User age group identification method and related apparatus

Abstract

This invention belongs to the field of identity authentication and discloses a method and related apparatus for user age group identification. The method includes acquiring multi-axis motion sensor data from a device and constructing user age group identification feature data; and calling a pre-trained age group identification model to obtain the age group identification result. The pre-trained model is obtained by first training an initial identification model using generated sample data and then fine-tuning it using real sample data. The generated sample data is produced by a feature generation model based on a generative adversarial network framework combined with pattern-specific normalization technology, using age group labels as conditions, to simulate the unimodal or multimodal distribution differences of behavioral characteristics of different age groups. This invention overcomes the bottlenecks of scarce real labeled data and high acquisition costs through generative behavior modeling and pre-training fine-tuning strategies, improving the accuracy and generalization ability of age group identification in complex scenarios, and avoiding the privacy and ethical risks associated with the collection of sensitive biometric features.

Description

Technical Field

[0001] This invention belongs to the field of identity authentication and relates to a method and related apparatus for identifying user age groups. Background Technology

[0002] With the widespread use of smartphones among minors, mobile phones have become a core terminal for children to access learning resources and engage in social entertainment. However, due to a lack of self-control, minors face risks such as gaming addiction, cyberbullying, and exposure to harmful information. Therefore, how to accurately and in real-time identify whether the current user is a child during device use and implement corresponding protective measures has become a critical issue that urgently needs to be addressed in the fields of mobile security and child protection. Early identity verification methods relied heavily on traditional biometric technologies, such as fingerprint and facial recognition. These methods are typically only applied when a user first accesses the device and cannot provide continuous identity protection throughout the entire usage process after the device is unlocked. Furthermore, while facial recognition is widely used, it raises significant privacy and ethical issues, especially in applications involving minors, where strict privacy regulations limit the large-scale collection of sensitive biometric features.

[0003] Therefore, current work is increasingly shifting towards more continuous and implicit identification methods, namely, identifying users by analyzing sensor behavior patterns during smartphone use. Despite significant efforts in feature design, existing behavior recognition methods still face several serious challenges. First, child behavior recognition suffers from severe data scarcity. Due to high costs of field data collection and stringent privacy restrictions, existing child behavior datasets are small in scale and lack ecological validity, making it difficult for models to cover the diverse behavioral distributions in real-world scenarios. Second, most existing recognition models rely on limited real-world samples, exhibiting weak generalization ability. When faced with complex real-world scenarios and subtle age differences, their recognition accuracy is often unsatisfactory. Furthermore, group behavior itself is highly complex; children of different ages exhibit significant differences in movement intensity, rhythm, and coordination. Simple statistical features or rigid models struggle to capture these fine-grained behavioral patterns. Therefore, current child identification methods still fall short of meeting the high-precision, privacy-protected identification requirements in real-world open scenarios. A new child user identification scheme is urgently needed that can improve the robustness and coverage of smartphone child user identification while protecting privacy. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and related apparatus for identifying user age groups.

[0005] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention provides a method for identifying a user's age group, comprising: acquiring multi-axis motion sensor data generated by a device based on user usage; constructing user age group identification feature data based on the multi-axis motion sensor data; and calling a pre-trained age group identification model based on the user age group identification feature data to obtain the user's age group identification result; wherein the pre-trained age group identification model is obtained by: acquiring several real sample data and several generated sample data; training an initial age group identification model using several generated sample data, and fine-tuning the trained age group identification model using several real sample data to obtain a pre-trained age group identification model; wherein the real sample data includes real age group identification feature data and corresponding age group labels; the generated sample data includes generated age group identification feature data and corresponding age group labels; the generated age group identification feature data is generated by a pre-trained feature generation model; the feature generation model is constructed based on a generative adversarial network framework combined with pattern-specific normalization technology, using age group labels as conditional data.

[0006] Optionally, the multi-axis motion sensor data includes triaxial accelerometer data and / or triaxial gyroscope data; wherein, both triaxial accelerometer data and triaxial gyroscope data include sampled values, timestamp information, and device metadata.

