1 and less than both (z - 1) and (x - 1);

[0024] Calculate the auxiliary key g s =g ―1 (z―1)(x―1);

[0025] A random string ρ is introduced to generate the encryption seed.

[0026] Preferably, the private data includes musculoskeletal X-ray images, and the training set, validation set, and test set are divided in a ratio of 8:1:1.

[0027] Preferably, during the server-side training, negative pixel reconstruction loss and negative feature reconstruction loss are minimized to prevent the feature vector from being reversely decoded into the original data.

[0028] Preferably, the client is a consumer electronic product with a memory capacity of 1-4GB, and the server is a cloud server or an edge server.

[0029] A secure and intelligent transmission system for private data based on binary sequential segmentation of the original network, including:

[0030] Client device, integrating binary sequence segmentation module and BulletFish encryption module;

[0031] The server-side device is configured with a segmentation scheme alignment module and a security storage module;

[0032] The communication module uses the TLS1.3 protocol to transmit encrypted feature vectors.

[0033] (3) Beneficial effects

[0034] Compared with the existing technology, the present invention provides a method for secure and intelligent transmission of private data based on binary sequence segmentation of the original network, which has the following beneficial effects:

[0035] 1. This method for secure and intelligent transmission of private data based on binary sequential segmentation of the original network splits the original network in binary order. By doing so, the client only transmits feature vectors instead of the original private data, reducing the risk of sensitive information exposure from the source. The BulletFish algorithm uses prime number keys and random hash encryption to make the transmitted data exist in ciphertext form, and cooperates with adversarial reconstruction learning to minimize the possibility of reverse decoding, forming a triple privacy barrier of "data segmentation-dynamic encryption-model protection". Hash verification before cloud storage ensures data integrity and avoids tampering and leakage, realizing full-link privacy protection from data extraction, transmission to storage, effectively resisting brute force attacks and reverse engineering, and meeting the privacy compliance requirements of highly sensitive scenarios.

[0036] 2. This method for secure and intelligent transmission of private data based on binary sequential segmentation of the original network uses loop tiling technology to decompose client computing tasks into cache-friendly sub-blocks, reducing data loss from the cache and increasing memory utilization by more than 30%, making it suitable for consumer electronics products with 1-4GB of memory. The segmentation scheme alignment design eliminates inter-layer data reorganization overhead and combines with the TLS1.3 protocol to achieve low-latency transmission within 200ms. The distributed architecture offloads intensive computing to the server side, and the client is only responsible for lightweight feature extraction, which improves overall computing efficiency by 40%. This not only reduces the power consumption of smartphones and wearable devices, but also ensures model training speed through the powerful computing power of cloud / edge servers, balancing device performance and computing needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Schematic diagram of the method for secure and intelligent transmission of private data of the present invention;

[0038] Figure 2 Schematic diagram of analog output of the present invention;

[0039] Figure 3 Schematic diagram of the number of iterations and loss curve of the present invention;

[0040] Figure 4 Schematic diagram of the accuracy and loss curve of the present invention;

[0041] Figure 5 This is a bar diagram of the accuracy evaluation of the present invention;

[0042] Figure 6 This is a schematic diagram of error analysis comparison of the present invention;

[0043] Figure 7 This is a safety level analysis and comparison diagram of the present invention. DETAILED DESCRIPTION

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the embodiments of the present invention and the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0045] See also Figure 1-7 , a secure and intelligent transmission method for private data based on binary sequential segmentation of the original network, comprising the following steps:

[0046] S1. Perform binary sequential segmentation on the deep neural network in the spatial dimension, determine the segmentation boundary by calculating the vector average magnitude absolute difference d(A, B) and vector sum angle θ(A, B) of sets A and B, and form a client sub-network and a server sub-network.

[0047] S2. Feature transmission: The client performs cyclic tiling processing on the private data, extracts the feature vector, and then encrypts and transmits it to the server.

[0048] S3, Bulletfish encryption: The server uses the Bulletfish algorithm to encrypt the feature vector through prime number selection, key generation and hash function combination. The ciphertext formula is:

[0049]

[0050] Among them, Y(Q n ) indicates DPQ n The encrypted text of

[0051] S4. Secure storage: The server stores the encrypted data in the cloud and performs integrity verification through hash value comparison.

[0052] Example 1:

[0053] This embodiment specifically implements the binary sequence segmentation and feature transmission process of deep neural network.

[0054] First, the deep neural network is divided into binary order in the spatial dimension, and the vector average absolute difference of sets A and B is calculated. and vectors and angles Determine the segmentation boundary, where A client sub-network and a server sub-network are formed.

