Wireless human body action recognition method and device, equipment, storage medium and product

Abstract

The invention discloses a wireless human body action recognition method and device, equipment, a storage medium and a product, and relates to the technical field of wireless communication, and the wireless human body action recognition method comprises the steps: responding to a recognition starting signal, and obtaining wireless channel state information data; performing diffusion restoration on the wireless channel state information data according to a preset diffusion model to obtain restored channel state information data; and inputting the wireless channel state information data and the recovered channel state information data into a preset human body action recognition model to obtain a human body action recognition result. According to the method, when the identification starting signal is received, the diffusion model is adopted to perform diffusion restoration on the wireless channel state information data, so that sample data enhancement is realized, action identification is performed through the trained human body action identification model, and the identification effect of human body action identification is improved; and finally, the accuracy of human body action recognition is improved.

Description

Technical Field

[0001] The present application relates to the field of wireless communication technology, and in particular to wireless human motion recognition methods, devices, equipment, storage media and products. Background Art

[0002] With the continuous advancement of science and technology, wireless human motion recognition has gradually become a popular research direction. In wireless human motion recognition technology, CSI (Channel State Information) is often used to evaluate and optimize wireless communication performance. By collecting the wireless channel status in real time, it can provide necessary feedback for the wireless communication system.

[0003] However, due to force majeure environmental factors, such as radio interference and object occlusion, missing values ​​and interference data are prone to appear in the actual collected CSI data. The appearance of missing values ​​damages the data integrity and cannot fully and accurately reflect the channel status. The introduction of interference data further deteriorates the data quality and reduces the accuracy of human motion recognition, resulting in low accuracy of wireless human motion recognition based on CSI.

[0004] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of this application is to provide a wireless human motion recognition method, device, equipment, storage medium and product, aiming to solve the technical problem of low accuracy of wireless human motion recognition.

[0006] To achieve the above objectives, the present application proposes a wireless human motion recognition method, the method comprising: In response to the start identification signal, acquiring wireless channel state information data; Performing diffusion restoration on the wireless channel state information data according to a preset diffusion model to obtain restored channel state information data; The wireless channel state information data and the restored channel state information data are input into a preset human motion recognition model to obtain a human motion recognition result.

[0007] In one embodiment, the step of performing diffusion restoration on the wireless channel state information data according to a preset diffusion model to obtain restored channel state information data includes: Preprocessing the wireless channel state information data to obtain original state information data; According to the diffusion model, the original state information data is diffused and restored to obtain restored data, and the corresponding time step is updated; Determine whether the time step satisfies a preset termination condition; if not, update the original state information data based on the restored data, return to the step of performing diffusion restoration on the original state information data according to the diffusion model; if satisfied, output the restored data as restored channel state information data.

[0008] In one embodiment, the step of performing diffusion restoration on the original state information data according to the diffusion model to obtain restored data and updating the corresponding time step includes: Get the original time step sequence; According to a preset time step subsequence extraction rule, subsequence extraction is performed on the original time step sequence to obtain a target time step subsequence; Sampling the original state information data according to a preset sampling rule to obtain sampled state information data; The sampled state information data is diffused and restored to obtain restored data, and the corresponding time step is updated.

[0009] In one embodiment, the step of preprocessing the wireless channel state information data to obtain original state information data includes: Performing missing value processing on the wireless channel state information data according to a preset first ratio to obtain a first preprocessing result; The predicted values ​​of the first preprocessing result are randomly set to zero according to a preset second ratio to obtain original state information data.

[0010] In one embodiment, before the step of inputting the wireless channel state information data and the restored channel state information data into a preset human motion recognition model to obtain a human motion recognition result, the step includes: Obtaining wireless channel state data samples; Discretize the wireless channel state data samples to obtain secondary state data samples; Performing diffusion restoration on the secondary state data sample through the diffusion model to obtain a secondary restored state data sample; Binding the secondary state data sample with the secondary restored state data sample to obtain a positive sample pair; Extracting features of the positive sample pair to obtain similar feature representations; Binding data samples of the wireless channel state data samples excluding the secondary state data samples into a negative sample pair; Extracting features of the negative sample pair to obtain a dissimilar feature representation; Inputting the secondary state data sample and the secondary restored state data sample into a preset deep neural network to be trained to obtain a human action recognition result, wherein the deep neural network to be trained is obtained by training with a preset contrast loss function to minimize the distance between similar feature representations and maximize the distance between dissimilar feature representations; Determine whether the action recognition result meets the preset loss convergence condition. If not, return to the step of inputting the secondary state data sample and the secondary recovery state data sample into the preset deep neural network to be trained to obtain the human action recognition result, until the action recognition result meets the loss convergence condition to obtain the human action recognition model.

[0011] In one embodiment, the step of inputting the secondary state data sample and the secondary restored state data sample into a preset deep neural network to be trained to obtain a human motion recognition result comprises: Extracting local information and spatial structure from the secondary state data sample and the secondary restored state data sample through a preset one-dimensional neural network to obtain low-level features; Based on the self-attention mechanism, the global relationship and long-distance dependency in the low-level features are modeled to obtain high-level features; The high-level features are classified and normalized to obtain a human action recognition result.