[0007] Optionally, before constructing user age group identification feature data based on the multi-axis motion sensor data, the method further includes: performing cascaded filtering preprocessing on the multi-axis motion sensor data using a median filter and a Butterworth low-pass filter.

[0008] Optionally, constructing user age group identification feature data based on the multi-axis motion sensor data includes: segmenting the multi-axis motion sensor data using a fixed step-size sliding window to obtain several window data; performing statistical aggregation of user behavior statistical descriptive quantities on each window data to obtain user age group identification feature data for each window data; wherein, the dimensions of user behavior statistical descriptive quantities include one or more of the following: activity intensity, motion smoothness, rhythm, explosiveness, coordination, and posture.

[0009] Optionally, the step of calling a pre-trained age group identification model based on the user age group identification feature data to obtain the user's age group identification result includes: inputting the user age group identification feature data of each window data into the pre-trained age group identification model to obtain the age group identification result of each window data; and performing majority voting based on the age group identification result of each window data to obtain the user's age group identification result.

[0010] Optionally, the generation of age group identification feature data is obtained by the following method: inputting random noise and target age group labels into a pre-trained feature generation model to generate generation age group identification feature data of target age group labels, and repeating this process several times; performing utility evaluation screening and physical constraint filtering on each generation age group identification feature data; wherein, the utility evaluation screening is used to screen generation age group identification feature data with confidence levels higher than a preset confidence threshold and generation age group identification feature data with distances less than a preset distance from the decision boundary.

[0011] Optionally, the feature-based model generation includes a generator and a discriminator; the generator includes multiple cascaded network layers, and at least one network layer has a pattern-specific normalization module; wherein, the pattern-specific normalization module is used to receive age group labels, generate normalization adjustment parameters corresponding to the age group labels, and apply the normalization adjustment parameters to normalize the output data of the corresponding network layer before outputting it to the next network layer; the generator is used to receive random noise and age group labels, and output generated age group identification feature data; the discriminator is used to receive the generated age group identification feature data and the corresponding age group labels, and output a judgment result on whether the generated age group identification feature data is real data and whether it matches the corresponding age group label.

[0012] In a second aspect, the present invention provides a user age group identification system, comprising: a data acquisition module for acquiring multi-axis motion sensor data generated by a device based on user usage; a feature construction module for constructing user age group identification feature data based on the multi-axis motion sensor data; and an age identification module for calling a pre-trained age group identification model based on the user age group identification feature data to obtain the user's age group identification result; wherein the pre-trained age group identification model is obtained by: acquiring several real sample data and several generated sample data; training an initial age group identification model using several generated sample data, and fine-tuning the trained age group identification model using several real sample data to obtain the pre-trained age group identification model; wherein the real sample data includes real age group identification feature data and corresponding age group labels; the generated sample data includes generated age group identification feature data and corresponding age group labels; the generated age group identification feature data is generated by a pre-trained feature generation model; the feature generation model is constructed based on a generative adversarial network framework combined with pattern-specific normalization technology, using age group labels as conditional data.

[0013] In a third aspect, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the user age group identification method.

[0014] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the user age group identification method.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a user age group identification method that acquires multi-axis motion sensor data naturally generated by device sensors during user use and constructs user age group identification feature data. This achieves fully implicit and continuous identity recognition, overcoming the shortcomings of traditional biometric technologies that can only verify at the initial point of access and cannot provide all-time protection. It also avoids the privacy and ethical risks associated with collecting sensitive biometric information such as facial features. Furthermore, by constructing a generative adversarial network conditioned on age group labels and introducing pattern-specific normalization technology, the feature generation model can accurately model the distribution differences of behavioral characteristics among different age groups, synthesizing high-fidelity generated sample data. Through a two-stage training strategy—first training the initial identification model with generated samples and then fine-tuning it with real samples—this effectively overcomes the bottleneck of scarce large-scale labeled data caused by the high cost of collecting real behavioral data and privacy restrictions. This allows the final age group identification model to fully learn diverse behavioral patterns while accurately adapting to real physical scenarios, significantly improving the robustness and generalization ability of identifying different age groups, especially children, in complex open environments. Attached Figure Description

[0016] Figure 1 This is a flowchart of a user age group identification method according to an embodiment of the present invention.

