Digital fingerprint generation method for terminal equipment

Abstract

The invention discloses a digital fingerprint generation method for terminal equipment, and relates to the technical field of digital fingerprint generation, comprising the following steps: S1, collecting multi-dimensional hierarchical features; s2, privacy enhancement; s3, fingerprint anti-regeneration generation is carried out; and S4, fingerprint life cycle management. According to the terminal equipment digital fingerprint generation method, multi-dimensional fine-grained feature acquisition is realized from a non-sensitive layer, a semi-sensitive layer and a high-sensitive layer, a feature weight is optimized through a federated learning framework, fingerprint drift caused by hardware replacement and software updating is effectively resisted, and the user experience is improved. Dynamic Gaussian noise is injected into numerical semi-sensitive features to balance privacy protection and feature effectiveness, 2048-bit key Paillier homomorphic encryption is adopted for high-sensitive features, meanwhile, cross-scene tracking risks are blocked through scenarized sub-fingerprint generation and an original feature immediate destruction mechanism, and the privacy protection and feature effectiveness are improved. And an anti-duplication architecture fusing a Merkle tree and a bloom filter is introduced to greatly reduce the fingerprint coincidence collision rate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital fingerprint generation, in particular to a terminal device digital fingerprint generation method. BACKGROUND

[0002] In the field of terminal device identity recognition, the existing digital fingerprint generation process generally collects feature data from the dimensions of device hardware, software environment and network state, such as CPU model, operating system version, MAC address and other information, and integrates these features to generate a unique digital fingerprint through a hash algorithm or other data processing method for device identity recognition and security verification.

[0003] In the existing digital fingerprint generation process of terminal devices, the feature collection dimensions are limited and lack dynamic adjustment: existing methods rely on fixed dimension features, generally only collecting browser UA or hardware ID, resulting in serious fingerprint drift after hardware replacement and system update, thereby reducing the accuracy of identity recognition. SUMMARY

[0004] To overcome the deficiencies of the prior art, the present application provides a terminal device digital fingerprint generation method, which solves the problems raised in the background art.

[0005] To achieve the above purpose, the present application realizes the following technical scheme: a terminal device digital fingerprint generation method, comprising the following steps: S1, multi-dimensional hierarchical feature collection: Establish a three-level feature collection layer of non-sensitive, semi-sensitive and high-sensitive, non-sensitive features including screen physical resolution, operating system kernel compilation time; semi-sensitive features including CPU core number, total memory capacity; high-sensitive features including a spliced string composed of CPU model, motherboard model and baseband version; S2, privacy enhancement: In the terminal hardware security area, inject Gaussian noise to numerical semi-sensitive features, use 128-bit dynamic salt value and SHA-3-512 algorithm for salted hash of semi-sensitive features, and use 2048-bit key Paillier homomorphic encryption for high-sensitive features; S3, fingerprint anti-repetition generation: Construct a distributed training system, use a random forest model to train feature weights locally on the terminal, with a learning rate , and use FedAvg algorithm to aggregate the model every 7 days on the server, fuse features through Merkle tree and Bloom filter, and generate anti-repetition digital fingerprints; S4, fingerprint life cycle management: The main fingerprint and the scene identifier are used to generate a sub-fingerprint through HMAC-SHA256, the main fingerprint is automatically updated every 30 days, and the original feature data is destroyed in real time.

[0006] Further, in the step S1, the screen physical resolution in the non-sensitive feature is accurate to 1 pixel, and the operating system kernel compilation time is accurate to seconds. The total memory capacity in the semi-sensitive feature is accurate to 0.1 GB.

[0007] Further, in the step S1, the sampling frequency of the non-sensitive feature is once per session, the sampling frequency of the semi-sensitive feature is 24 hours once and needs user authorization, and the high-sensitive feature is only collected when the device is first or reset.

[0008] Further, in the step S2, the Gaussian noise , is 1.5 or 0.8, is the feature sensitivity; The 128-bit dynamic salt value is updated every 24 hours.

