Training method, device, electronic device and storage medium for target perception model

Abstract

The present disclosure discloses a method, apparatus, electronic device, and storage medium for training a target perception model. The method includes: calculating multiple spectrograms corresponding to CSI signals within multiple time periods; mixing them to form sample data; inputting the data into the target perception model, and training the target perception model by performing the following tasks: encoding each input feature pair to obtain corresponding spectrogram features, so that the spectrogram features can be decoded to restore the corresponding input features; representing the noise floor features corresponding to CSI signals without targets as low-dimensional noise floor vectors, such that the similarity between low-dimensional noise floor vectors in the same Wi-Fi environment is greater than or equal to a first preset threshold, and the similarity between low-dimensional noise floor vectors in different Wi-Fi environments is greater than a second threshold; extracting domain-invariant features that are independent of the Wi-Fi environment from the input feature pairs; and identifying target information based on the input feature pairs.

Description

Technical Field

[0001] The present disclosure relates to the field of communication technology, and in particular to a training method, device, electronic device and storage medium for a target perception model. Background Art

[0002] Currently common human behavior recognition or detection technologies have promising applications, but they also have certain drawbacks: computer vision-based technologies are limited by lighting conditions and are prone to involving sensitive user information; wearable devices require specialized supporting equipment; and small radars are not suitable for large-scale deployment due to their high cost. Wireless communication technology (also commonly referred to as Wi-Fi technology), in addition to providing communication functions, is also suitable for indoor scene perception tasks such as person detection, location, and human gesture recognition. Wi-Fi sensing technology can reuse existing communication frameworks and, based on the propagation characteristics of electromagnetic waves in the deployment environment, detect target status by calculating the multipath propagation characteristics of Wi-Fi signals within the deployment environment. It offers advantages such as all-weather operation, multi-tasking, no direct information collection, and low deployment costs.

[0003] Due to the propagation characteristics of electromagnetic waves, Wi-Fi signals experience multipath propagation within the deployment environment. In indoor environments, reflections from walls, surfaces, the ground, and moving people all cause electromagnetic waves to take different propagation paths. These signals ultimately combine to form the Wi-Fi signal received by the receiver. The receiver measures the received signal to determine the effects of scattering, fading, and power attenuation in the target's environment from transmission to reception. This measured value is called Channel State Information (CSI). If the effects on the signal can be linked to the state of the target, tasks such as target detection and identification can be performed, achieving Wi-Fi awareness.

[0004] For simpler perception tasks, such as intrusion detection, statistical models are often used to achieve better perception performance. Existing technologies have disclosed indoor personnel fall detection methods based on statistical models. For tasks such as indoor personnel behavior recognition, due to the difficulty of statistically modeling the target state, data-driven machine learning or deep learning models are often used. For data-driven models, since the signal received by the perception device receiver is also affected by the deployment environment to which the perception target machine belongs, the differences in the deployment environment, the relative position and orientation of the perception target to the perception device, and the individual differences in the perception target are the parts of the perception model that have the greatest impact on perception precision or accuracy. Existing data-driven models usually adopt a "one scene, one model" experimental setting. Summary of the Invention

[0005] The embodiments of the present disclosure provide a model method, device, electronic device and storage medium for a target perception model.

[0006] In a first aspect, embodiments of the present disclosure provide a method for obtaining a CSI signal in a Wi-Fi environment; the Wi-Fi environment includes a laboratory environment and a deployment environment; the CSI signal in the laboratory environment corresponds to a real label; the CSI signal in the deployment environment does not have a real label; the real label is used to represent target information present in the Wi-Fi environment;

[0007] Calculating multiple spectrograms in multiple time periods corresponding to the CSI signal, each spectrogram serving as an input feature in each of the time periods;

[0008] The input features of the CSI signal in the laboratory environment within the time period and the input features of the CSI signal in the deployment environment within the time period are mixed to form sample data; each sample data is an input feature pair, and the input feature pair includes the data features of the CSI signal in the laboratory environment with a target in the time period and the background noise features of the CSI signal in the deployment environment without a target in the time period, or the data features of the CSI signal in the deployment environment with a target in the time period and the background noise features of the CSI signal in the deployment environment without a target in the time period;

[0009] The input feature pairs are fed into a target perception model, and the target perception model is trained by performing a combination of one or more of the following tasks:

[0010] Task 1: Encode each of the input feature pairs to obtain corresponding spectrogram features, so that the spectrogram features can be decoded to restore the corresponding input features;

[0011] Task 2: Represent the noise floor features corresponding to CSI signals without targets as low-dimensional noise floor vectors, ensuring that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in the same Wi-Fi environment is greater than or equal to a first preset threshold, and that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in different Wi-Fi environments is greater than a second threshold.

[0012] Task 3: Extract domain-invariant features that are independent of the Wi-Fi environment from the input feature pairs;

[0013] Task 4: Identify target information in the CSI signal based on the input feature pair.

[0014] Furthermore, calculating multiple spectrograms in multiple time periods corresponding to the CSI signal, with each spectrogram serving as an input feature in each of the time periods, includes:

[0015] Calculating multiple time series about CSI signal amplitudes based on multiple subcarriers of the CSI signal;

[0016] After performing noise reduction on each of the time series related to the CSI signal amplitude, the time series are cut into time subsequences within multiple time periods of the same length;

[0017] Performing short-time Fourier transform on the time subsequences corresponding to the multiple time series about the CSI signal amplitudes in each time period to obtain multiple spectrograms;

[0018] An average of the multiple spectrograms is taken as the input feature in the time period corresponding to the CSI signal.

[0019] Furthermore, the input features of the CSI signal in the laboratory environment within the time period and the input features of the CSI signal in the deployment environment within the time period are mixed to form sample data, including:

[0020] For the data features of the CSI signal in the laboratory environment that has a true label and indicates the presence of a target, randomly extract the background noise features of the CSI signal in the laboratory environment that has a true label and indicates the absence of a target, to form an input feature pair;

[0021] For the data features of the CSI signal in the deployment environment where the target exists, the background noise features of the CSI signal in the deployment environment where the target does not exist are randomly extracted to form an input feature pair.