[0017] Figure 2 This is a block diagram of the user age group identification system according to an embodiment of the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] The present invention will now be described in further detail with reference to the accompanying drawings: See Figure 1 In one embodiment of the present invention, a method for identifying user age groups is provided, which effectively overcomes the shortcomings of existing technologies in identifying user age groups, such as the lack of large-scale datasets, high costs of on-site collection, and insufficient generalization ability of existing models in complex open scenarios. It can effectively achieve high-accuracy identification of child users on smartphones and effectively alleviate the risks of privacy leakage and ethical concerns in data collection.

[0021] Specifically, the user age group identification method of the present invention includes the following steps: S1: Acquire multi-axis motion sensor data generated by the device based on user usage.

[0022] S2: Construct user age group identification feature data based on the multi-axis motion sensor data.

[0023] S3: Based on the user age group identification feature data, call the pre-trained age group identification model to obtain the user's age group identification result.

[0024] The pre-trained age group identification model is obtained in the following manner: Acquire several real sample data and several generated sample data; train an initial age group identification model using several generated sample data, and fine-tune the trained age group identification model using several real sample data to obtain a pre-trained age group identification model.

[0025] The real sample data includes real age group identification feature data and corresponding age group labels; the generated sample data includes generated age group identification feature data and corresponding age group labels; the generated age group identification feature data is generated through a pre-trained feature generation model; the feature generation model is constructed based on a generative adversarial network framework combined with pattern-specific normalization technology, using age group labels as conditional data.

[0026] Explanatoryly, the user age group identification method of the present invention uses multi-axis motion sensor data generated by the device based on user usage as the identification basis. It can identify user age groups in complex open scenarios by using only low-frequency sensor data, and can achieve data compliance, privacy protection and continuous protection of minors.

[0027] Specifically, in terms of data acquisition and feature representation, multi-axis motion sensor data can be collected without the user's awareness during the user's natural use of devices such as smartphones (e.g., game interaction).

[0028] Interpretive data augmentation focuses on expanding the coverage of real-world datasets using synthetic data. Specifically, in the pre-training process of the age group identification model, a data substitution or hybrid augmentation strategy was employed. This involves pre-training the model using high-quality synthetic datasets in the early stages of training, enabling the model to learn a broad distribution of behavioral patterns; subsequently, fine-tuning is performed using a small amount of real-world labeled data to ensure the model's adaptability to real-world physical scenarios. This method effectively expands the training sample size and improves the model's robustness and generalization ability without relying on large-scale field data collection.

[0029] For example, the feature generation model employs a generative adversarial network framework, consisting of a generator and a discriminator engaging in a game of strategy. Pattern-specific normalization technology is embedded in the generator's network layer. This technology enables the generator to dynamically calculate age-specific normalization parameters based on the input age group label, allowing for differentiated modeling of the potential unimodal or multimodal distributions of behavioral characteristics across different age groups. Compared to traditional generative models that treat all age data indiscriminately and easily lose subtle behavioral differences between groups, this approach, through the combination of conditionalization and pattern-specific normalization, can synthesize high-fidelity samples that statistically closely approximate the real data of each age group. This provides downstream recognition models with ample and diverse training data, effectively overcoming privacy restrictions and scale bottlenecks in collecting real children's behavioral data, and significantly improving the generalization ability and fine-grained age recognition accuracy of the recognition model in complex open scenarios.

[0030] This invention provides a user age group identification method that acquires multi-axis motion sensor data naturally generated by device sensors during user use and constructs user age group identification feature data. This achieves fully implicit and continuous identity recognition, overcoming the shortcomings of traditional biometric technologies that can only verify at the initial point of access and cannot provide all-time protection. It also avoids the privacy and ethical risks associated with collecting sensitive biometric information such as facial features. Furthermore, by constructing a generative adversarial network conditioned on age group labels and introducing pattern-specific normalization technology, the feature generation model can accurately model the distribution differences of behavioral characteristics among different age groups, synthesizing high-fidelity generated sample data. Through a two-stage training strategy—first training the initial identification model with generated samples and then fine-tuning it with real samples—this effectively overcomes the bottleneck of scarce large-scale labeled data caused by the high cost of collecting real behavioral data and privacy restrictions. This allows the final age group identification model to fully learn diverse behavioral patterns while accurately adapting to real physical scenarios, significantly improving the robustness and generalization ability of identifying different age groups, especially children, in complex open environments.