[0009] Further, in the step S3, the Merkle tree: The Merkle tree is used to verify the integrity and consistency of the data.

[0010] Further, in the step S3, the false positive rate p of the Bloom filter is 0.0001, and the number of optimized hash functions and the bit vector length are used to realize the lightweight storage and fast retrieval of the feature set; wherein, is the number of inserted elements, is the number of filter bits.

[0011] Further, in the step S4, the main fingerprint is the anti-duplicate digital fingerprint generated in the step S3, and the scene identifier is a group of feature information used for marking and distinguishing different application environments, business operation scenarios or device use scenarios in the main fingerprint regeneration process.

[0012] Further, the feature information includes but is not limited to time information, geographic location information, business type, device use mode, and through the scene identifier, the generation of the digital fingerprint is more suitable for the specific use scenario requirements.

[0013] Further, the terminal device digital fingerprint generation method is applied to the technical field of terminal device anti-duplicate digital fingerprint generation.

[0014] The present application provides a terminal device digital fingerprint generation method, which has the following beneficial effects: 1. The terminal device digital fingerprint generation method, which sets up multi-dimensional three-level feature collection to realize multi-dimensional fine-grained feature collection from non-sensitive, semi-sensitive and high-sensitive layers, ensures that the features cover the device hardware core attributes and dynamic environment parameters, optimizes the feature weights through a federated learning framework, effectively resists fingerprint drift caused by hardware replacement and software update, injects dynamic Gaussian noise into numerical semi-sensitive features to balance privacy protection and feature effectiveness, uses 2048-bit key Paillier homomorphic encryption for high-sensitive features, supports feature fusion calculation in an encrypted state, simultaneously blocks cross-scene tracking risks through a scene-based sub-fingerprint generation and original feature instant destruction mechanism, and introduces a Merkle tree and Bloom filter combined anti-duplication architecture to automatically add random features for recalculation of overlapping digital fingerprints, thereby greatly reducing the fingerprint overlap collision rate. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 It is a step flowchart of the terminal device digital fingerprint generation method of the application. Figure 2 It is a comparison chart of the terminal device digital fingerprint generation method based on the embodiment and the prior art effect of the application. DETAILED DESCRIPTION

[0016] The embodiments of the application will be further described in detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the application, but cannot be used to limit the scope of the application.