[0055] The client performs loop tiling processing on musculoskeletal X-ray privacy data, breaking down large calculations into cache-friendly sub-blocks, processing one small block at a time to improve memory utilization. After extracting the feature vector, it uses the Bulletfish algorithm for encryption. The encryption process includes selecting prime numbers z and y to calculate K = z × x, selecting an integer g that satisfies the requirement of being greater than 1 and less than (z-1) and (x-1), and calculating the auxiliary key g. s =g ―1 (z―1)(x―1), introduce a random string ρ to generate the encryption seed, and generate the ciphertext through the hash function F:

[0056]

[0057] After encryption is completed, the feature vector is transmitted to the server. After receiving it, the server performs hash value comparison to complete the integrity check, and finally stores the encrypted data in the cloud. The training set, validation set, and test set are divided into 8:1:1.

[0058] Example 2:

[0059] This embodiment focuses on segmentation assertion conditions and solution alignment requirements.

[0060] Binary sequence segmentation must meet the assertion conditions:

[0061] θ(F,R)>θ(R1,R2)

[0062] d(F,R)>d(R1,R2)

[0063] Among them, F is the original feature set, R, R1, R2 are the feature subsets after segmentation; by comparing the vectors, angles and average amplitude absolute differences of the feature subsets before and after segmentation, it is ensured that the differences in the feature subsets after segmentation meet the requirements to ensure the rationality of network segmentation.

[0064] At the same time, the output segmentation scheme of the client sub-network On Input partitioning scheme I with server terminal network (n+1) Need to be aligned in the spatial dimension, satisfying:

[0065] lc(I i )=O i s―q b

[0066] uc(I i )=(O i ―1)s+k―q b

[0067] Where lc / uc is the input index boundary, q b is the starting filling value; this alignment design eliminates the overhead of data reorganization between layers, so that each part of the input can be processed independently within the layer without the need for intermediate connections or split operations, thereby improving network processing efficiency. In each training iteration, the server guides the model L by optimizing the error rate of the discriminator. f =log(1―D(f(X priv ))), and the loss function is invariant to fluctuations in the training labels.

[0068] Example 3:

[0069] This embodiment further expands on the details of the BulletFish algorithm.

[0070] First, select prime numbers z and y, for example, z = 3, y = 5, and calculate K = z × x = 15;

[0071] Then select an integer g that satisfies the conditions of being greater than 1 and less than both (z - 1) and (x - 1), here g = 2;

[0072] Calculate the auxiliary key g s =g ―1 (z-1)(x-1)=2 ―1 ×2×4=4;

[0073] Introduce a random string ρ to generate an encryption seed, use traditional encryption methods to encode the seed, use the starting key and key to generate the encryption seed and input Q n The corresponding ciphertext, the encryption seed is represented by "signifies", the random string is represented by the symbol "ρ", and the cryptographic hash function is represented by "F", through:

[0074]

[0075] Generate ciphertext, where Y(Q n ) indicates DPQ nDuring the encryption process, the authorized party needs to obtain the random string, encryption seed, initial key and secret key to safely restore the original data. The above steps are used to encrypt the feature vector and ensure the security of data transmission and storage.

[0076] Example 4:

[0077] This embodiment further expands on the system architecture and training requirements.

[0078] The client is a consumer electronic product with a memory capacity of 1-4GB, such as a smartphone or wearable device. It integrates a binary sequence segmentation module and a BulletFish encryption module, responsible for segmenting and encrypting private data for transmission.

[0079] The server side is a cloud server or edge server, equipped with a segmentation scheme alignment module and a secure storage module. During training, it minimizes the negative pixel reconstruction loss and negative feature reconstruction loss to prevent the feature vector from being reversely decoded into the original data. The communication module uses the TLS1.3 protocol to transmit encrypted feature vectors to ensure the security and low latency of the transmission process (within 200ms).

[0080] The client is only responsible for lightweight feature extraction, offloading intensive calculations to the server, improving overall computing efficiency by 40%. At the same time, it improves memory utilization by more than 30% through loop tiling technology, adapting to the resource limitations of consumer electronics products, and achieving full-link privacy protection from data extraction, transmission to storage, effectively resisting brute force attacks and reverse engineering, and meeting the privacy compliance requirements of highly sensitive scenarios.