[0012] In addition, to achieve the above-mentioned purpose, the present application also proposes a wireless human motion recognition device, the wireless human motion recognition device comprising: an acquisition module, for acquiring wireless channel state information data in response to a start recognition signal; A diffusion module, used for performing diffusion restoration on the wireless channel state information data according to a preset diffusion model to obtain restored channel state information data; The recognition module is used to input the wireless channel state information data and the restored channel state information data into a preset human motion recognition model to obtain a human motion recognition result.

[0013] In addition, to achieve the above-mentioned purpose, the present application also proposes a wireless human motion recognition device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the wireless human motion recognition method as described above.

[0014] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the wireless human motion recognition method described above are implemented.

[0015] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the wireless human motion recognition method described above are implemented.

[0016] One or more technical solutions proposed in this application have at least the following technical effects: Compared with the related art, due to force majeure environmental factors, missing values ​​and interference data are prone to appear in the channel state information data actually collected, which damages the data integrity and cannot fully and accurately reflect the channel state, resulting in low accuracy of wireless human motion recognition. In response to the start recognition signal, the present application obtains wireless channel state information data; according to a preset diffusion model, the wireless channel state information data is diffusely restored to obtain restored channel state information data; the wireless channel state information data and the restored channel state information data are input into a preset human motion recognition model to obtain a human motion recognition result. It is understandable that when the present application receives the start recognition signal, the diffusion model is used to diffusely restore the wireless channel state information data to achieve sample data enhancement, and the wireless channel state information data and the restored channel state information data are used to perform motion recognition through the trained human motion recognition model, so as to improve the recognition effect of human motion recognition and finally improve the accuracy of human motion recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0019] Figure 1 A flowchart of the first embodiment of the wireless human motion recognition method of the present application is provided; Figure 2 A schematic diagram provided for Embodiment 2 of the wireless human motion recognition method of the present application; Figure 3 A brief flowchart of the wireless human motion recognition method provided in Example 2 of the present application; Figure 4 This is a schematic diagram of the module structure of the wireless human motion recognition device according to an embodiment of the present application; Figure 5Schematic diagram of the device structure of the hardware operating environment involved in the wireless human motion recognition method in the embodiment of the present application.

[0020] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0021] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0022] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0023] The main solutions of the embodiments of this application are: In response to the start identification signal, acquiring wireless channel state information data; Performing diffusion restoration on the wireless channel state information data according to a preset diffusion model to obtain restored channel state information data; The wireless channel state information data and the restored channel state information data are input into a preset human motion recognition model to obtain a human motion recognition result.

[0024] In this embodiment, the present application uses a wireless human motion recognition device as the execution subject. For ease of description, it is specifically described below as a "device".

[0025] The existing technology uses CSI (Channel State Information) to evaluate and optimize wireless communication performance. CSI can provide necessary feedback to the wireless communication system by collecting the wireless channel status in real time. However, due to force majeure environmental factors, such as radio interference and object obstruction, missing values ​​and interference data are prone to appear in the CSI data actually collected. The appearance of missing values ​​impairs the integrity of the data and cannot fully and accurately reflect the channel status. The introduction of interference data further deteriorates the data quality and reduces the accuracy of human motion recognition, resulting in low accuracy of wireless human motion recognition based on CSI.

[0026] The present application provides a solution. When a start recognition signal is received, a diffusion model is used to perform diffusion restoration on the wireless channel state information data to achieve sample data enhancement. The trained human motion recognition model is used to perform motion recognition on the wireless channel state information data and the restored channel state information data to achieve improved recognition effect of human motion recognition, and ultimately improve the accuracy of human motion recognition.

[0027] Based on this, the present application embodiment provides a wireless human motion recognition method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the wireless human motion recognition method of the present application.

[0028] In this embodiment, the wireless human motion recognition method includes steps S10 to S30: Step S10, in response to the start identification signal, obtaining wireless channel state information data; It should be noted that responding to the start recognition signal is a trigger mechanism. When the system receives a specific start recognition signal, it will start the human motion recognition process. This signal is a software instruction, hardware trigger or user operation result. Wireless channel state information data is data that describes the characteristics of the wireless channel, including signal attenuation, reflection, multipath effect, etc., including but not limited to channel state information (CSI).

[0029] It can be understood that, in human motion recognition, obtaining wireless channel state information data can be used to capture human motion and position information.

[0030] Step S20, performing diffusion restoration on the wireless channel state information data according to a preset diffusion model to obtain restored channel state information data; It should be noted that the preset diffusion model is a mathematical model or algorithm for processing and restoring wireless channel state information data. The diffusion model refers to a model that simulates the physical diffusion process, and is used here to refer to an algorithm that can process wireless channel state information data to better reflect the characteristics of human body movements.

[0031] Diffusion restoration involves using a diffusion model to process wireless channel state information data to restore or enhance the information in the original signal that is distorted due to the influence of the wireless propagation environment. The purpose of restoration is to obtain more accurate wireless channel state information data for human action recognition.

[0032] The restored channel state information data is the CSI data after diffusion restoration processing. These data are considered to be closer to the original, undisturbed human motion information, and are therefore more suitable for subsequent motion recognition processes.