[0017] As shown in the drawings, Figures 1-2 The application provides a technical solution: a terminal device digital fingerprint generation method, which comprises the following steps: S1, multi-dimensional hierarchical feature collection: A non-sensitive, semi-sensitive and high-sensitive three-level feature collection layer is established, the non-sensitive features include screen physical resolution and operating system kernel compilation time, the semi-sensitive features include CPU core number and total memory capacity, and the high-sensitive features include a spliced string composed of CPU model, motherboard model and baseband version; The screen physical resolution in the non-sensitive features is accurate to 1 pixel, and the operating system kernel compilation time is accurate to seconds; the total memory capacity in the semi-sensitive features is accurate to 0.1 GB; The sampling frequency of the non-sensitive features is once per session, the sampling frequency of the semi-sensitive features is 24 hours once and needs user authorization, and the high-sensitive features are collected only when the device is first used or reset; Non-sensitive features are used for basic scenarios with high real-time requirements and no privacy leakage risk, and can be used for application interface adaptive adaptation scenarios, for example, a video platform or e-commerce APP automatically adjusts the interface layout, font size or video playback clarity when the user opens the session each time, ensuring that the display effect adapts to the device hardware and improves the user experience. Semi-sensitive features are used for scenarios that require certain device performance information but need to protect user core privacy, and can be used for application performance optimization recommendation scenarios, for example, a game platform collects CPU core number once a day after obtaining user authorization, analyzes the computing power of the device in combination with other performance characteristics, and recommends game quality settings that adapt to the CPU performance of the user. Highly sensitive features are used for device unique identity, tamper-proofing or high security level verification scenarios, and can be used for device tracing or anti-theft verification scenarios, for example, operators or device manufacturers collect the feature when the device is activated for the first time and store it encrypted, when the device is lost or stolen, the unique identity of the device is confirmed through feature comparison in an encrypted state, assisting the police or the user in locating the device; or in the first registration scenario of a financial APP, the encrypted feature is used to verify the authenticity of the device hardware, prevent fake devices or emulators from logging into the account, and improve the security level of the account. S2, privacy enhancement: In the terminal hardware security area, Gaussian noise is injected into numerical semi-sensitive features, a 128-bit dynamic salt value is used for salted hash of semi-sensitive features using SHA-3-512 algorithm, and a 2048-bit key is used for Paillier homomorphic encryption of highly sensitive features; Gaussian noise , 1.5 or 0.8, feature sensitivity, memory capacity =0.1GB, delay =5ms; The 128-bit dynamic salt value is updated every 24 hours as follows: Gaussian noise is injected into numerical semi-sensitive features such as memory capacity and delay, and the noise standard deviation The calculation formula is: , where is the privacy budget, the ordinary scenario =1.5, high security scenario =0.8, ordinary scenarios such as daily life and office scenarios, after noise injection, 1 decimal place accuracy is retained, such as the original memory 12.0GB processed as 12.1GB, the error is within the allowable range; A 128-bit dynamic salt value is generated based on the random number generator TRNG of the device security chip. It is automatically updated every 24 hours. The salt value is stored only in the hardware security area and adopts the SHA-3-512 algorithm, which is the 512-bit version of Secure Hash Algorithm 3 (SHA-3). The feature and the salt value are concatenated to calculate the hash, such as CPU model + salt value → hash value. The original feature is destroyed immediately. S3, fingerprint anti-duplicate generation: A distributed training system is constructed, and the feature weights are trained locally on the terminal using a random forest model with a learning rate of [missing information]. The server uses the FedAvg algorithm to aggregate models every 7 days, and fuses features through Merkle trees and Bloom filters to generate anti-duplicate digital fingerprints; Merkle tree: Merkle trees are used to verify data integrity and consistency. Bloom filters have a false positive rate of p=0.0001, achieved by optimizing the number of hash functions. By considering the bit vector length, lightweight storage and fast retrieval of feature sets can be achieved; among them, The number of elements to be inserted. The filter bit length is the core data structure of the Bloom filter, where the bit vector length is [value]. Essentially, it's a continuous binary bit sequence composed of 0s and 1s, used to indicate whether an element might exist in the set. express; Specifically, each terminal device acts as an independent federated learning node, deploying a random forest model as the core training framework. This model effectively reduces model variance and improves the generalization ability to terminal device features by constructing multiple decision tree sets and using the Bagging strategy for sample sampling. The training rounds are deeply coupled with the feature dimensions, meaning that the number of training rounds is strictly equal to the number of feature dimensions. For example, when the feature vector of a terminal device is 10-dimensional, the training will be performed in 10 rounds, with the learning rate fixed at α=0.01. This approach achieves the best performance in balancing training speed and convergence accuracy, ensuring that the model approaches the optimal solution with a stable step size in each iteration. During training, the random forest model dynamically calculates the importance score of each feature and generates a feature weight vector using metrics such as Gini index or information gain, providing a quantitative basis for subsequent digital fingerprint generation. The federated server employs a periodic aggregation strategy with an aggregation interval of 7 days. It uses the FedAvg federated averaging algorithm to allocate aggregation weights based on the proportion of devices on each node. Specifically, nodes with a higher proportion of devices have greater weight in the global aggregation of their uploaded model parameters, ensuring that the final global model can fully reflect the overall feature distribution of a large number of devices. To alleviate network transmission pressure, the global model parameters are compressed with a compression rate of 50%. Lightweight compression algorithms such as sparsity processing and quantization encoding are used to significantly reduce the amount of model transmission without losing key feature information, thereby improving the overall efficiency of the federated learning system. For highly stable features such as hardware combined hashes, a tiered weight adjustment strategy is adopted: the initial weight is set to 0.4 to ensure the basic weight ratio of the device's core identifier. The system will perform a verification every 24 hours. If the feature value does not drift during three consecutive verifications, that is, the deviation value is less than the set threshold of 0.01, the weight will be increased by 5%. When the weight reaches the upper limit of 0.6, it enters a stable state and no longer increases. This mechanism enhances the anti-interference ability of digital fingerprints by gradually strengthening the weight of stable features. For dynamic features such as browser version, dynamic responsive weight management is adopted: the initial weight is 0.2, which provides flexible environmental parameter reference in device fingerprint calculation. Once a change in feature value is detected, the system will immediately trigger a temporary weight adjustment mechanism to reduce its weight to 0.1 in order to reduce the impact of temporary environmental changes on fingerprint stability. When the data is aggregated and calculated again, the weight will automatically be restored to the initial value of 0.2 to balance the relationship between dynamic changes and fingerprint stability. Merkle tree construction: A bottom-up construction strategy is adopted, using the preprocessed device feature hash values ​​as leaf nodes of the Merkle tree. The number of nodes strictly corresponds to the number of device features, and is constructed using a formula tree. The depth of the dynamically calculated tree ensures that the generated Merkle tree has good balance. This structure can not only quickly verify the integrity of feature data, but also achieve efficient comparison of device fingerprints through the root hash value, effectively reducing the risk of data tampering. Bloom filter: To meet the fingerprint storage requirements of large-scale devices, a false positive rate p=0.0001 is set to ensure high accuracy of fingerprint recognition; Bloom filter capacity... It can support fingerprint data storage for 100 million devices; according to the formula The number of hash functions is calculated, and by scientifically configuring the parameters, the probability of false positives is reduced while ensuring storage space efficiency. When the Bloom filter detects a duplicate hash value, the intelligent collision handling process is triggered. At this time, two random features such as real-time battery temperature, current running process quantity, etc. are automatically collected, and the device fingerprint hash value is recalculated. This process uses recursive verification until a unique and conflict-free fingerprint data is generated, ensuring the uniqueness and accuracy of each device fingerprint. S4, fingerprint life cycle management: Based on the main fingerprint and the scene identifier, a sub-fingerprint is generated by HMAC-SHA256. The main fingerprint is automatically updated every 30 days, and the original feature data is destroyed in real time. The key is managed by the device security chip. The main fingerprint is a duplicate-resistant digital fingerprint generated in step S3. The scene identifier is a set of feature information used to mark and distinguish different application environments, business operation scenarios, or device usage scenarios during the generation of the main fingerprint. Feature information includes but is not limited to time information, geographic location information, business type, and device usage mode. Through the scene identifier, the generation of the digital fingerprint is more suitable for specific usage scenario requirements. The main fingerprint generates a 256-bit hash value through the SHA-256 algorithm and is stored in the terminal local trusted execution environment (TEE). The hardware-level encryption is used to ensure data security. The server side implements a lightweight storage scheme, only saving the fingerprint hash value + dynamic device trust level combination data. The device trust level is updated in real time based on terminal behavior data and security detection results, and asymmetric encryption transmission is used. The server does not save any original feature data such as device model and hardware serial number throughout the process, which maximizes the reduction of data leakage risk. The terminal device digital fingerprint generation method is applied to the technical field of terminal device duplicate-resistant digital fingerprint generation.