[0022] Furthermore, the target perception model includes a shared feature extraction layer, a first codec pair, a second codec pair, a data feature extraction layer, and a background noise feature extraction layer; the first codec pair includes a first encoder and a first decoder, and the second encoder pair includes a second encoder and a second decoder; each of the input feature pairs is encoded to obtain a corresponding spectrogram feature, so that the spectrogram feature can restore the corresponding input feature after decoding, including:

[0023] Inputting the data features and background noise features in the input feature pair into the shared feature extraction layer to extract data shared features and background noise shared features respectively;

[0024] Input the data sharing feature into the first encoder to obtain a latent space data representation vector, and input the background noise sharing feature into the second encoder to obtain a latent space background noise representation vector;

[0025] The latent space data representation vector is input into the first decoder, and the decoding result is input into the data feature extraction layer to obtain the data restoration feature of the data feature. The latent space background noise representation vector is input into the second decoder, and the decoding result is input into the background noise feature extraction layer to obtain the background noise restoration feature of the background noise feature.

[0026] Furthermore, the target perception model also includes a vector dimensionality reduction module; representing the noise floor features corresponding to the CSI signals without the target as low-dimensional noise floor vectors, such that the similarity between the low-dimensional noise floor vectors corresponding to the CSI signals in the same Wi-Fi environment is greater than or equal to a first preset threshold, and the similarity between the low-dimensional noise floor vectors corresponding to the CSI signals in different Wi-Fi environments is greater than a second threshold, including:

[0027] After the latent space background noise representation vector is input into the vector dimensionality reduction module, a low-dimensional background noise vector corresponding to the background noise feature is obtained.

[0028] Furthermore, the target perception model further includes a vector aggregation module; extracting domain-invariant features that are independent of the Wi-Fi environment from the input feature pairs, including:

[0029] After inputting the latent space data representation vector into the vector dimensionality reduction module, a low-dimensional data vector of the data features corresponding to the CSI signal of the target is obtained;

[0030] The low-dimensional background noise vector and the low-dimensional data vector are input into the vector aggregation module to obtain domain-invariant features of the input feature pair that cannot be distinguished as coming from the laboratory environment or the deployment environment.

[0031] Furthermore, identifying target information in the CSI signal based on the input feature pair includes:

[0032] Selecting, from the plurality of domain-invariant features obtained by processing the plurality of input features respectively, a domain-invariant feature corresponding to the input feature of the laboratory environment;

[0033] A prediction result of target information of the CSI signal from a laboratory environment is identified based on the selected domain-invariant features, so as to train the target perception model based on the prediction result and the true label.

[0034] In a second aspect, an embodiment of the present disclosure provides a training device for a target perception model, comprising:

[0035] an acquisition module configured to acquire a CSI signal in a Wi-Fi environment; the Wi-Fi environment includes a laboratory environment and a deployment environment; the CSI signal in the laboratory environment corresponds to a real label; the CSI signal in the deployment environment does not have a real label; the real label is used to represent target information present in the Wi-Fi environment;

[0036] a calculation module configured to calculate a plurality of spectrograms in a plurality of time periods corresponding to the CSI signal, each spectrogram being used as an input feature in each of the time periods;

[0037] a mixing module configured to mix the input features of the CSI signal in the laboratory environment within the time period and the input features of the CSI signal in the deployment environment within the time period to form sample data; each sample data is an input feature pair, the input feature pair including the data features of the CSI signal in the laboratory environment with a target in the time period and the background noise features of the CSI signal in the deployment environment without a target in the time period, or the data features of the CSI signal in the deployment environment with a target in the time period and the background noise features of the CSI signal in the deployment environment without a target in the time period;

[0038] A training module is configured to input the input feature pairs into a target perception model and train the target perception model by performing a combination of one or more of the following tasks:

[0039] Task 1: Encode each of the input feature pairs to obtain corresponding spectrogram features, so that the spectrogram features can be decoded to restore the corresponding input features;

[0040] Task 2: Represent the noise floor features corresponding to CSI signals without targets as low-dimensional noise floor vectors, ensuring that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in the same Wi-Fi environment is greater than or equal to a first preset threshold, and that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in different Wi-Fi environments is greater than a second threshold.

[0041] Task 3: Extract domain-invariant features that are independent of the Wi-Fi environment from the input feature pairs;

[0042] Task 4: Identify target information in the CSI signal based on the input feature pair.

[0043] The functions can be implemented by hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the functions.

[0044] In one possible design, the apparatus includes a memory and a processor. The memory is configured to store one or more computer instructions that enable the apparatus to perform the corresponding method, and the processor is configured to execute the computer instructions stored in the memory. The apparatus may also include a communication interface for communicating with other devices or a communication network.

[0045] In a third aspect, an embodiment of the present disclosure provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method described in any one of the above aspects.

[0046] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium for storing computer instructions used by any of the above-mentioned devices, and when the computer instructions are executed by a processor, they are used to implement the method described in any of the above-mentioned aspects.

[0047] In a fifth aspect, an embodiment of the present disclosure provides a computer program product, which includes computer instructions, and when the computer instructions are executed by a processor, they are used to implement the method described in any of the above aspects.

[0048] The technical solutions provided by the embodiments of the present disclosure may have the following beneficial effects:

[0049] During the training process of the target perception model in the embodiment of the present disclosure, sample data in a laboratory environment with known true labels and data in a deployment environment with unknown true labels are mixed, and then trained using four tasks. This allows the trained target perception model to accurately identify target information in the deployment environment. That is, it utilizes labeled data collected in the laboratory environment to achieve unlabeled learning in the deployment environment and achieve higher prediction accuracy.