[0033] It is understandable that the purpose of processing the original wireless channel state information data using a preset diffusion model is to restore or enhance this data. Since the wireless channel state information data is affected by factors such as multipath effects and signal attenuation during transmission, data distortion occurs. The purpose of diffusion restoration is to restore the original state of these data as much as possible in order to more accurately capture human motion information. The diffusion model also enhances the features of the wireless channel state information data, making the features of human motion more obvious, thereby improving the performance of subsequent recognition models. The restored wireless channel state information data provides higher quality input data for the next step of human motion recognition, which is the basis for achieving high-accuracy recognition.

[0034] Step S30, inputting the wireless channel state information data and the restored channel state information data into a preset human motion recognition model to obtain a human motion recognition result.

[0035] It should be noted that the human motion recognition model is a machine learning or deep learning model specifically designed to recognize and classify human motions from wireless channel state information data. The model can extract features from the restored wireless channel state information data and map them to specific motion categories. The human motion recognition result is the output of the human motion recognition model, which represents the model's prediction of the human motion recognized by the input wireless channel state information data. These results can be used in various applications, such as smart home control, health monitoring, or security monitoring.

[0036] Exemplarily, the original wireless channel state information data and the restored wireless channel state information data are provided as input to a preset human motion recognition model. This dual-input design is to utilize the difference between the original data and the restored data to increase the model's ability to capture motion features. Through the idea of ​​comparative learning, the model will consider both the original wireless channel state information data and the restored wireless channel state information data to achieve feature fusion. This fusion helps the model understand human motion from different perspectives and improve the robustness and accuracy of recognition. The human motion recognition model analyzes the input wireless channel state information data, extracts key features, maps them to specific human motion categories, and finally outputs human motion recognition results. The obtained human motion recognition results are the direct output of the model, which can be used in various application scenarios, such as automated control, health monitoring, etc.

[0037] This embodiment provides a wireless human motion recognition method. When a start recognition signal is received, a diffusion model is used to perform diffusion restoration on the wireless channel state information data to achieve sample data enhancement. The trained human motion recognition model is used to perform motion recognition on the wireless channel state information data and the restored channel state information data to achieve improved recognition effect of human motion recognition, and ultimately improve the accuracy of human motion recognition.

[0038] In a feasible implementation, step S20 may include: Preprocessing the wireless channel state information data to obtain original state information data; According to the diffusion model, the original state information data is diffused and restored to obtain restored data, and the corresponding time step is updated; Determine whether the time step satisfies a preset termination condition; if not, update the original state information data based on the restored data, return to the step of performing diffusion restoration on the original state information data according to the diffusion model; if satisfied, output the restored data as restored channel state information data.

[0039] It should be noted that the wireless channel state information data is data captured by the wireless communication system, which reflects the channel characteristics of the signal during transmission, such as signal strength, phase, time delay, etc. In human motion recognition, these data may contain information about channel changes caused by human motion. Preprocessing refers to a series of processing steps performed on the original wireless channel state information data before inputting the data into the model to improve the data quality or make it more suitable for subsequent analysis, including denoising, normalization, feature extraction and other operations. The original state information data is the wireless channel state information data obtained after preprocessing, which is called the original state information data. These data are the results of the preprocessing steps and provide a basis for the next step of diffusion restoration. The diffusion model is a mathematical or computational model used to simulate and process the diffusion process in the wireless channel state information data. In human motion recognition, the diffusion model is used to restore data damaged by interference or distortion during signal propagation. Diffusion restoration refers to the use of the diffusion model to process the original state information data to restore or enhance the useful information in the data. It aims to improve the accuracy and reliability of the data so as to better identify human motion. The restored data is the data obtained after the diffusion restoration process, which is called the restored data. These data are considered to be closer to the original, undisturbed human motion information and are therefore more suitable for the subsequent action recognition process.

[0040] A time step refers to a specific time point or time period in the processing process. In the process of diffusion restoration, each time step corresponds to an update or processing cycle of the data. The preset termination condition is a condition set in the algorithm or model to determine when to stop the processing process. The termination condition includes but is not limited to reaching a certain restoration quality, processing time or number of iterations. The restored channel state information data is the final output restoration data when the preset termination condition is met. It is called the restored channel state information data. These data are the final results of the diffusion restoration process and can be used as the input of the human action recognition model to achieve high-accuracy recognition.

[0041] It can be understood that in order to generate the missing WiFi CSI value in the real sampling environment, the diffusion model SDM proposed in this application will scale the original CSI data of the target to The missing values ​​are randomly set to 0. The time dimension of the later time dimension of the randomly selected CSI samples is set to 0 for use in predicting the true value.

[0042] After training, data sampling can be performed time-step by time-step to slowly restore the CSI data. Therefore, the model sampling process based on this framework can be formally described as follows.

[0043] Algorithm 1: CSI synthesis.

[0044] Input: Preprocessing stage according to Missing values ​​or The original data sample with the predicted value randomly set to 0 .

[0045] Output: Synthesized restored / enhanced samples .

[0046] 1: Randomly extract noise from a Gaussian distribution ; 2: for time step do; 3: Encode the label and time step to obtain embedded features ; 4: Randomly extract noise from Gaussian distribution ; 5: Gradually restore CSI;

[0047] in, ,variance , , ,noise , time step, , is the restored / enhanced sample recovered from the extracted noise at the corresponding time step, represents the denoising function, represents the sample from time step t The noise is estimated.

[0048] 6: end for.