[0018] Embodiment: Taking a smartphone as an example, its operating system is Android 14, equipped with Snapdragon 8Gen3 chip, supporting SE chip. The SE chip is a highly secure embedded microcontroller that provides protection for sensitive data such as digital fingerprint information and user keys through hardware encryption and secure storage technology, effectively resisting external attacks and enhancing the security and reliability of the terminal device digital fingerprint generation process. Step 1, feature collection: Non-sensitive features: resolution 2796x1290, Chrome 118.0.5993, DNS delay 32ms; Semi-sensitive features: 8-core CPU, 16.0GB memory, after injecting =0.16 noise, 15.9GB, accelerometer sampling rate 100Hz; Highly sensitive features: Snapdragon 8Gen3 + SM8650 motherboard and MDM9250 baseband combination to generate hash, SE chip public key hash; Step two, local processing: Semi-sensitive features are processed by SHA-3-512 hash algorithm with 128-bit salt value; high-sensitive features are encrypted by 2048-bit Paillier key to generate 2048-bit key pair for encryption protection; Step three, fingerprint generation: Federal weights use hierarchical weighting strategy, and the contribution proportion of each feature dimension to digital fingerprint is determined by combining entropy weight method with expert experience: Hardware combination, weight 0.45: including CPU model, GPU architecture, sensor array and other core hardware information, as the basic identity of the device; Resolution, weight 0.15: covering screen pixel density, camera imaging parameters and other display-related features, to assist in distinguishing similar devices; Memory hash, weight 0.12: SHA-256 hash of runtime memory snapshot to capture the real-time running state of the device; Others (weight 0.28): including device serial number, firmware version, network MAC address and other supplementary information; In the feature processing stage, first, the original data of the above key dimensions is standardized to generate a fixed-length hash value as the Merkle tree leaf node. Through the hierarchical hash operation of Merkle tree, the root hash value is finally obtained. In order to ensure the uniqueness and accuracy of the fingerprint, the root hash is detected for collision by using the Bloom filter. After confirming that there is no conflict, a 256-bit device digital fingerprint is generated: 0x9b2f3a765c8d1e24a9f0b1c3d4e5f67890abcdef1234567890abcdef, here the 16 hexadecimal string is generated by SHA-256 encryption of device hardware ID, operating system version hash value, network MAC address and other information, and the first 32 characters are reserved as the core identifier of the device digital fingerprint. Subsequent dynamic update and anti-fake check can be realized by appending timestamp and check code. This fingerprint can be used as the unique identity certificate of the device in the federal system, for subsequent device authentication and access control; Step four, fingerprint management: The generation of Alipay sub-fingerprint uses HMAC-SHA256 encryption algorithm. The specific process is to take the hash value of the main fingerprint and the Alipay application package name "com.eg.android.AlipayGphone" as input parameters, and calculate the HMAC-SHA256 encryption to obtain: The Alipay sub-fingerprint = HMAC-SHA256 (main fingerprint + "com.eg.android.AlipayGphone" hash), the effective period of the sub-fingerprint is set to 24 hours, and the sub-fingerprint will be automatically invalidated after 24 hours, and a new sub-fingerprint needs to be calculated and generated to ensure the security of the transaction and the timeliness of the device identification.