[0031] In one possible implementation, the multi-axis motion sensor data includes triaxial accelerometer data and / or triaxial gyroscope data; wherein both the triaxial accelerometer data and the triaxial gyroscope data include sampled values, timestamp information, and device metadata.

[0032] Explaining this, using smartphone game interaction as an example, the method acquires raw behavioral data generated by device sensors during user interaction while playing games on a smartphone, including three-axis accelerometer data and three-axis gyroscope data. Both the three-axis accelerometer and three-axis gyroscope data include sampled values ​​along the three orthogonal axes (x, y, and z), as well as timestamp information corresponding to each sample point. They may also be associated with device metadata. Furthermore, the user's age group label is associated when collecting sample data. By collecting data in natural interaction states (such as game scenarios), the shortcomings of traditional biometric technologies (such as facial recognition and fingerprint recognition) in terms of continuous monitoring and privacy protection are avoided.

[0033] In one possible implementation, before constructing user age group identification feature data based on the multi-axis motion sensor data, the method further includes: performing cascaded filtering preprocessing on the multi-axis motion sensor data using a median filter and a Butterworth low-pass filter.

[0034] Interpretive, cascaded filtering is performed on multi-axis motion sensor data. First, a median filter is used to suppress outlier pulses caused by the sensor hardware. Then, a Butterworth low-pass filter is used to smooth high-frequency motion noise to preserve low-frequency components that reflect human motion patterns.

[0035] Specifically, by combining median filtering and low-pass filtering, it is possible to effectively remove transient abnormal shocks and electronic noise generated during hardware acquisition while retaining the dominant frequency components related to natural human movement behavior. This significantly improves the signal-to-noise ratio and stability of the original sensor data, providing a cleaner and more reliable motion signal input for subsequent sliding window segmentation and feature extraction. It also avoids distortion of behavioral representation caused by noise interference, ultimately improving the generalization ability and recognition accuracy of the age group identification model.

[0036] In one possible implementation, constructing user age group identification feature data based on the multi-axis motion sensor data includes: segmenting the multi-axis motion sensor data using a fixed step-size sliding window to obtain several window data; and performing statistical aggregation of user behavior statistical descriptive quantities on each window data to obtain user age group identification feature data for each window data.

[0037] Interpretive, a fixed-step sliding window is used to segment the preprocessed multi-axis motion sensor data, dividing the long-term behavior stream into independent behavior segments containing temporal context information.

[0038] Specifically, by dividing the data into sliding windows with a fixed step size, continuous and unbounded sensor data streams can be transformed into behavioral units with a unified time scale and contextual relevance, thereby effectively capturing the local motion patterns and dynamic changes of users in different time segments. On this basis, statistical aggregation of multi-axis motion sensor data within each window can extract behavioral representation vectors covering multiple dimensions, forming a structured feature expression. This not only preserves the key behavioral information in the original sequence but also significantly reduces the data dimensionality, providing information-rich and stable input features for subsequent generative modeling and classification, effectively improving the model's ability to distinguish behavioral differences among different age groups.

[0039] Optionally, the dimensions of the user behavior statistics description include one or more of the following: activity intensity, motion smoothness, rhythm, explosiveness, coordination, and posture.