[0081] Embodiment 5:

[0082] This example quantitatively verifies the accuracy, privacy protection capability, and computational efficiency of a method for secure and intelligent transmission of private data based on binary sequential segmentation of the original network. The specific steps are as follows:

[0083] The experiment is based on the PyTorch framework and uses the ResNet-20 model to classify musculoskeletal X-ray images (7 parts such as elbows and fingers). The dataset is divided into training, validation, and test sets at an 8:1:1 ratio. The client retains the first three convolutional layers as cutting layers, and the server processes the remaining network layers. The input data is decomposed into cache-friendly sub-blocks through loop tiling technology, which improves client memory utilization by more than 30%. The simulation output is as follows: Figure 2 As shown;

[0084] To find the total loss incurred during a specific time period, the loss function analyzes the entire dataset; a numerical description of the evaluation is provided; data loss is an inevitable consequence of the iterative curve; compared to the resource utilization of other methods, the observed curve is mainly determined by the time required to train and test the classifier; due to the significant difference between training loss and validation loss, the proposed model may be underfitting; increasing the sample size has the potential to reduce training loss; (either in terms of the total number of layers or the specific number of neurons in each layer, according to a more qualitative approach).

[0085] Figure 3 The dataset used to calculate validation loss is shown; the validation loss metric is a way to measure how well a model performs on a validation dataset for consumer electronics. To evaluate the model's performance, a subset of the dataset was selected and labeled as the validation set. The test loss is obtained when the total number of false positives in the training and validation sets is added together. In this case, the method shows that the loss can be significantly reduced.

[0086] Figure 4 The empirical data shown shows a discrepancy between validation and training accuracy, which may be due to the model being more familiar with the training data due to previous exposure to it; in contrast, the validation data contains data points that the model has not seen before; as shown in Figure 4 As shown, the proposed classifier can accurately identify views; the dataset is divided into multiple network layers, and then the accuracy scores are estimated based on the information in Table I.

[0087] Several performance measures were used to evaluate the effectiveness of the presented techniques; accuracy, privacy, and efficiency were the three main criteria used to evaluate model performance; seeing how well an attacker can piece together the input is one way to find out how private something is; metrics frequently used in this field include structural similarity (SSIM), peak signal-to-noise ratio (PSNR), encryption and decryption time, and F-SIM; the F-SIM method uses a third-party network that takes the original and reconstructed inputs as variables; next, properties are compared by computing cosine similarity.

[0088] Table 1. Performance evaluation of different segmentation learning strategies

[0089]

[0090]

[0091] Table 1 Figure 5 and Figure 6 The results show that the proposed method outperforms existing methods in reducing error rate and improving accuracy.

[0092] Table 2. Computational cost of different encryption algorithms

[0093]

[0094] Table 2 presents a visual and chronological presentation of various file sizes measured in megabytes (Mb), including encryption, decryption, and execution operations. The table shows the time required to encrypt and decode the data. The proposed method was extensively tested, and its results were compared with popular algorithms such as Bulletfish, RSA, DES, and AES. The proposed Bulletfish method reduces the time required for data encryption, decryption, and execution.

[0095] Table 3, Safety Analysis Table

[0096]

[0097] Table 3 and Figure 7 A comprehensive analysis and comparison of the security levels provided by different encoding schemes are presented separately; for a 20MB file, the most secure encryption algorithms are DES (78% success rate), RSA (77.23% success rate), and AES (73% success rate); a comprehensive review of security measures is also conducted for data ranging from 40 to 100 gigabytes; as shown in the above figure, the proposed system provides a higher degree of security than current encryption solutions.

[0098] The data showed that the proposed approach was sufficiently successful in protecting the privacy and security of patients’ medical records.

[0099] Compared with traditional methods, the proposed solution has significant improvements in accuracy, privacy protection, and computational efficiency, especially in consumer electronics and mobile medical scenarios, showing the advantages of lightweight and high security.

[0100] In summary, this method for secure and intelligent transmission of private data based on binary sequential segmentation of the original network divides the original network into binary sequences. By dividing the original network into binary sequences, the client only transmits feature vectors instead of original private data, reducing the risk of sensitive information exposure from the source. The BulletFish algorithm uses prime number keys and random hash encryption to make the transmitted data exist in ciphertext form, and cooperates with adversarial reconstruction learning to minimize the possibility of reverse decoding, forming a triple privacy barrier of "data segmentation-dynamic encryption-model protection". The hash check before cloud storage ensures data integrity and avoids tampering and leakage, realizing full-link privacy protection from data extraction, transmission to storage, effectively resisting brute force attacks and reverse engineering, and meeting the privacy compliance requirements of highly sensitive scenarios.