[0019] Based on the embodiment, the effect is compared with the prior art, the comparison result is as shown in Figure 2 The feature reverse success rate is designed as follows: Sample size: 1000 different models of smart phones are selected, covering Android 12-14, iOS 16-17 systems, including high, medium and low hardware configurations; Feature data set: 100 pieces of 3 types of features are collected for each device, 200 pieces of non-sensitive features per device, 150 pieces of semi-sensitive features per device, and 50 pieces of high-sensitive features per device, with a total sample size of 1000x(200+150+50)=400,000 pieces; Attack model: simulate 3 typical attack scenarios: Attack model 1: brute force attack, numerical reverse inference based on known feature distribution; Attack model 2: side channel attack, reverse original features through encryption and hash algorithm output rules; Attack model 3: feature correlation analysis, infer using potential correlation between non-sensitive features and sensitive features; Experimental steps: Step 1: Process the original features according to the S2 step of the present application, inject Gaussian noise σ=1.5×s to the semi-sensitive features, and hash with 128-bit dynamic salt value; encrypt the high-sensitive features with 2048-bit Paillier; Step 2: Store the processed features and the original features separately, upload the processed features to the simulation server, and store the original features only on the terminal and destroy them immediately; Step 3: For each device's processed features, use 3 types of attack models to attempt 100 times of reverse inference, record the number of successful reverse inference of original features, and the success standard is that the error between the reverse value and the original value is less than or equal to 5%; The experimental results are as follows: Attack model Total attack times Successful backtracking times Backtracking success rate Brute force cracking 100,000 12 0.012% Side-channel attack 100,000 8 0.008% Feature correlation analysis 100,000 5 0.005% Total 300,000 25 0.0083% Result analysis: Gaussian noise generates controllable offset for numerical semi-sensitive features such as memory and delay through dynamic parameters =1.5 or 0.8), making it difficult for brute force attack to accurately match the original value; ​128-bit dynamic salt value is updated every 24 hours and only stored in the SE chip, making the SHA-3-512 hash output irregular, and blocking side-channel attack feature speculation; 2048-bit Paillier homomorphic encryption enables high-sensitive features to be calculated in encrypted state but cannot be decrypted, completely eliminating the correlation analysis of the original features. The experimental results show that the total reverse success rate is 0.0083% (<0.1%), which verifies the effectiveness of the privacy enhancement measures.