[0040] For example, the activity intensity dimension may include indicators such as the mean, peak, and standard deviation of three-axis acceleration to measure the overall amplitude and force of the user's movements; the motion smoothness dimension may include indicators such as the variance of the difference between adjacent sampling points of the acceleration signal and the statistics of accelerometer, to reflect the smoothness of the movements and whether there are high-frequency jitters or stutters; the rhythm dimension may include indicators such as the zero-crossing rate of the acceleration signal and the mean and standard deviation of the peak interval, to characterize the speed of the user's operation frequency and its temporal regularity; the explosiveness dimension may include indicators such as acceleration kurtosis, skewness, and the ratio of peak to mean, to capture whether the user has sudden, brief bursts of high-intensity operations during interaction; the coordination dimension may include indicators such as the Pearson correlation coefficient between pairs of three-axis accelerations and the correlation between acceleration and gyroscope signals, to assess the degree of coordination between different directions and types of user movement; the posture dimension may include indicators such as the average tilt angle of each axis obtained by gyroscope integration and its variance, to characterize the main orientation of the user holding the phone and the stability of that orientation.

[0041] In one possible implementation, the step of calling a pre-trained age group identification model based on the user age group identification feature data to obtain the user's age group identification result includes: inputting the user age group identification feature data of each window data into the pre-trained age group identification model to obtain the age group identification result of each window data; and performing majority voting processing based on the age group identification result of each window data to obtain the user's age group identification result.

[0042] Interpretively, user age group identification feature data extracted in real time from multiple windows are sequentially input into a pre-trained age group identification model to obtain a series of age group identification results corresponding to the data from each window. For each user, a majority vote is performed on all age group identification results generated within the monitoring period, and the age group category with the highest frequency is taken as the user's final age group identification result. In this implementation, decision fusion across time windows effectively eliminates misjudgment interference caused by random fluctuations in individual actions within a single behavior segment, improving the accuracy and stability of age group identification.

[0043] In one possible implementation, the generation of age group identification feature data is obtained by: inputting random noise and target age group labels into a pre-trained feature generation model to generate generation age group identification feature data of the target age group labels, and repeating this process several times; performing utility evaluation screening and physical constraint filtering on each generation age group identification feature data; wherein, the utility evaluation screening is used to screen generation age group identification feature data with a confidence level higher than a preset confidence threshold and generation age group identification feature data with a distance from the decision boundary less than a preset distance.

[0044] While the generated age group identification feature data produced by the feature generation model can simulate the real distribution, it may produce redundant samples with low contribution to downstream tasks or invalid samples that violate physical laws. Therefore, to enhance the effectiveness of the data, this implementation introduces a dual filtering mechanism: First, a utility evaluation screening is performed, selecting high-confidence within-class samples and difficult samples close to the decision boundary; second, a physical constraint filtering mechanism is introduced, verifying the numerical physical consistency of the data on each axis in the generated age group identification feature data to ensure that the generated age group identification feature data strictly conforms to the physical laws of motion in the real world.

[0045] Specifically, this implementation refines the candidate generated data output by the feature generation model through a dual filtering mechanism combining utility evaluation screening and physical constraint filtering. On the one hand, utility evaluation screening prioritizes retaining typical samples with high classification confidence for the recognition model and difficult samples located near the decision boundary, thereby enhancing the relevance and training efficiency of the generated data for downstream recognition tasks and avoiding redundant samples that waste computational resources. On the other hand, physical constraint filtering verifies the numerical physical consistency between the data of each axis in the generated feature vector, eliminating invalid samples that violate real motion laws and ensuring the physical rationality of the synthesized data. This dual filtering mechanism ensures that the generated samples used for model training are statistically close to real data, focus on key discrimination regions in terms of task utility, and conform to kinematic laws in terms of physical logic, thereby significantly improving the generalization ability and recognition accuracy of the age group recognition model trained based on this data.

[0046] In one possible implementation, the feature generation model includes a generator and a discriminator; the generator includes multiple cascaded network layers, and at least one network layer has a pattern-specific normalization module; wherein, the pattern-specific normalization module is used to receive age group labels, generate normalization adjustment parameters corresponding to the age group labels, and apply the normalization adjustment parameters to normalize the output data of the corresponding network layer before outputting it to the next network layer; the generator is used to receive random noise and age group labels, and output generated age group identification feature data; the discriminator is used to receive the generated age group identification feature data and the corresponding age group labels, and output a judgment result on whether the generated age group identification feature data is real data and whether it matches the corresponding age group label.