[0049] For example, assume that the device collects CSI data from a wireless router, which contains signal strength and phase changes caused by people walking around the room. The preprocessing steps include removing background noise, adjusting the amplitude range of the signal, and converting the signal into a format suitable for model processing. After these steps, the device obtains "raw state information data", which is CSI data that is clean and ready for further analysis. Next, the device uses a diffusion model to process these raw CSI data. This model is a deep learning network that can simulate the propagation of signals in space and try to restore signals distorted by multipath effects and other wireless channel interference. After model processing, the device obtains "restored data", which more accurately reflects the signal changes caused by human motion. At the same time, the model records the current processing time step. The device sets a termination condition, such as the number of model iterations reaches 100. If the current time step has not met these conditions, the device will continue to use the diffusion model to process new CSI data and update the time step. If the time step reaches the preset termination condition, the device stops processing and outputs the final restored data as "restored channel state information data". These data will be used for the next step of human action recognition analysis.

[0050] In this example, the device obtains high-quality CSI data through preprocessing, diffusion restoration and iterative updating, which can be used to accurately identify and analyze human movements. This process shows how to achieve non-invasive monitoring of human movements through changes in wireless signals.

[0051] In a feasible implementation manner, the step of performing diffusion restoration on the original state information data according to the diffusion model to obtain restored data and updating the corresponding time step includes: Get the original time step sequence; According to a preset time step subsequence extraction rule, subsequence extraction is performed on the original time step sequence to obtain a target time step subsequence; Sampling the original state information data according to a preset sampling rule to obtain sampled state information data; The sampled state information data is diffused and restored to obtain restored data, and the corresponding time step is updated.

[0052] It should be noted that the original time step sequence refers to the sequence of time steps originally set. The preset time step subsequence extraction rules are a set of predefined rules or criteria for selecting specific time steps or data segments from the original time step sequence. These rules may be based on time intervals, thresholds of signal changes, or other specific features, with the goal of extracting a data subset that is particularly useful for human action recognition.

[0053] The target time step subsequence is a data subset extracted from the original time step sequence according to the preset time step subsequence extraction rules. This subsequence contains time steps that are particularly meaningful for identifying human actions, probably because these time steps contain significant signal changes caused by human actions.

[0054] The preset sampling rules are a set of rules used to determine how to select samples from the original state information data. The sampling rules may involve sampling frequency, sampling point selection criteria, etc. The purpose is to select representative samples from a large amount of data to reduce the amount of data and improve processing efficiency.

[0055] The sampled state information data is sample data selected from the original state information data according to the preset sampling rules. These data are screened and represent the key information in the original data set, which is used in the subsequent diffusion restoration and action recognition process.

[0056] Diffusion restoration is a processing step that uses a specific model or algorithm to process the sampled state information data to restore or enhance the useful information in the data. In human action recognition, diffusion restoration is used to reduce noise, correct signal distortion, or enhance the signal change characteristics caused by human action.

[0057] The restored data is the enhanced or recovered data obtained after the diffusion restoration process. These data are considered to more accurately reflect the signal changes caused by human motion, and are therefore more suitable for subsequent human motion recognition analysis. The corresponding time step is updated during the diffusion restoration process, and the data of each time step will be updated to reflect the latest restoration results. This step ensures the temporal continuity and consistency of the data, which is particularly important for time series analysis.

[0058] In this implementation, data acquisition, extraction, sampling, restoration and time step updating are involved, which constitute the key steps in the wireless human motion recognition process.

[0059] For example, it is easy to know from the framework of Algorithm 1 that the diffusion model needs to use the Markov chain idea to gradually restore the data, and the computational overhead of the sampling synthesis CSI process is positively correlated with the total number of diffusion time steps. For example, if the number of diffusion steps is 1000, then generating one sample requires 1000 model inferences. The present application proposes a relaxed sampling condition so that the model accelerates sampling under non-Markov chain conditions and obtains the final enhanced data. Specifically, a sampling process similar to the sampling process of the diffusion model framework of step-by-step sampling is first introduced, and then the assumption is relaxed, and the formula is not required to conform to the Markov property, but the sampling process is in the form shown in formula (1).

[0060] (1)

[0061] Among them, formula (1) strictly satisfies , represents the denoising function, , yes The coefficient of It's noise.

[0062] From the original sequence of time steps Define a length of subsequence of Iterate sampling repeatedly through formula (1) You can get .make , .when When , the last term in formula (1) will also become 0, so the sampling process becomes a deterministic process and is not random.

[0063] It can be seen that the model can be used without affecting the boundary distribution In the context of In other words, the model can skip several time steps to speed up the generation process. In the actual implementation process, the subsequence values ​​of the overall sampling steps are often selected empirically. After repeated experiments, the final deterministic sampling subsequence is determined in this application. The empirical value is 10. After the optimization, the process of SDM sampling and restoring CSI data can be described as shown in Algorithm 2.

[0064] Algorithm 2 SDM model synthesis CSI Input: Preprocessing stage according to Missing values ​​or The original data sample with the predicted value randomly set to 0

[0065] Output: Synthesized restored / enhanced samples

[0066] 1: Randomly extract noise from a Gaussian distribution ; 2: for time step subsequence do; 3: Label With time step Encoding, obtaining embedded features ; 4: Randomly extract noise from Gaussian distribution ; 5: Recover CSI using Equation (1) by skipping time steps;

[0067] Among them, formula (1) strictly satisfies , represents the denoising function, , yes The coefficient of noise .