[0050] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Other features, objectives and advantages of the present disclosure will become more apparent through the following detailed description of non-limiting embodiments in conjunction with the accompanying drawings. In the accompanying drawings:

[0052] Figure 1 A flowchart illustrating a method for training a target perception model according to an embodiment of the present disclosure is shown;

[0053] Figure 2 A flowchart illustrating a method for training a target perception model according to an embodiment of the present disclosure is shown;

[0054] Figure 3A schematic diagram of the target perception model structure according to an embodiment of the present disclosure is shown;

[0055] Figure 4 A structural block diagram of a training device for an object perception model according to an embodiment of the present disclosure is shown;

[0056] Figure 5 It is a structural diagram of an electronic device suitable for implementing a training method for a target perception model according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0057] Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement them. In addition, for the sake of clarity, parts not related to the description of the exemplary embodiments are omitted in the accompanying drawings.

[0058] In the present disclosure, it should be understood that terms such as "including" or "having" are intended to indicate the presence of features, numbers, steps, behaviors, components, parts, or combinations thereof disclosed in the present specification, and do not exclude the possibility that one or more other features, numbers, steps, behaviors, components, parts, or combinations thereof exist or are added.

[0059] It should also be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present disclosure may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0060] Existing technologies use data-driven models to learn from data collected in specific scenarios to develop a perception model. Specifically, CSI signals collected over a certain period of time and their corresponding true labels are fed into the model as sample data. Based on a specific model update strategy, the model is trained using an appropriate learning algorithm. Finally, the trained model is used to analyze and predict CSI signals collected in deployment scenarios. In particular, due to the powerful data-fitting capabilities of deep learning models, some existing technologies have achieved good performance on Wi-Fi perception tasks using deep learning models. However, these models suffer from poor generalization and are only applicable to specific scenarios.

[0061] Existing data-driven Wi-Fi sensing technologies have high learning and training costs, poor generalization, and many limitations. This is due to the following two reasons:

[0062] Currently, for more complex perception tasks that traditional statistical models struggle to address, Wi-Fi perception technologies mostly employ data-driven models and a "one-model-per-scenario" approach. This is because Wi-Fi propagation signals are affected by the relative position and orientation of the sensing target and device, as well as environmental characteristics. This can overwhelm valid features within the received signal, requiring the model to filter the received signal features. However, the signal features that are effective for a task are not independent of the background environment, resulting in a filtering model often being applicable only to a specific environment or relative location. This results in poor universality and high learning and training costs for current Wi-Fi perception models.

[0063] In some deployment scenarios, the training data required for data-driven models is often unavailable. For example, a fall detection model for vulnerable individuals cannot be deployed due to the prohibitive cost of acquiring CSI signals, which would indicate a vulnerable person falling in a real-world deployment, and the poor generalizability of existing models. Consequently, Wi-Fi sensing technology is only suitable for scenarios where real data and labels can be collected, with environments and sensing targets that are nearly identical to those in the deployment scenario.

[0064] Therefore, to address the problem in existing data-driven models where CSI signals are significantly affected by target-independent features such as the environment, a deep learning neural network structure based on a dual-stream autoencoder is proposed to learn the ambient noise floor and obtain environment-independent features. This allows data collected in a specific deployment environment to drive model learning in other deployment environments, improving the model's generalization and accuracy and reducing the application constraints of Wi-Fi sensing technology.

[0065] The details of the embodiments of the present disclosure are described in detail below through specific examples.

[0066] Figure 1 FIG. 1 is a flow chart showing a method for training a target perception model according to an embodiment of the present disclosure. Figure 1 As shown, the training method of the target perception model includes the following steps:

[0067] In step S101, a CSI signal in a Wi-Fi environment is obtained; the Wi-Fi environment includes a laboratory environment and a deployment environment; the CSI signal in the laboratory environment corresponds to a real label; the CSI signal in the deployment environment does not have a real label; the real label is used to represent target information in the Wi-Fi environment;

[0068] In step S102, a plurality of spectrograms in a plurality of time periods corresponding to the CSI signal are calculated, and each spectrogram is used as an input feature in each of the time periods;

[0069] In step S103, the input features of the CSI signal in the laboratory environment within the time period and the input features of the CSI signal in the deployment environment within the time period are mixed to form sample data; each sample data is an input feature pair, and the input feature pair includes the data features of the CSI signal in the laboratory environment with a target in the time period and the background noise features of the CSI signal in the deployment environment without a target in the time period, or the data features of the CSI signal in the deployment environment with a target in the time period and the background noise features of the CSI signal in the deployment environment without a target in the time period;

[0070] In step S104, the input feature pair is input to a target perception model, and the target perception model is trained by performing a combination of one or more of the following tasks:

[0071] Task 1: Encode each of the input feature pairs to obtain corresponding spectrogram features, so that the spectrogram features can be decoded to restore the corresponding input features;

[0072] Task 2: Represent the noise floor features corresponding to CSI signals without targets as low-dimensional noise floor vectors, ensuring that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in the same Wi-Fi environment is greater than or equal to a first preset threshold, and that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in different Wi-Fi environments is greater than a second threshold.

[0073] Task 3: Extract domain-invariant features that are independent of the Wi-Fi environment from the input feature pairs;

[0074] Task 4: Identify target information in the CSI signal based on the input feature pair.

[0075] In this embodiment, the method can be executed on the server. After the device is initialized, a period of CSI data with real labels and laboratory background noise CSI data corresponding to these data can be randomly selected from the existing laboratory data set. The so-called background noise CSI data is CSI data collected in the absence of a target in the laboratory environment. The randomly selected period of CSI data with real labels has a total length of about 20 seconds, for example. The specific length can be determined based on actual needs and is not specifically limited here. The real label may include but is not limited to whether there is a target in the laboratory environment when the CSI data is collected and other information about the target. Other information about the target may include but is not limited to the number of targets, target classification, target status, etc.

[0076] Afterwards, the device is placed in the deployment environment, allowing it to collect noise-free CSI data for a period of time when no targets are present in the deployment environment. Afterwards, the device can collect CSI data in real-time under real-world conditions in the deployment environment.