[0047] Interpretive, normalized adjustment parameters generally include scaling and offset parameters, and different age group labels correspond to independent sets of parameters.

[0048] In this embodiment, the feature generation model employs a conditional generative adversarial network architecture, consisting of a generator and a discriminator engaging in a game of mutual competition. Specifically, the user's behavioral feature space is treated as a generation canvas, with fine-grained age group labels serving as conditional guidance, enabling the model to map the behavioral distribution patterns of different age groups. The generator consists of multiple cascaded network layers, with at least one layer embedding a pattern-specific normalization module. This module receives the currently specified age group label, dynamically calculates the age group-specific normalization adjustment parameters based on the label, and uses these parameters to stretch and shift the intermediate feature data output by the layer, completing the style injection for that age group before passing it to the next layer. Through the aforementioned pattern-specific normalization technique, when the behavioral features of different age groups exhibit a unimodal or multimodal distribution in the statistical space, the generator can specifically adopt a modeling strategy that matches the distribution pattern, thereby generating high-fidelity candidate synthetic behavioral samples that statistically closely approximate the real data. The discriminator simultaneously receives the synthetic feature data and its corresponding age group label, and comprehensively judges whether the data is real and matches the age group label, thereby guiding the generator to continuously improve the fidelity and age specificity of the synthetic data.

[0049] In one possible implementation, the age group identification model employs a one-dimensional convolutional neural network or residual network as the backbone network to extract features and classify age groups from the input user age group identification feature data. Specifically, the model extracts local short-term correlations in behavioral sequences layer by layer along the time dimension through stacked multi-layer one-dimensional convolutional operations, and gradually aggregates them into global long-term motion trend features as the network depth increases; finally, it outputs the predicted probability of the behavioral segment belonging to each preset age group through a fully connected classification layer, which serves as the age group identification result for the window data.

[0050] In one possible implementation, the network layers of the generator and discriminator of the feature generation model employ several fully connected layers cascaded in sequence.

[0051] In summary, the user age group identification method of this invention, by constructing a feature generation model based on age conditions and introducing a constraint filtering mechanism, can perform excellently in complex open scenarios using only low-frequency sensor data, thereby achieving data compliance, privacy protection, and continuous protection of minors.

[0052] In terms of data acquisition and feature representation, the system collects raw three-axis accelerometer and gyroscope data during natural smartphone use (such as game interaction) and transforms it into multi-dimensional behavioral representation vectors containing dimensions such as activity intensity, rhythm, and posture through cascaded filtering and sliding window techniques. Regarding generative modeling and quality control, user behavior is modeled as a conditional generation process guided by age labels. An age-conditional generation network captures the differences in behavioral patterns among different age groups in the feature space, generating high-fidelity candidate synthetic behavioral data. Simultaneously, to ensure the utility and authenticity of the synthetic data, the system implements dual filtering: behavioral perception filtering focuses on the contribution of each sample to the downstream recognition task, selecting high-confidence and difficult samples near the decision boundary; while physical constraint filtering provides physical dimension verification, ensuring that the synthetic behavior conforms to the motion laws of the real world by verifying the numerical logical relationships between features. These two types of constraints complement each other: task-oriented focuses on discriminative and heuristic properties, while physical constraints focus on authenticity and compliance. By integrating rigorously selected high-quality synthetic datasets, a multi-dimensional simulation of real-world behavioral distribution is achieved. The enhanced feature representations are fed into a deep neural network for training and prediction, which significantly improves the accuracy and robustness of child identification in data-scarce and complex scenarios.

[0053] This invention presents a user age group identification method that can be used for accurate age group identification in unrestricted scenarios. It is the first technical framework to address the scarcity of children's behavioral data through generative behavior modeling. By fully utilizing generative artificial intelligence technology to uncover potential behavioral distributions, it effectively alleviates the challenges of data collection under privacy compliance. Furthermore, it introduces stringent physical constraints and task-oriented screening to avoid the inconsistent sample quality issues associated with traditional generative models. This invention demonstrates good feasibility and effectiveness in improving the accuracy of child protection and identity verification, providing an innovative data augmentation and identification path for mobile security research.