[0068] 6: end for.

[0069] In a feasible implementation manner, the step of preprocessing the wireless channel state information data to obtain original state information data includes: Performing missing value processing on the wireless channel state information data according to a preset first ratio to obtain a first preprocessing result; The predicted values ​​of the first preprocessing result are randomly set to zero according to a preset second ratio to obtain original state information data.

[0070] It can be understood that in order to generate the missing WiFi CSI value in the real sampling environment, the diffusion model SDM proposed in this application will scale the original CSI data of the target to The missing values ​​are randomly set to 0. The time dimension of the later time dimension of the randomly selected CSI samples is set to 0 for use in predicting the true value.

[0071] In a feasible implementation manner, before the step of inputting the wireless channel state information data and the restored channel state information data into a preset human motion recognition model to obtain a human motion recognition result, the step includes: Obtaining wireless channel state data samples; Discretize the wireless channel state data samples to obtain secondary state data samples; Performing diffusion restoration on the secondary state data sample through the diffusion model to obtain a secondary restored state data sample; Binding the secondary state data sample with the secondary restored state data sample to obtain a positive sample pair; Extracting features of the positive sample pair to obtain similar feature representations; Binding data samples of the wireless channel state data samples excluding the secondary state data samples into a negative sample pair; Extracting features of the negative sample pair to obtain a dissimilar feature representation; Inputting the secondary state data sample and the secondary restored state data sample into a preset deep neural network to be trained to obtain a human action recognition result, wherein the deep neural network to be trained is obtained by training with a preset contrast loss function to minimize the distance between similar feature representations and maximize the distance between dissimilar feature representations; Determine whether the action recognition result meets the preset loss convergence condition. If not, return to the step of inputting the secondary state data sample and the secondary recovery state data sample into the preset deep neural network to be trained to obtain the human action recognition result, until the action recognition result meets the loss convergence condition to obtain the human action recognition model.

[0072] It should be noted that the wireless channel state data samples refer to the original data samples obtained from the wireless channel state information (CSI). These samples contain the state of the wireless signal at a specific moment, such as signal strength, phase, etc. These states can reflect the channel changes caused by human body movements. Discretization refers to converting continuous wireless channel state data samples into discrete form in order to simplify the processing process or because the model can only process discrete data. This involves mapping continuous signal values ​​to a finite set of values. Secondary state data samples are wireless channel state data samples obtained after discrete processing. These samples are simplified or transformed versions of the original samples and are used for subsequent diffusion restoration and feature extraction. The diffusion model is a model for processing wireless channel state data, which simulates the diffusion process of the signal in space and is used to restore the signal distorted by multipath effects or other interference. The secondary restored state data samples are data samples obtained by restoring the secondary state data samples through the diffusion model. These samples are considered to be closer to the original, undisturbed signal state. A positive sample pair is a sample pair obtained by binding a secondary state data sample with its corresponding secondary restored state data sample. Since these sample pairs come from the same original signal, they are considered similar. Similar feature representations are features extracted from positive sample pairs that can represent the similarity between samples. In human action recognition, similar feature representations can help the model recognize the same action. Negative sample pairs are sample pairs obtained by binding the part of the wireless channel state data sample excluding the secondary state data sample with other samples. These sample pairs come from different original signals and are therefore considered dissimilar.

[0073] Dissimilar feature representations are features extracted from negative sample pairs that can represent the dissimilarity between samples. In human action recognition, dissimilar feature representations can help the model distinguish different actions. The deep neural network to be trained is a deep learning model that has not yet completed training and will be used to learn how to recognize human actions from feature representations. The model will be trained using a contrastive loss function. The contrastive loss function is a loss function used to train deep neural networks. The goal is to minimize the distance between similar feature representations and maximize the distance between dissimilar feature representations. This helps the model better distinguish different actions. The loss convergence condition is a condition preset when training a deep neural network to determine whether the model training has achieved satisfactory results. When the loss value of the model meets this condition, the model is considered to have converged. The human action recognition model is a deep neural network model that has been trained and meets the loss convergence condition and can be used to recognize human actions. The model recognizes different human actions by learning the differences between similar and dissimilar feature representations.

[0074] It can be understood that the enhanced CSI obtained by diffusion model enhancement is used together with the original CSI as the multi-view input for training the human activity recognition network model. After joint training of the contrastive learning branch and the backbone classification model, a working human activity recognition model can be finally obtained.

[0075] In order to fully integrate the enhanced CSI data with the original CSI data and mine the correlation between them, this paper uses Figure 2 The design idea of ​​the supervised contrastive learning framework shown in the figure regards the original CSI and the enhanced CSI as different perspectives of the same sample.

[0076] Therefore, a certain original CSI and its enhanced CSI sent in the same training batch can be bound as a positive sample pair, while this CSI and other CSI samples in the same batch constitute a negative sample pair relationship. By constructing a contrastive learning framework that combines two perspectives and using the contrastive loss described by formula (2), the model can perform holistic training of the human activity recognition and classification process based on contrastive learning guidance.

[0077] (2) in, is a positive sample; is a negative sample; is the temperature coefficient used to set the smoothness of the loss; for each input anchor point sample ; Is to remove After itself and the rest A data set consisting of the indexes of positive samples that have consistent labels; is the size of the data sample included in the set.