[0077] Through the above process, a total of four types of CSI data are obtained, namely: the CSI signal when there is a target in the laboratory environment; the CSI signal when there is no target in the laboratory environment, that is, the background noise data; the CSI signal when there is a target in the deployment environment; the CSI signal when there is no target in the deployment environment, that is, the background noise data. It should be noted that the true label of the CSI signal when there is a target in the laboratory environment is known, while the true label of the CSI signal when there is a target in the deployment environment is unknown. The true label is used to represent the target information existing in the WIFI environment, and the target information may include but is not limited to whether there is a target, the number of targets, the target type, the target status, etc. In some embodiments, the target status may include but is not limited to the location of the target, the target posture, etc.

[0078] In some embodiments, CSI data in laboratory and deployment environments can be understood as follows: Wi-Fi signals are present in these environments. As electromagnetic waves, Wi-Fi signals undergo multipath propagation in these environments. In indoor environments, reflections from walls, surfaces, the ground, and moving human bodies cause the electromagnetic waves to take different propagation paths. These signals ultimately combine to form the Wi-Fi signal received by the receiver. The receiver measures the received Wi-Fi signal to determine the effects of scattering, fading, and power attenuation on the Wi-Fi signal from transmission to reception in the target's environment. This measurement is called Channel State Information (CSI).

[0079] It is understood that the laboratory environment is an existing environment for obtaining the correspondence between real labels and experimental data. In this embodiment, the experimental data is the measured values ​​of the Wi-Fi signal received when the target is present and absent. The deployment environment is a real environment, for example, an environment where target information needs to be continuously detected in subsequent real-life situations. In this deployment environment, the target information is unknown and needs to be detected.

[0080] After the above data are collected, they can be preprocessed. During the data preprocessing process, a spectrum of each CSI signal in multiple time periods can be obtained, and the spectrum is used as the input feature of the CSI signal in the time period.

[0081] Based on the input features, multiple sample data can be generated for model training. Each sample data includes an input feature pair; the input feature pair includes an input feature pair in a laboratory environment or an input feature pair in a deployment environment. The input feature pair in the laboratory environment includes the data features of the CSI signal with a target in the time period and the background noise features of the CSI signal without a target in the time period; the input feature pair in the deployment environment includes the data features of the CSI signal with a target in the time period and the background noise features of the CSI signal without a target in the time period.

[0082] After the above preprocessing, the sample data corresponding to the CSI signal in each time period can be obtained, and the target perception model can be trained based on these sample data. The model training process can be as follows: Figure 2 As shown in the figure, during the training of the target perception model, input feature pairs are input one at a time. These input feature pairs are processed by the target perception model and trained on multiple tasks. After training on a large number of input feature pairs, the target perception model is trained. The trained target perception model can accurately predict the target information in the CSI signal of the target in the input feature pair.

[0083] The tasks of training object perception models include one or more of the following combinations:

[0084] Task 1: Encode each of the input feature pairs to obtain corresponding spectrogram features, so that the spectrogram features can be decoded to restore the corresponding input features;

[0085] Task 2: Represent the noise floor features corresponding to CSI signals without targets as low-dimensional noise floor vectors, ensuring that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in the same Wi-Fi environment is greater than or equal to a first preset threshold, and that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in different Wi-Fi environments is greater than a second threshold.

[0086] Task 3: Extract domain-invariant features that are independent of the Wi-Fi environment from the input feature pairs;

[0087] Task 4: Identify target information in the CSI signal based on the input feature pair.

[0088] Among them, the purpose of Task 1 is that the target perception model can encode the input feature pair, and the encoded spectrum graph features can also be decoded by the target perception model to restore the original input data. That is, the data features corresponding to the CSI signal with the target in the input feature pair are encoded by the target perception model to obtain spectrum graph features. After the spectrum graph features are decoded by the target perception model, the data features corresponding to the CSI signal with the target in the input feature pair can be restored. Similarly, the background noise features corresponding to the CSI signal without the target in the input feature pair are encoded by the target perception model to obtain spectrum graph features. After the spectrum graph features are decoded by the target perception model, the background noise features corresponding to the CSI signal without the target in the input feature pair can be restored.

[0089] The purpose of Task 2 is to represent the noise floor features corresponding to CSI signals without targets as low-dimensional noise floor vectors, so that these low-dimensional noise floor vectors can represent environmental information. Specifically, the low-dimensional noise floor vectors corresponding to CSI signals in the same Wi-Fi environment have high similarity, for example, a similarity greater than or equal to a first preset threshold, while the low-dimensional noise floor vectors corresponding to CSI signals in different Wi-Fi environments have low similarity, for example, a similarity less than or equal to a second preset threshold. Ideally, the low-dimensional noise floor vectors corresponding to CSI signals in the same Wi-Fi environment are closely distributed, while the low-dimensional noise floor vectors corresponding to CSI signals in different Wi-Fi environments are approximately orthogonal.

[0090] The purpose of Task 3 is to extract domain-invariant features from the input feature pairs that are independent of the Wi-Fi environment. In other words, these domain-invariant features do not reflect environmental information, making it impossible for the target perception model to determine which Wi-Fi environment the input feature corresponds to based on the CSI signal.

[0091] The purpose of Task 4 is to identify the target information in the target CSI signal based on the input feature pair.

[0092] During the specific training process of the target perception model, the target perception model is first initialized, and the following steps are implemented cyclically: multiple input feature pairs are taken as an input feature pair batch (for example, 64 input feature pairs in a laboratory environment and 64 input feature pairs in a deployment environment can be used and mixed to generate a batch); after the input feature pairs are input into the target perception model once, gradient backpropagation and model parameter update are performed based on the loss function corresponding to each task; until the model training stop condition is reached and the loop is exited, a target perception model suitable for the current environment can be obtained. Model training stop conditions may include but are not limited to the number of training times exceeding a preset limit and the convergence of model parameters.