[0054] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not disclosed in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0055] See Figure 2 In another embodiment of the present invention, a user age group identification system is provided, which can be used to implement the above-mentioned user age group identification method. Specifically, the user age group identification system includes a data acquisition module, a feature construction module, and an age recognition module.

[0056] The system comprises three modules: a data acquisition module for acquiring multi-axis motion sensor data generated by the device based on user usage; a feature construction module for constructing user age group identification feature data based on the multi-axis motion sensor data; and an age recognition module for calling a pre-trained age group recognition model based on the user age group identification feature data to obtain the user's age group identification result. The pre-trained age group recognition model is obtained through the following methods: acquiring several real sample data and several generated sample data; training an initial age group recognition model using the generated sample data, and fine-tuning the trained age group recognition model using the real sample data to obtain the pre-trained age group recognition model. The real sample data includes real age group identification feature data and corresponding age group labels; the generated sample data includes generated age group identification feature data and corresponding age group labels; the generated age group identification feature data is generated through a pre-trained feature generation model; and the feature generation model is constructed based on a generative adversarial network framework combined with pattern-specific normalization technology, using age group labels as conditional data.

[0057] In one possible implementation, the multi-axis motion sensor data includes triaxial accelerometer data and / or triaxial gyroscope data; wherein both the triaxial accelerometer data and the triaxial gyroscope data include sampled values, timestamp information, and device metadata.

[0058] In one possible implementation, before constructing user age group identification feature data based on the multi-axis motion sensor data, the method further includes: performing cascaded filtering preprocessing on the multi-axis motion sensor data using a median filter and a Butterworth low-pass filter.

[0059] In one possible implementation, constructing user age group identification feature data based on the multi-axis motion sensor data includes: segmenting the multi-axis motion sensor data using a fixed-step sliding window to obtain several window data; performing statistical aggregation of user behavior statistical descriptors on each window data to obtain user age group identification feature data for each window data; wherein, the dimensions of the user behavior statistical descriptors include one or more of the following: activity intensity, motion smoothness, rhythm, explosiveness, coordination, and posture.

[0060] In one possible implementation, the step of calling a pre-trained age group identification model based on the user age group identification feature data to obtain the user's age group identification result includes: inputting the user age group identification feature data of each window data into the pre-trained age group identification model to obtain the age group identification result of each window data; and performing majority voting processing based on the age group identification result of each window data to obtain the user's age group identification result.

[0061] In one possible implementation, the generation of age group identification feature data is obtained by: inputting random noise and target age group labels into a pre-trained feature generation model to generate generation age group identification feature data of the target age group labels, and repeating this process several times; performing utility evaluation screening and physical constraint filtering on each generation age group identification feature data; wherein, the utility evaluation screening is used to screen generation age group identification feature data with a confidence level higher than a preset confidence threshold and generation age group identification feature data with a distance from the decision boundary less than a preset distance.

[0062] In one possible implementation, the feature generation model includes a generator and a discriminator; the generator includes multiple cascaded network layers, and at least one network layer has a pattern-specific normalization module; wherein, the pattern-specific normalization module is used to receive age group labels, generate normalization adjustment parameters corresponding to the age group labels, and apply the normalization adjustment parameters to normalize the output data of the corresponding network layer before outputting it to the next network layer; the generator is used to receive random noise and age group labels, and output generated age group identification feature data; the discriminator is used to receive the generated age group identification feature data and the corresponding age group labels, and output a judgment result on whether the generated age group identification feature data is real data and whether it matches the corresponding age group label.

[0063] All relevant content of each step involved in the aforementioned embodiments of the user age group identification method can be referenced to the functional description of the corresponding functional module of the user age group identification system in the embodiments of the present invention, and will not be repeated here.

[0064] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0065] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of a user age group identification method.

[0066] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the user age group identification method in the above embodiments.