[0020] The following comparison summary can be drawn: The prior art has low recognition accuracy due to fixed feature collection dimension and static weight, which is difficult to cope with device dynamic changes such as hardware replacement and system update. The present application dynamically collects non-sensitive, semi-sensitive and high-sensitive three-level features, such as the screen resolution of the smart phone in the embodiment, hardware combination hash and other multi-dimensional features, and optimizes the feature weight by combining the federated learning framework, i.e. random forest model training + 7-day FedAvg aggregation. The weight of high-stability features is strengthened, such as the hardware combination hash weight reaching 0.45, which significantly improves the anti-interference ability of fingerprints to device changes, and the recognition accuracy is improved to 99.7% within 30 days. The prior art generally has the problem of sensitive features being exposed, and centralized storage, with high privacy leakage risk. The present application relies on terminal hardware security areas such as smart phone SE chips, and uses Gaussian noise injection , 128-bit dynamic salt value hash (SHA-3-512), 2048-bit Paillier homomorphic encryption and other technologies to realize that the original features do not leave the device and the encrypted features can be calculated. The original feature destruction rate reaches 100%, and the feature reverse success rate is reduced to less than 0.1%, effectively avoiding the risk of data leakage and cross-scene tracking. The prior art lacks systematic anti-collision design, and is prone to fingerprint duplication in large-scale device scenarios. The present application innovatively introduces Merkle tree to dynamically calculate the depth to ensure integrity and Bloom filter false positive rate p=0.0001 combination mechanism, combined with collision dynamic processing logic, i.e. adding random features to recalculate, which reduces the collision rate to 8.3 ×10-9 in a 1 billion device scale, which is 3 orders of magnitude lower than the prior art, especially solving the problem of fingerprint duplication of standardized devices in the same batch.

[0021] Based on the above description, the application sets up multi-dimensional three-level feature collection, realizes multi-dimensional fine-grained feature collection from non-sensitive, semi-sensitive and high-sensitive layers, ensures that feature covers device hardware core attributes and dynamic environment parameters, optimizes feature weight through a federal learning framework, effectively resists fingerprint drift caused by hardware replacement and software update, injects dynamic Gaussian noise into numerical semi-sensitive features to balance privacy protection and feature effectiveness, uses 2048-bit key Paillier homomorphic encryption for high-sensitive features, and supports feature fusion calculation in an encrypted state, at the same time, blocks cross-scene tracking risk through a scene-based sub-fingerprint generation and original feature instant destruction mechanism, and introduces a Merkle tree and Bloom filter fusion anti-repetition architecture to automatically add random features for recalculation of overlapping digital fingerprints, thereby greatly reducing fingerprint overlap collision rate.