[0093] During the training process of the target perception model in the embodiment of the present disclosure, sample data in a laboratory environment with known true labels and data in a deployment environment with unknown true labels are mixed, and then trained using four tasks. This allows the trained target perception model to accurately identify target information in the deployment environment. That is, it utilizes labeled data collected in the laboratory environment to achieve unlabeled learning in the deployment environment and achieve higher prediction accuracy.

[0094] In an optional implementation of this embodiment, step S102, i.e., calculating multiple spectrograms in multiple time periods corresponding to the CSI signal, with each spectrogram serving as an input feature in each of the time periods, further includes the following steps:

[0095] Calculating multiple time series about CSI signal amplitudes based on multiple subcarriers of the CSI signal;

[0096] After performing noise reduction on each of the time series related to the CSI signal amplitude, the time series are cut into time subsequences within multiple time periods of the same length;

[0097] Performing short-time Fourier transform on the time subsequences corresponding to the multiple time series about the CSI signal amplitudes in each time period to obtain multiple frequency spectrograms;

[0098] An average of the multiple spectrograms is taken as the input feature in the time period corresponding to the CSI signal.

[0099] In this optional implementation, it is assumed that each CSI signal corresponds to c subcarriers. The amplitude of the CSI signal of each subcarrier is calculated separately to obtain c time series about the CSI signal amplitude. After the amplitude in the time series is denoised, it is cut using a sliding window with a length of w steps and a step size of s steps. C time subsequences are obtained in each time period. The c time subsequences of length w obtained in the same time period are short-time Fourier transformed to obtain c spectrograms. The c spectrograms are averaged and can be used as input features for the short period of time.

[0100] In an optional implementation of this embodiment, step S103, i.e., calculating multiple spectrograms in multiple time periods corresponding to the CSI signal, with each spectrogram serving as an input feature in each of the time periods, further includes the following steps:

[0101] For the data features of the CSI signal in the laboratory environment that has a true label and indicates the presence of a target, randomly extract the background noise features of the CSI signal in the laboratory environment that has a true label and indicates the absence of a target, to form an input feature pair;

[0102] For the data features of the CSI signal in the deployment environment where the target exists, the background noise features of the CSI signal in the deployment environment where the target does not exist are randomly extracted to form an input feature pair.

[0103] In this optional implementation, after data preprocessing to obtain input features in a laboratory environment and input features in a deployment environment, input feature pairs are formed. One input feature pair can be a sample data.

[0104] For the data feature corresponding to the CSI signal with a target in a laboratory environment, a random noise feature is extracted from the background noise feature in the laboratory environment to form an input feature pair with the data feature. For the data feature corresponding to the CSI signal with a target in a deployment environment, a background noise feature corresponding to the CSI signal without a target in the deployment environment is also randomly extracted to form an input feature pair.

[0105] It should be noted that the CSI signal in the laboratory environment has real labels, and it is possible to determine which are data features and which are background noise features based on the real labels.

[0106] The CSI signals in the deployment environment do not have real labels. However, when collecting data, the CSI signals without targets are collected separately. The input features corresponding to this batch of collected data can be determined to be background noise features. The rest of the data (generally collected when targets are present) can be determined to be CSI data with targets, and its corresponding input features are also data features.

[0107] In an optional implementation of this embodiment, the target perception model includes a shared feature extraction layer, a first codec pair, a second codec pair, a data feature extraction layer, and a background noise feature extraction layer; the first codec pair includes a first encoder and a first decoder, and the second encoder pair includes a second encoder and a second decoder; the step of encoding each of the input feature pairs to obtain corresponding spectrogram features, so that the spectrogram features can restore the corresponding input features after decoding, further includes the following steps:

[0108] Inputting the data features and background noise features in the input feature pair into the shared feature extraction layer to extract data shared features and background noise shared features respectively;

[0109] Input the data sharing feature into the first encoder to obtain a latent space data representation vector, and input the background noise sharing feature into the second encoder to obtain a latent space background noise representation vector;

[0110] The latent space data representation vector is input into the first decoder, and the decoding result is input into the data feature extraction layer to obtain the data restoration feature of the data feature. The latent space background noise representation vector is input into the second decoder, and the decoding result is input into the background noise feature extraction layer to obtain the background noise restoration feature of the background noise feature.

[0111] In this optional implementation, the structure of the target perception model can be as follows Figure 3 As shown. The target perception model includes a shared feature extraction layer, a first codec pair, a second codec pair, a data feature extraction layer, and a background noise feature extraction layer; the first codec pair includes a first encoder and a first decoder, and the second encoder pair includes a second encoder and a second decoder. Figure 3 As shown in FIG, the data input features (i.e., data spectrum s1) and noise input features (i.e., noise spectrum s2) corresponding to the input feature pair in Task 1 (i.e., the data-noise pair mixed in the laboratory environment and the deployment environment) are respectively passed through the shared feature layer to extract the data shared features and noise shared features; the data shared features are encoded by the first encoder to obtain the latent space data representation vector, and the noise shared features are encoded by the second encoder to obtain the latent space noise representation vector.

[0112] After the latent space data representation vector is input into the first decoder, it passes through the data feature extraction layer to obtain the data restoration feature of the data feature. It should be noted that the data restoration feature may be different from the data feature during the model training process, and the purpose of Task 1 in the model training is to ensure that the data restoration feature is the same as the data feature after the model training is completed.

[0113] The latent space noise representation vector is input into the second decoder and then passes through the noise feature extraction layer to obtain the noise restoration feature of the noise feature. It should be noted that the noise restoration feature may be different from the noise feature during the model training process. The purpose of Task 1 in model training is to ensure that the noise restoration feature is the same as the data feature after model training is completed.

[0114] In the algorithm framework, the input of the algorithm in a batch gradient descent is Where b is the pre-set batch size. is the input feature pair in the laboratory environment, y (i) yes The corresponding true label; is the input feature pair in the deployment environment.