[0101] In addition, this method for secure and intelligent transmission of private data based on binary sequential segmentation of the original network uses cyclic tiling technology to decompose client computing tasks into cache-friendly sub-blocks, reducing data loss from the cache and improving memory utilization by more than 30%, making it suitable for consumer electronics products with 1-4GB of memory; the segmentation scheme alignment design eliminates the overhead of inter-layer data reorganization, and combined with the TLS1.3 protocol to achieve low-latency transmission within 200ms; the distributed architecture offloads intensive computing to the server side, and the client is only responsible for lightweight feature extraction, which improves overall computing efficiency by 40%. It not only reduces the power consumption of smartphones and wearable devices, but also ensures the model training speed through the powerful computing power of cloud / edge servers, balancing device performance and computing requirements, and solving two major defects of existing technologies: one is that the client needs to bear a high computing load, and the other is that the intermediate feature transmission still has the risk of privacy leakage; existing encryption technologies, such as fully homomorphic encryption, can protect data, but the computing and transmission overhead is too high, making it difficult to adapt to resource-constrained consumer electronics products such as smartphones and wearable devices.

[0102] The relevant modules involved in this system are all hardware system modules or functional modules that combine computer software programs or protocols with hardware in the existing technology. The computer software programs or protocols involved in the functional modules are themselves technologies that are well known to those skilled in the art and are not improvements to this system. The improvements to this system are the interaction or connection relationships between the modules, that is, improvements to the overall structure of the system to solve the corresponding technical problems to be solved by this system.

[0103] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A secure and intelligent transmission method for private data based on binary sequence segmentation of the original network, characterized by: The following steps are involved: S1. Perform binary sequential segmentation on the deep neural network in the spatial dimension, determine the segmentation boundary by calculating the vector average magnitude absolute difference d(A, B) and vector sum angle θ(A, B) of sets A and B, and form a client sub-network and a server sub-network. S2. Feature transmission: The client performs cyclic tiling processing on the private data, extracts the feature vector, and then encrypts and transmits it to the server. S3, Bulletfish encryption: The server uses the Bulletfish algorithm to encrypt the feature vector through prime number selection, key generation and hash function combination. The ciphertext formula is: Among them, Y(Q n ) indicates DPQ n The encrypted text of S4. Secure storage: The server stores the encrypted data in the cloud and performs integrity verification through hash value comparison.

2. The method for secure and intelligent transmission of private data based on binary sequence segmentation of the original network according to claim 1 is characterized in that: The binary sequence segmentation satisfies the assertion condition: θ(F,R)>θ(R1,R2) d(F,R)>d(R1,R2) Among them, F is the original feature set, and R, R1, and R2 are the feature subsets after segmentation.

3. The method for secure and intelligent transmission of private data based on binary sequence segmentation of the original network according to claim 1 is characterized in that: The output partitioning scheme O of the client sub-network n Input partitioning scheme I with server terminal network (n+1) Aligned in the spatial dimension, satisfying: lc(I i )=O i s―q b uc(I i )=(O i ―1)s+k―q b Where lc / uc is the input index boundary, q b is the starting fill value.

4. The method for secure and intelligent transmission of private data based on binary sequence segmentation of the original network according to claim 1 is characterized in that: The BulletFish algorithm includes: Select prime numbers z and y and calculate K = z × x; Then, an integer g is selected that satisfies the conditions that it is greater than 1 and less than both (z - 1) and (x - 1); Calculate the auxiliary key g s =g ―1 (z―1)(x―1); A random string ρ is introduced to generate the encryption seed.

5. The method for secure and intelligent transmission of private data based on binary sequence segmentation of the original network according to claim 1 is characterized in that: The private data includes musculoskeletal X-ray images, and the training set, validation set, and test set are divided into 8:1:

1.

6. The method for secure and intelligent transmission of private data based on binary sequence segmentation of the original network according to claim 1 is characterized in that: During the server-side training, negative pixel reconstruction loss and negative feature reconstruction loss are minimized to prevent the feature vector from being reversely decoded into the original data.

7. The method for secure and intelligent transmission of private data based on binary sequence segmentation of the original network according to claim 1 is characterized in that: The client is a consumer electronic product with a memory capacity of 1-4GB, and the server is a cloud server or an edge server.

8. A secure intelligent transmission system for private data based on binary sequence segmentation of the original network, characterized by: include: Client device, integrating binary sequence segmentation module and BulletFish encryption module; The server-side device is configured with a segmentation scheme alignment module and a security storage module; The communication module uses the TLS1.3 protocol to transmit encrypted feature vectors.

Images