[0067] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0068] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0069] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0070] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for identifying user age groups, characterized in that, include: Acquire multi-axis motion sensor data generated by the device based on user usage; Based on the multi-axis motion sensor data, construct user age group identification feature data; Based on the user age group identification feature data, a pre-trained age group identification model is invoked to obtain the user's age group identification result; The pre-trained age group identification model is obtained in the following manner: Obtain several real sample data and several generated sample data; An initial age group identification model is trained using several generated sample data, and a pre-trained age group identification model is obtained by fine-tuning the trained model using several real sample data. The real sample data includes real age group identification feature data and corresponding age group labels; the generated sample data includes generated age group identification feature data and corresponding age group labels; the generated age group identification feature data is generated through a pre-trained feature generation model; the feature generation model is constructed based on a generative adversarial network framework combined with pattern-specific normalization technology, using age group labels as conditional data.

2. The user age group identification method according to claim 1, characterized in that, The multi-axis motion sensor data includes triaxial accelerometer data and / or triaxial gyroscope data; wherein, both triaxial accelerometer data and triaxial gyroscope data include sampled values, timestamp information, and device metadata.

3. The user age group identification method according to claim 1, characterized in that, Before constructing user age group identification feature data based on the multi-axis motion sensor data, the method further includes: performing cascaded filtering preprocessing on the multi-axis motion sensor data using a median filter and a Butterworth low-pass filter.

4. The user age group identification method according to claim 1, characterized in that, The step of constructing user age group identification feature data based on the multi-axis motion sensor data includes: The data from the multi-axis motion sensor is segmented using a fixed-step sliding window to obtain several window data points. The statistical aggregation of user behavior descriptive metrics for each window data is performed to obtain user age group identification feature data for each window data; the dimensions of user behavior descriptive metrics include one or more of the following: activity intensity, movement smoothness, rhythm, explosiveness, coordination and posture.

5. The user age group identification method according to claim 4, characterized in that, The step of calling a pre-trained age group identification model based on the user's age group identification feature data to obtain the user's age group identification result includes: The user age group identification feature data of each window data is input into the pre-trained age group identification model to obtain the age group identification results of each window data; and the majority voting process is performed based on the age group identification results of each window data to obtain the user's age group identification results.

6. The user age group identification method according to claim 1, characterized in that, The generated age group identification feature data is obtained in the following manner: Random noise and target age group labels are input into a pre-trained feature generation model to generate age group identification feature data for the target age group labels, and this process is repeated several times. Each generated age group identification feature data is subjected to utility evaluation screening and physical constraint filtering; the utility evaluation screening is used to screen generated age group identification feature data with confidence levels higher than a preset confidence threshold and generated age group identification feature data with distances less than a preset distance from the decision boundary.

7. The user age group identification method according to claim 1, characterized in that, The model generated based on the features includes a generator and a discriminator; the generator includes multiple cascaded network layers, and at least one network layer has a pattern-specific normalization module. The pattern-specific normalization module receives age group labels, generates normalization adjustment parameters corresponding to the age group labels, and applies the normalization adjustment parameters to normalize the output data of the corresponding network layer before outputting it to the next network layer. The generator receives random noise and age group labels and outputs generated age group identification feature data. The discriminator receives the generated age group identification feature data and the corresponding age group labels, and outputs a judgment result on whether the generated age group identification feature data is real data and whether it matches the corresponding age group labels.

8. A user age group identification system, characterized in that, include: The data acquisition module is used to acquire multi-axis motion sensor data generated by the device based on user usage; The feature construction module is used to construct user age group identification feature data based on the multi-axis motion sensor data; The age recognition module is used to call a pre-trained age group recognition model based on the user's age group recognition feature data to obtain the user's age group recognition result; The pre-trained age group identification model is obtained in the following manner: Obtain several real sample data and several generated sample data; An initial age group identification model is trained using several generated sample data, and a pre-trained age group identification model is obtained by fine-tuning the trained model using several real sample data. The real sample data includes real age group identification feature data and corresponding age group labels; the generated sample data includes generated age group identification feature data and corresponding age group labels; the generated age group identification feature data is generated through a pre-trained feature generation model; the feature generation model is constructed based on a generative adversarial network framework combined with pattern-specific normalization technology, using age group labels as conditional data.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the user age group identification method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the user age group identification method as described in any one of claims 1 to 7.

Images