[0115] remember Let there be a rearrangement σ(1,2,…,b) of (1,2,…,b).

[0116] Construct source tags And s=(s (k) ) k＝1,2,…,b and l are rearranged according to σ(1,2,…,b) to obtain s σ and l σ .

[0117] For task one, the loss function can be designed:

[0118] When the value of this function approaches 0, the target perception model can accurately restore the corresponding data spectrum or background noise spectrum using the compressed vector. The latent space representation output by the encoder part of the target perception model contains rich information about the spectrum.

[0119] In an optional implementation of this embodiment, the target perception model further includes a vector dimensionality reduction module; the step of representing the noise floor features corresponding to CSI signals without targets as low-dimensional noise floor vectors, such that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in the same Wi-Fi environment is greater than or equal to a first preset threshold, and the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in different Wi-Fi environments is greater than a second threshold, further includes the following steps:

[0120] After the latent space background noise representation vector is input into the vector dimensionality reduction module, a low-dimensional background noise vector corresponding to the background noise feature is obtained.

[0121] In this optional implementation, Figure 3 As shown, the purpose of Task 2 is to train the target perception model so that the low-dimensional background noise vector output by the latent space background noise representation vector after the vector dimensionality reduction module can represent the WIFI environment information of the CSI signal. In some embodiments, the training target of Task 2 is to obtain a low-dimensional background noise vector after the latent space background noise representation vector under the same WIFI environment passes through the vector dimensionality reduction module. Tightly distributed, and the low-dimensional background noise vector E corresponding to different WIFI environments j are mutually orthogonal.

[0122] For task 2, the loss function term can be involved where i represents the environment number to which the low-dimensional noise background vector w belongs, and j∈{1,2,…,J}, where J represents the sum of the number of laboratory environments and the number of deployment environments. sim(·,·) is a vector similarity function, typically set to the cosine of the angle between two vectors. Represents the mean of all low-dimensional noise vectors w for all environments numbered j in the previous optimization round.

[0123] When this loss function value When it approaches 0, the low-dimensional background noise vector corresponding to the CSI signal collected from the WIFI environment numbered i in each time period The set falls on a straight line in the space, and the low-dimensional background noise vector corresponding to the CSI signal collected from other WIFI environments (numbered j, j≠i) in each time period Pairwise orthogonality. This loss function forces the target perception model to output similar low-dimensional noise background vectors in the target vector space when receiving noise background features from similar Wi-Fi environments, while distinguishing between noise background features from different Wi-Fi environments. Specifically, when receiving noise background features from similar Wi-Fi environments, the low-dimensional noise background vectors output in the target vector space are less similar or orthogonal. This task can assist in extracting domain-invariant features in Task 3.

[0124] exist Figure 3 In the contrastive learning module shown, the loss function value corresponding to task two is calculated through the above-mentioned loss function term, and then the model parameters of the target perception model are updated using the loss function value through the gradient descent method.

[0125] In an optional implementation of this embodiment, the target perception model further includes a vector aggregation module; and the step of extracting domain-invariant features that are independent of the Wi-Fi environment from the input feature pairs further includes the following steps:

[0126] After inputting the latent space data representation vector into the vector dimensionality reduction module, a low-dimensional data vector of the data features corresponding to the CSI signal of the target is obtained;

[0127] The low-dimensional background noise vector and the low-dimensional data vector are input into the vector aggregation module to obtain domain-invariant features of the input feature pair that cannot be distinguished as coming from the laboratory environment or the deployment environment.

[0128] In this optional implementation, the target perception model also includes a vector aggregation module. The latent space data representation vector undergoes dimensionality reduction processing in the vector dimensionality reduction module to obtain a low-dimensional data vector. This low-dimensional data vector and the low-dimensional background noise vector are then aggregated in the vector aggregation module to produce domain-invariant features. The purpose of Task 3 is to train the target perception features so that the output domain-invariant features do not include environmental information. The target perception model cannot use these domain-invariant features to identify whether the input features corresponding to the low-dimensional data vector and the low-dimensional background noise vector originate from a laboratory environment or a deployment environment.

[0129] Design loss function for the task case in The target perception model accepts input features s i The target vector set u generated later i, the model output obtained by the discriminator. In Task 3, the purpose of training is to require As large as possible, ensuring that the supervised target perception model outputs target vectors with no significant distributional differences when receiving input features corresponding to CSI signals in a laboratory environment and those in a deployment environment. This ensures that the target vectors output by the target perception model are the domain-invariant features required for Task 3. This allows the target perception model to accurately classify CSI signals in a deployment environment without real labels using CSI signals from a laboratory environment with real labels. Furthermore, Task 3 competes with Task 2 to a certain extent, ensuring that the domain-invariant features obtained by the target perception model are not trivial solutions containing only irrelevant information.

[0130] The discriminator in Task 3 is used to determine whether the domain-invariant feature u originates from the lab environment or the deployment environment. Using the corresponding loss function, the target perception model is ultimately trained so that the discriminator cannot distinguish whether the target vector originates from the lab environment or the deployment environment. The discriminator is actually responsible for calculating the loss function, and the loss function value calculated by the discriminator is used to update the model parameters of the target perception model.

[0131] In an optional implementation of this embodiment, the step of identifying target information in the CSI signal based on the input feature pair further includes the following steps:

[0132] Selecting, from the plurality of domain-invariant features obtained by processing the plurality of input features respectively, a domain-invariant feature corresponding to the input feature of the laboratory environment;

[0133] A prediction result of target information of the CSI signal from a laboratory environment is identified based on the selected domain-invariant features, so as to train the target perception model based on the prediction result and the true label.

[0134] In this optional implementation, the purpose of Task 4 is to identify target information in the CSI signal corresponding to the input features. During training, multiple domain-invariant features corresponding to multiple input feature pairs within a batch are obtained. Domain-invariant features with true labels from a laboratory environment are selected from these domain-invariant features. Predictions of target information in CSI signals from the laboratory environment are then identified based on these selected domain-invariant features. The difference between these predictions and the true labels is used to update the model parameters of the target perception model.