[0022] Embodiments of the application are presented by way of example and description, and are not intended to be exhaustive or to limit the application to the form disclosed. Many modifications and variations will be apparent to those skilled in the art. Embodiments are chosen and described in order to best explain the principles of the application and its practical application, and to enable others skilled in the art to understand the application for various embodiments with various modifications as are suited to the particular use contemplated.

Claims

1. A method for generating a digital fingerprint for a terminal device, characterized in that: Includes the following steps: S1. Multidimensional hierarchical feature acquisition: A three-level feature acquisition layer is established, consisting of non-sensitive, semi-sensitive, and highly sensitive features. Non-sensitive features include screen physical resolution and operating system kernel compilation time; semi-sensitive features include the number of CPU cores and total memory capacity; and highly sensitive features include a concatenated string composed of CPU model, motherboard model, and baseband version. S2, Enhanced Privacy: Within the secure area of ​​the terminal hardware, Gaussian noise is injected into numerical semi-sensitive features, salted hashing is performed on semi-sensitive features using a 128-bit dynamic salt value and the SHA-3-512 algorithm, and Paillier homomorphic encryption with a 2048-bit key is used for highly sensitive features. S3, fingerprint anti-duplicate generation: A distributed training system is constructed, and the feature weights are trained locally on the terminal using a random forest model with a learning rate of [missing information]. The server uses the FedAvg algorithm to aggregate models every 7 days, and fuses features through Merkle trees and Bloom filters to generate anti-duplicate digital fingerprints; S4. Fingerprint Lifecycle Management: Sub-fingerprints are generated using HMAC-SHA256 based on the master fingerprint and scene identifier. The master fingerprint is automatically updated every 30 days, and the original feature data is destroyed immediately.

2. The method for generating a digital fingerprint for a terminal device according to claim 1, characterized in that: In step S1, the screen physical resolution of the non-sensitive features is accurate to 1 pixel, and the operating system kernel compilation time is accurate to seconds. The total memory capacity in the semi-sensitive feature is accurate to 0.1GB.

3. The method for generating a digital fingerprint for a terminal device according to claim 1, characterized in that: In step S1, the sampling frequency for non-sensitive features is once per session, the sampling frequency for semi-sensitive features is once every 24 hours and requires user authorization, and highly sensitive features are only collected when the device is first installed or reset.

4. The method for generating a digital fingerprint for a terminal device according to claim 1, characterized in that: In step S2, Gaussian noise , It is 1.5 or 0.

8. For feature sensitivity; The 128-bit dynamic salt value is updated every 24 hours.

5. The method for generating a digital fingerprint for a terminal device according to claim 1, characterized in that: In step S3, the Merkle tree is: Merkle trees are used to verify the integrity and consistency of data.

6. The method for generating a digital fingerprint for a terminal device according to claim 1, characterized in that: In step S3, the false positive rate of the Bloom filter is p=0.0001, which is achieved by optimizing the number of hash functions. By considering the bit vector length, lightweight storage and fast retrieval of feature sets can be achieved; among them, The number of elements to be inserted. This represents the number of bits in the filter.

7. The method for generating a digital fingerprint for a terminal device according to claim 1, characterized in that: In step S4, the master fingerprint is the anti-duplication digital fingerprint generated in step S3; the scene identifier refers to a set of feature information used to mark and distinguish different application environments, business operation scenarios or device usage scenarios during the process of master fingerprint regeneration.

8. The method for generating a digital fingerprint for a terminal device according to claim 7, characterized in that: The feature information includes, but is not limited to, time information, geographical location information, business type, and device usage mode. Through scene identification, it is used to make the generation of digital fingerprints more in line with the specific use case requirements.

9. A method for generating a digital fingerprint for a terminal device according to any one of claims 1-8, characterized in that: The terminal device digital fingerprint generation method is applied to the technical field of terminal device anti-replication digital fingerprint generation.

Images