[0079] Since the contrastive learning branch provides supervised loss during the model training phase and assists the model in extracting more discriminative model features, during reasoning, the network does not go through the contrastive learning branch, but directly performs refined feature extraction through the feature extraction network and finally identifies human activities.

[0080] In a feasible implementation manner, the step of inputting the secondary state data sample and the secondary restored state data sample into a preset deep neural network to be trained to obtain a human motion recognition result includes: Extracting local information and spatial structure from the secondary state data sample and the secondary restored state data sample through a preset one-dimensional neural network to obtain low-level features; Based on the self-attention mechanism, the global relationship and long-distance dependency in the low-level features are modeled to obtain high-level features; The high-level features are classified and normalized to obtain a human action recognition result.

[0081] It should be noted that the secondary state data samples are discretely processed wireless channel state information data samples, which are simplified versions of the original CSI data and are used for further feature extraction and analysis.

[0082] The secondary restored state data samples are data samples obtained by restoring the secondary state data samples through the diffusion model, and are intended to restore or enhance the information in the original CSI data that is damaged due to interference or distortion during signal propagation. A one-dimensional neural network is a deep learning model that is specifically used to process one-dimensional data sequences, such as time series data. In this scenario, it is used to extract local information and spatial structure from the secondary state data samples and the secondary restored state data samples.

[0083] Local information is the characteristics of a local area or a short period of time in a data sample, which usually reflects the instantaneous state of the signal or changes in a small range. Spatial structure is the spatial relationship between different parts of a data sample, which can reveal the geometric or topological properties of the signal.

[0084] Low-level features are preliminary features extracted by one-dimensional neural networks, which usually contain local information and spatial structure of the data, but do not yet contain more complex global relationships or long-distance dependencies. The self-attention mechanism is a deep learning technique that allows the model to assign different attention weights to each element in the sequence when processing sequence data, thereby capturing long-distance dependencies and global relationships in the sequence.

[0085] Global relationships are the overall relationships between different parts of a data sample, which can reveal the global characteristics and contextual information of the signal. Long-range dependencies are the dependencies between elements that are far apart in a data sample, which are very important for understanding the long-term behavior and trends of the signal. High-level features are deeper features extracted based on the self-attention mechanism, which contain global relationships and long-range dependencies of the data and are more critical for human action recognition. Classification processing is the process of mapping high-level features to specific categories (such as different human actions). This usually involves the use of classification algorithms such as support vector machines, decision trees, or neural networks.

[0086] Normalization is the process of scaling features so that their values ​​fall within a specific range, such as [0, 1] or [-1, 1], to improve the performance and stability of the model.

[0087] The human action recognition result is the final output, which represents the human action recognized by the model for the input CSI data sample. This result is obtained based on the extracted high-level features and classification and normalization processing.

[0088] Exemplarily, the enhanced CSI obtained by diffusion model enhancement is used together with the original CSI as the multi-view input for training the human activity recognition network model. After the joint training of the contrastive learning branch and the backbone classification model, a workable human activity recognition model can be finally obtained. In the process of feature extraction and classification, the CNN-1D network CNN is first used to process the local information and spatial structure in the input data and extract low-level features. Subsequently, the classification feature map output is connected to a Transformer encoder, and its self-attention mechanism is used to model the global relationship and long-distance dependency in the sequence data, and further extract high-level feature representations. This combination can enhance the modeling ability of the model for sequence data while retaining the sensitivity of CNN to spatial information, thereby improving the performance of the model in various tasks, especially when processing complex data containing both image features and sequence features. Finally, the model is used to convert the output value of the final model into the probability of predicting that the CSI belongs to each specific activity through a linear layer classification head and the final SoftMax operation. In short, the classification framework can expand the overall data set based on the original CSI, and use the contrastive learning strategy and the network advantages of both CNN and Transformer to improve the generalization performance of the model in the task of human activity recognition.

[0089] In addition, in order to fully integrate the enhanced CSI data with the original CSI data and mine the correlation between them, the present application uses Figure 2 The design idea of ​​the supervised contrastive learning framework shown in the figure regards the original CSI and the enhanced CSI as different perspectives of the same sample.

[0090] For example, in order to help understand the implementation process of the wireless human motion recognition method obtained by combining this embodiment with the above-mentioned embodiment 1, please refer to Figure 3 , Figure 3 A brief flow chart of a wireless human motion recognition method is provided, specifically: The device first obtains wireless channel state information (CSI) data samples, which contain the state information of the wireless signal at a specific moment. The continuous CSI data samples are discretized to obtain secondary state data samples. This step may involve mapping the continuous signal values ​​into a finite set of values ​​for subsequent processing.

[0091] The secondary state data samples are restored through the diffusion model to obtain secondary restored state data samples. This process aims to restore or enhance the information in the original CSI data that is damaged due to interference or distortion during signal propagation.

[0092] A one-dimensional neural network (CNN) is used to process the secondary state data samples and the secondary restored state data samples to extract local information and spatial structure and obtain low-level features.

[0093] Based on the self-attention mechanism, low-level features are processed to model global relationships and long-distance dependencies to obtain high-level features. This helps capture complex patterns and relationships in the data.

[0094] The secondary state data samples are bound to the secondary restored state data samples to form a positive sample pair. Meanwhile, the portion of the wireless channel state data samples excluding the secondary state data samples is bound to other samples to form a negative sample pair.