[0135] In some embodiments, after obtaining domain-invariant features corresponding to b input features in a batch gradient descent, only b1 vectors with l=1 (i.e., from the laboratory environment) are taken out and the cross entropy classification loss is calculated according to the true label y. The loss function term can be constructed as

[0136] The classifier in Task 4 is used to classify the domain-invariant features corresponding to the CSI signal from the laboratory environment and determine target information, such as whether a target exists, the type of target, and the posture of the target.

[0137] In some embodiments, when using the above tasks 1 to 4 to train the target perception model, the loss function can be set to

[0138] The following are embodiments of the apparatus disclosed herein, which can be used to execute embodiments of the method disclosed herein.

[0139] Figure 4 The structure block diagram of the target perception model training device according to one embodiment of the present disclosure is shown. The device can be implemented as part or all of an electronic device through software, hardware, or a combination of both. Figure 4 As shown, the training device of the target perception model includes:

[0140] Acquisition module 401 is configured to acquire CSI signals in a Wi-Fi environment; the Wi-Fi environment includes a laboratory environment and a deployment environment; the CSI signals in the laboratory environment correspond to real labels; the CSI signals in the deployment environment do not have real labels; the real labels are used to represent target information present in the Wi-Fi environment;

[0141] A calculation module 402 is configured to calculate a plurality of spectrograms in a plurality of time periods corresponding to the CSI signal, each spectrogram being used as an input feature in each of the time periods;

[0142] The mixing module 403 is configured to mix the input features of the CSI signal in the laboratory environment within the time period and the input features of the CSI signal in the deployment environment within the time period to form sample data; each sample data is an input feature pair, the input feature pair including the data features of the CSI signal in the laboratory environment with a target in the time period and the background noise features of the CSI signal in the deployment environment without a target in the time period, or the data features of the CSI signal in the deployment environment with a target in the time period and the background noise features of the CSI signal in the deployment environment without a target in the time period;

[0143] The training module 404 is configured to input the input feature pairs into the object perception model and train the object perception model by performing a combination of one or more of the following tasks:

[0144] Task 1: Encode each of the input feature pairs to obtain corresponding spectrogram features, so that the spectrogram features can be decoded to restore the corresponding input features;

[0145] Task 2: Represent the noise floor features corresponding to CSI signals without targets as low-dimensional noise floor vectors, ensuring that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in the same Wi-Fi environment is greater than or equal to a first preset threshold, and that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in different Wi-Fi environments is greater than a second threshold.

[0146] Task 3: Extract domain-invariant features that are independent of the Wi-Fi environment from the input feature pairs;

[0147] Task 4: Identify target information in the CSI signal based on the input feature pair.

[0148] The training device for the target perception model in this embodiment corresponds to the training method for the target perception model described above. For specific details, please refer to the description of the training method for the target perception model described above, which will not be repeated here.

[0149] Figure 5 It is a structural diagram of an electronic device suitable for implementing a training method for a target perception model according to an embodiment of the present disclosure.

[0150] like Figure 5 As shown, the electronic device 500 includes a processing unit 501, which can be implemented as a processing unit such as a CPU, a GPU, an FPGA, an NPU, etc. The processing unit 501 can perform various processes in the embodiment of any of the above methods of the present disclosure according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage portion 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic device 500 are also stored. The processing unit 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0151] The following components are connected to the I / O interface 505: an input section 506 including a keyboard, a mouse, and the like; an output section 507 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 508 including a hard disk; and a communication section 509 including a network interface card such as a LAN card or a modem. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the I / O interface 505 as needed. A removable medium 511, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 510 as needed, so that computer programs read therefrom can be installed into the storage section 508 as needed.

[0152] In particular, according to embodiments of the present disclosure, any of the methods described above with reference to the embodiments of the present disclosure may be implemented as a computer software program. For example, embodiments of the present disclosure include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program comprising program code for executing any of the methods described in the embodiments of the present disclosure. In such embodiments, the computer program may be downloaded and installed from a network via the communication portion 509 and / or installed from a removable medium 511.

[0153] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment or part of code, and the module, program segment or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0154] The units or modules described in the embodiments of the present disclosure may be implemented in software or hardware. The units or modules described may also be provided in a processor, and the names of these units or modules do not, in certain circumstances, limit the units or modules themselves.

[0155] As another aspect, the present disclosure further provides a computer-readable storage medium. This computer-readable storage medium may be included in the apparatus described in the above embodiments, or may be a standalone computer-readable storage medium not incorporated into the apparatus. The computer-readable storage medium stores one or more programs, which are used by one or more processors to execute the methods described in the present disclosure.

[0156] The above description is merely a preferred embodiment of the present disclosure and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention herein is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also encompasses other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the inventive concept. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this disclosure.

Claims

1. A method for training a target perception model, wherein: include: Get the CSI signal in the WIFI environment; The WIFI environment includes a laboratory environment and a deployment environment; the CSI signal in the laboratory environment corresponds to a real label; the CSI signal in the deployment environment does not have a real label; the real label is used to represent the target information in the WIFI environment; Calculating multiple spectrograms in multiple time periods corresponding to the CSI signal, each spectrogram serving as an input feature in each of the time periods; The input features of the CSI signal in the laboratory environment within the time period and the input features of the CSI signal in the deployment environment within the time period are mixed to form sample data; each sample data is an input feature pair, and the input feature pair includes the data features of the CSI signal in the laboratory environment with a target in the time period and the background noise features of the CSI signal in the deployment environment without a target in the time period, or the data features of the CSI signal in the deployment environment with a target in the time period and the background noise features of the CSI signal in the deployment environment without a target in the time period; The input feature pairs are fed into a target perception model, and the target perception model is trained by performing a combination of one or more of the following tasks: Task 1: Encode each of the input feature pairs to obtain corresponding spectrogram features, so that the spectrogram features can be decoded to restore the corresponding input features; Task 2: Represent the noise floor features corresponding to CSI signals without targets as low-dimensional noise floor vectors, ensuring that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in the same Wi-Fi environment is greater than or equal to a first preset threshold, and that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in different Wi-Fi environments is greater than a second threshold. Task 3: Extract domain-invariant features that are independent of the Wi-Fi environment from the input feature pairs; Task 4: Identify target information in the CSI signal based on the input feature pair.