[0095] High-level features are classified and normalized in preparation for human action recognition.

[0096] The positive and negative sample pairs are input into the preset deep neural network to be trained. The network is trained by the contrastive loss function to minimize the distance between similar feature representations and maximize the distance between dissimilar feature representations.

[0097] It is determined whether the action recognition result meets the preset loss convergence condition. If not, return to the step of inputting the positive sample pair and the negative sample pair into the preset deep neural network to be trained to continue training until the loss convergence condition is met, thereby obtaining a human action recognition model.

[0098] Finally, the trained model can output human action recognition results based on the extracted high-level features and classification and normalization processing.

[0099] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the wireless human motion recognition method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0100] This application also provides a wireless human motion recognition device, please refer to Figure 4 , the wireless human motion recognition device comprises: An acquisition module 10, configured to acquire wireless channel state information data in response to a start identification signal; A diffusion module 20, configured to perform diffusion restoration on the wireless channel state information data according to a preset diffusion model to obtain restored channel state information data; The recognition module 30 is used to input the wireless channel state information data and the restored channel state information data into a preset human motion recognition model to obtain a human motion recognition result.

[0101] And / or, the diffusion module 20 includes: A first preprocessing module, used to preprocess the wireless channel state information data to obtain original state information data; A first diffusion module, used for performing diffusion restoration on the original state information data according to the diffusion model to obtain restored data and update the corresponding time step; The first loop module is used to determine whether the time step meets a preset termination condition. If not, the original state information data is updated based on the restored data, and the step of performing diffusion restoration on the original state information data according to the diffusion model is returned. If satisfied, the restored data is output as restored channel state information data.

[0102] And / or, the first diffusion module includes: A first acquisition module, for acquiring an original time step sequence; A first extraction module is used to perform subsequence extraction on the original time step sequence according to a preset time step subsequence extraction rule to obtain a target time step subsequence; A first sampling module, used for sampling the original state information data according to a preset sampling rule to obtain sampled state information data; The first restoration module is used to perform diffusion restoration on the sampled state information data to obtain restored data and update the corresponding time step.

[0103] And / or, the first preprocessing module includes: A first missing value processing module, used to perform missing value processing on the wireless channel state information data according to a preset first ratio to obtain a first preprocessing result; The first prediction value processing module is used to perform a prediction value random zeroing process on the first preprocessing result according to a preset second ratio to obtain original state information data.

[0104] And / or, the wireless human motion recognition device comprises: A second acquisition module is used to acquire wireless channel state data samples; A first discretization module, used for discretizing the wireless channel state data samples to obtain secondary state data samples; A second restoration module, configured to perform diffusion restoration on the secondary state data sample through the diffusion model to obtain a secondary restored state data sample; A first binding module, used for binding the secondary state data sample with the secondary restored state data sample to obtain a positive sample pair; A first extraction module is used to extract features of the positive sample pair to obtain similar feature representations; A second binding module, configured to bind data samples excluding the secondary state data samples from the wireless channel state data samples into a negative sample pair; A second extraction module is used to extract features of the negative sample pair to obtain a dissimilar feature representation; A first recognition module, used for inputting the secondary state data sample and the secondary restored state data sample into a preset deep neural network to be trained to obtain a human action recognition result, wherein the deep neural network to be trained is obtained by training with a preset contrast loss function to minimize the distance between similar feature representations and maximize the distance between dissimilar feature representations; The second loop module is used to determine whether the action recognition result meets the preset loss convergence condition. If not, it returns to the step of inputting the secondary state data sample and the secondary recovery state data sample into the preset deep neural network to be trained to obtain the human action recognition result, until the action recognition result meets the loss convergence condition to obtain the human action recognition model.

[0105] And / or, the first identification module includes: A first low-level extraction module, configured to extract local information and spatial structure from the secondary state data sample and the secondary restored state data sample through a preset one-dimensional neural network to obtain low-level features; A first high-level extraction module, for modeling the global relationship and long-distance dependency in the low-level features based on a self-attention mechanism to obtain high-level features; The first normalization processing module is used to perform classification processing and normalization processing on the high-level features to obtain human action recognition results.

[0106] The wireless human motion recognition device provided by the present application adopts the wireless human motion recognition method in the above embodiment, which can solve the technical problem of low accuracy of wireless human motion recognition. Compared with the prior art, the beneficial effects of the wireless human motion recognition device provided by the present application are the same as the beneficial effects of the wireless human motion recognition method provided by the above embodiment, and other technical features in the wireless human motion recognition device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0107] The present application provides a wireless human motion recognition device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the wireless human motion recognition method in the above-mentioned embodiment one.

[0108] Reference below Figure 5, which shows a schematic diagram of the structure of a wireless human motion recognition device suitable for implementing the embodiment of the present application. The wireless human motion recognition device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, tablet computers, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 5 The wireless human motion recognition device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0109] like Figure 5 As shown, the wireless human motion recognition device may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the wireless human motion recognition device are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the wireless human motion recognition device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows a wireless human motion recognition device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or provided alternatively.

[0110] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0111] The wireless human motion recognition device provided by the present application adopts the wireless human motion recognition method in the above embodiment, which can solve the technical problem of low accuracy of wireless human motion recognition. Compared with the prior art, the beneficial effects of the wireless human motion recognition device provided by the present application are the same as the beneficial effects of the wireless human motion recognition method provided by the above embodiment, and other technical features in the wireless human motion recognition device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0112] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0113] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0114] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the wireless human motion recognition method in the above-mentioned embodiment.

[0115] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: Read Only Memory), an erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination of the above.

[0116] The computer-readable storage medium may be included in the wireless human motion recognition device; or may exist independently without being assembled into the wireless human motion recognition device.

[0117] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the wireless human motion recognition device, the wireless human motion recognition device: In response to the start identification signal, acquiring wireless channel state information data; Performing diffusion restoration on the wireless channel state information data according to a preset diffusion model to obtain restored channel state information data; The wireless channel state information data and the restored channel state information data are input into a preset human motion recognition model to obtain a human motion recognition result.

[0118] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0119] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square 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 square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0120] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0121] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned wireless human motion recognition method, and can solve the technical problem of low accuracy of wireless human motion recognition. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the wireless human motion recognition method provided in the above-mentioned embodiment, and will not be repeated here.

[0122] The present application also provides a computer program product, including a computer program, which implements the steps of the wireless human motion recognition method as described above when executed by a processor.

[0123] The computer program product provided by the present application can solve the technical problem of low accuracy of wireless human motion recognition. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the wireless human motion recognition method provided by the above embodiment, which will not be repeated here.

[0124] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A wireless human motion recognition method, characterized in that: The method includes: In response to the start identification signal, acquiring wireless channel state information data; Performing diffusion restoration on the wireless channel state information data according to a preset diffusion model to obtain restored channel state information data; The wireless channel state information data and the restored channel state information data are input into a preset human motion recognition model to obtain a human motion recognition result.

2. The method according to claim 1, characterized in that The step of performing diffusion restoration on the wireless channel state information data according to a preset diffusion model to obtain restored channel state information data comprises: Preprocessing the wireless channel state information data to obtain original state information data; According to the diffusion model, the original state information data is diffused and restored to obtain restored data, and the corresponding time step is updated; Determine whether the time step satisfies a preset termination condition; if not, update the original state information data based on the restored data, return to the step of performing diffusion restoration on the original state information data according to the diffusion model; if satisfied, output the restored data as restored channel state information data.

3. The method according to claim 2, characterized in that The step of performing diffusion restoration on the original state information data according to the diffusion model to obtain restored data and updating the corresponding time step includes: Get the original time step sequence; According to a preset time step subsequence extraction rule, subsequence extraction is performed on the original time step sequence to obtain a target time step subsequence; Sampling the original state information data according to a preset sampling rule to obtain sampled state information data; The sampled state information data is diffused and restored to obtain restored data, and the corresponding time step is updated.

4. The method according to claim 2, characterized in that The step of preprocessing the wireless channel state information data to obtain original state information data comprises: Performing missing value processing on the wireless channel state information data according to a preset first ratio to obtain a first preprocessing result; The predicted values ​​of the first preprocessing result are randomly set to zero according to a preset second ratio to obtain original state information data.

5. The method according to claim 1, characterized in that The step of inputting the wireless channel state information data and the restored channel state information data into a preset human motion recognition model to obtain a human motion recognition result includes: Obtaining wireless channel state data samples; Discretize the wireless channel state data samples to obtain secondary state data samples; Performing diffusion restoration on the secondary state data sample through the diffusion model to obtain a secondary restored state data sample; Binding the secondary state data sample with the secondary restored state data sample to obtain a positive sample pair; Extracting features of the positive sample pair to obtain similar feature representations; Binding data samples of the wireless channel state data samples excluding the secondary state data samples into a negative sample pair; Extracting features of the negative sample pair to obtain a dissimilar feature representation; Inputting the secondary state data sample and the secondary restored state data sample into a preset deep neural network to be trained to obtain a human action recognition result, wherein the deep neural network to be trained is obtained by training with a preset contrast loss function to minimize the distance between similar feature representations and maximize the distance between dissimilar feature representations; Determine whether the action recognition result meets the preset loss convergence condition. If not, return to the step of inputting the secondary state data sample and the secondary recovery state data sample into the preset deep neural network to be trained to obtain the human action recognition result, until the action recognition result meets the loss convergence condition to obtain the human action recognition model.

6. The method according to claim 5, characterized in that The step of inputting the secondary state data sample and the secondary restored state data sample into a preset deep neural network to be trained to obtain a human motion recognition result comprises: Extracting local information and spatial structure from the secondary state data sample and the secondary restored state data sample through a preset one-dimensional neural network to obtain low-level features; Based on the self-attention mechanism, the global relationship and long-distance dependency in the low-level features are modeled to obtain high-level features; The high-level features are classified and normalized to obtain a human action recognition result.

7. A wireless human motion recognition device, characterized in that: The device comprises: An acquisition module, configured to acquire wireless channel state information data in response to a start identification signal; A diffusion module, used for performing diffusion restoration on the wireless channel state information data according to a preset diffusion model to obtain restored channel state information data; The recognition module is used to input the wireless channel state information data and the restored channel state information data into a preset human motion recognition model to obtain a human motion recognition result.

8. A wireless human motion recognition device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the wireless human motion recognition method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the wireless human motion recognition method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the wireless human motion recognition method according to any one of claims 1 to 6 are implemented.

Images