2. The method according to claim 1, wherein Calculating multiple spectrograms in multiple time periods corresponding to the CSI signal, each spectrogram serving as an input feature in each of the time periods, including: Calculating multiple time series about CSI signal amplitudes based on multiple subcarriers of the CSI signal; After performing noise reduction on each of the time series related to the CSI signal amplitude, the time series are cut into time subsequences within multiple time periods of the same length; Performing short-time Fourier transform on the time subsequences corresponding to the multiple time series about the CSI signal amplitudes in each time period to obtain multiple spectrograms; An average of the multiple spectrograms is taken as the input feature in the time period corresponding to the CSI signal.

3. The method according to claim 1 or 2, wherein: Mixing the input features of the CSI signal in the laboratory environment within the time period and the input features of the CSI signal in the deployment environment within the time period to form sample data includes: For the data features of the CSI signal in the laboratory environment that has a true label and indicates the presence of a target, randomly extract the background noise features of the CSI signal in the laboratory environment that has a true label and indicates the absence of a target, to form an input feature pair; For the data features of the CSI signal in the deployment environment where the target exists, the background noise features of the CSI signal in the deployment environment where the target does not exist are randomly extracted to form an input feature pair.

4. The method according to claim 1 or 2, wherein: The target perception model includes a shared feature extraction layer, a first codec pair, a second codec pair, a data feature extraction layer, and a background noise feature extraction layer; the first codec pair includes a first encoder and a first decoder, and the second codec pair includes a second encoder and a second decoder; each of the input feature pairs is encoded to obtain a corresponding spectrogram feature, so that the spectrogram feature can be restored to the corresponding input feature after decoding, including: Inputting the data features and background noise features in the input feature pair into the shared feature extraction layer to extract data shared features and background noise shared features respectively; Input the data sharing feature into the first encoder to obtain a latent space data representation vector, and input the background noise sharing feature into the second encoder to obtain a latent space background noise representation vector; The latent space data representation vector is input into the first decoder, and the decoding result is input into the data feature extraction layer to obtain the data restoration feature of the data feature. The latent space background noise representation vector is input into the second decoder, and the decoding result is input into the background noise feature extraction layer to obtain the background noise restoration feature of the background noise feature.

5. The method according to claim 4, wherein The target perception model also includes a vector dimensionality reduction module; representing the noise floor features corresponding to the CSI signals without the target as low-dimensional noise floor vectors, so that the similarity between the low-dimensional noise floor vectors corresponding to the CSI signals in the same Wi-Fi environment is greater than or equal to a first preset threshold, and the similarity between the low-dimensional noise floor vectors corresponding to the CSI signals in different Wi-Fi environments is greater than a second threshold, including: After the latent space background noise representation vector is input into the vector dimensionality reduction module, a low-dimensional background noise vector corresponding to the background noise feature is obtained.

6. The method according to claim 5, wherein: The target perception model further includes a vector aggregation module; extracting domain-invariant features that are independent of the Wi-Fi environment from the input feature pairs, including: After inputting the latent space data representation vector into the vector dimensionality reduction module, a low-dimensional data vector of the data features corresponding to the CSI signal of the target is obtained; The low-dimensional background noise vector and the low-dimensional data vector are input into the vector aggregation module to obtain domain-invariant features of the input feature pair that cannot be distinguished as coming from the laboratory environment or the deployment environment.

7. The method according to claim 6, wherein: Identifying target information in the CSI signal based on the input feature pair includes: Selecting, from the plurality of domain-invariant features obtained by processing the plurality of input features respectively, a domain-invariant feature corresponding to the input feature of the laboratory environment; A prediction result of target information of the CSI signal from a laboratory environment is identified based on the selected domain-invariant features, so as to train the target perception model based on the prediction result and the true label.

8. A training device for a target perception model, wherein: include: An acquisition module is configured to acquire a CSI signal in a Wi-Fi environment; The WIFI environment includes a laboratory environment and a deployment environment; the CSI signal in the laboratory environment corresponds to a real label; the CSI signal in the deployment environment does not have a real label; the real label is used to represent the target information in the WIFI environment; a calculation module configured to calculate a plurality of spectrograms in a plurality of time periods corresponding to the CSI signal, each spectrogram being used as an input feature in each of the time periods; a mixing module configured to mix the input features of the CSI signal in the laboratory environment within the time period and the input features of the CSI signal in the deployment environment within the time period to form sample data; each sample data is an input feature pair, the input feature pair including the data features of the CSI signal in the laboratory environment with a target in the time period and the background noise features of the CSI signal in the deployment environment without a target in the time period, or the data features of the CSI signal in the deployment environment with a target in the time period and the background noise features of the CSI signal in the deployment environment without a target in the time period; A training module is configured to input the input feature pairs into a target perception model and train the target perception model by performing a combination of one or more of the following tasks: Task 1: Encode each of the input feature pairs to obtain corresponding spectrogram features, so that the spectrogram features can be decoded to restore the corresponding input features; Task 2: Represent the noise floor features corresponding to CSI signals without targets as low-dimensional noise floor vectors, ensuring that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in the same Wi-Fi environment is greater than or equal to a first preset threshold, and that the similarity between the low-dimensional noise floor vectors corresponding to CSI signals in different Wi-Fi environments is greater than a second threshold. Task 3: Extract domain-invariant features that are independent of the Wi-Fi environment from the input feature pairs; Task 4: Identify target information in the CSI signal based on the input feature pair.

9. An electronic device, wherein: The method comprises a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer instructions stored thereon, wherein: When the computer instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images