RFID contactless human behavior recognition method and device based on world model

Abstract

The application discloses an RFID non-contact human behavior recognition method and device based on a World Model, and the method comprises the following steps: acquiring a human behavior signal and pre-processing the human behavior signal to obtain a real sample sequence data set; an initial network model is constructed, the real sample sequence data set is input into the initial network model, long-time time sequence evolution is modeled through an RSSM dynamic module in the initial network model, a Decoder reconstruction constraint is added to a classification loss, a KL divergence loss of a 'free energy lower limit' skill is introduced to guarantee feature effectiveness, an Imagination mechanism is introduced to generate virtual sample expansion data, and a well-trained human behavior recognition model is obtained; the well-trained human behavior recognition model is used for human behavior recognition, and a recognition result corresponding to a to-be-predicted behavior signal is obtained; and therefore, the recognition accuracy and robustness are considered.

Description

Technical Field

[0001] This invention relates to the field of radio frequency identification (RFID) technology, and in particular to a World Model-based RFID contactless human behavior identification method, a computer-readable storage medium, a computer device, and a World Model-based RFID contactless human behavior identification device. Background Technology

[0002] Among related technologies, Human Activity Recognition (HAR) aims to monitor and identify human behavior (walking, lying down, etc.) by analyzing data acquired from sensors, thereby achieving the goal of automatically detecting and interpreting human behavior. In recent years, the method of contactless human activity recognition using RFID passive tags has continued to attract researchers' attention. Compared with traditional methods such as camera video sensing and wearable device sensing, using RFID signals for human activity recognition has advantages such as low cost, low energy consumption, good privacy protection, and flexible deployment. Its low energy consumption and maintenance costs make it particularly advantageous for application in places such as health and wellness centers that require long-term stable operation. However, in actual deployment and application, existing contactless RFID behavior recognition systems face the following key problems: (1) Low sample efficiency. Specifically, current traditional RFID human behavior recognition methods generally rely on large-scale labeled sample sets during the model training stage. In order to cover common human behaviors such as walking, sitting, and waving, as well as behavioral differences among people of different body types, volunteers often need to perform the same behavior dozens or even hundreds of times in a strictly controlled experimental environment. Effective signal segments are manually extracted and labeled from massive amounts of raw signals. During the labeling process, it is also necessary to check whether the signal matches the time node of the corresponding behavior. Especially in special scenarios such as medical care and nursing, considering the special nature of the data collection objects, the difficulty and cost of sample collection increase sharply, which is far from meeting the amount of data required for model training, thus failing to provide reliable safety monitoring. (2) The problem of insufficient long-term dynamic dependence capture. Specifically, traditional RFID human behavior recognition methods mostly adopt short-sequence signal analysis mode. They usually extract static features such as amplitude, phase and frequency of RFID signals in the short segment, focusing on the short-term signal changes corresponding to each step of the action to realize the recognition of human behavior. However, for human behaviors with long-term dynamic characteristics (such as getting up and walking), the completion of such behaviors usually lasts for several seconds, and there is an obvious temporal dependence between different stages of the action. Since traditional methods cannot establish the long-term correlation of such behaviors on the time axis, they often misjudge different stages of the same behavior as independent different behaviors, or the overall behavior recognition accuracy drops significantly due to the loss of key temporal information, which cannot meet the needs of practical applications. Summary of the Invention

[0003] This invention aims to at least partially solve one of the technical problems in the aforementioned technologies. Therefore, one objective of this invention is to propose a World Model-based RFID contactless human behavior recognition method that can significantly improve sample utilization efficiency, enhance long-term action modeling capabilities, and improve robustness in noisy environments, while possessing stronger generalization ability for unseen action variants, thus balancing recognition accuracy and robustness.

[0004] A second objective of this invention is to provide a computer-readable storage medium.

[0005] The third objective of this invention is to provide a computer device.

[0006] The fourth objective of this invention is to propose an RFID contactless human behavior recognition device based on the World Model.

[0007] To achieve the above objectives, a first aspect of the present invention proposes a World Model-based RFID contactless human behavior recognition method, comprising: acquiring human behavior signals collected by an RFID device; preprocessing the human behavior signals to obtain a real sample sequence dataset; constructing an initial network model by inputting the real sample sequence dataset into the initial network model to obtain the deterministic and random states at the current moment through a Transition dynamic module, and obtaining a real sample loss based on the deterministic and random states; obtaining the deterministic and random states of virtual samples using the Transition dynamic module based on the deterministic and random states of the last moment of the real sample sequence, and training based on the deterministic and random states of the virtual samples to obtain a virtual sample loss; performing a weighted fusion of the real sample loss and the virtual sample loss to obtain a weighted fusion total loss, and updating the model parameters of the initial network model based on the weighted fusion loss to obtain a trained human behavior recognition model; and using the trained human behavior recognition model to perform human behavior recognition to obtain the recognition result corresponding to the predicted behavior signal.

[0008] The RFID contactless human behavior recognition method based on the World Model according to embodiments of the present invention has the following advantages: it models long-term temporal evolution through the RSSM dynamic model, adds Decoder reconstruction constraints in addition to classification loss and introduces KL divergence loss with the "free energy lower limit" technique to ensure feature effectiveness, and introduces the Imagination mechanism to generate virtual samples to expand data, which significantly improves sample utilization efficiency, enhances long-term action modeling ability, improves robustness in noisy environments, and has a stronger generalization ability to unseen action variants, thus balancing recognition accuracy and robustness.

[0009] In addition, the RFID contactless human behavior recognition method based on the World Model proposed in the above embodiments of the present invention may also have the following additional technical features:

[0010] Optionally, the human behavior signal is preprocessed to obtain a real sample sequence dataset, including: dividing the human behavior signal into frames, i.e., extracting action segments according to a fixed time window; filling missing values ​​using linear interpolation; and standardizing intra-frame features to obtain a real sample sequence.

[0011] Optionally, the deterministic and random states at the current moment are obtained through the Transition dynamic module, including: performing iterative updates through GRU units, updating the input including the random state at the previous moment and the encoded features of the human behavior signal at the current moment; concatenating and fusing the random state at the previous moment and the encoded features of the human behavior signal at the current moment and inputting them into GRUCell, performing operations with the deterministic state at the previous moment to obtain the deterministic state at the current moment; generating a prior distribution of the random state based on the deterministic state at the current moment through a prior network; generating a posterior distribution of the random state by combining the deterministic state at the current moment with the encoded features of the RFID signal at the current moment through a posterior network; and sampling the prior and posterior distributions of the random state to obtain the current random state in order to characterize random fluctuations.

[0012] Optionally, the real sample loss includes reconstruction loss, classification loss, and KL divergence loss.

[0013] Alternatively, the loss of the true sample can be obtained according to the following formula:

[0014]

[0015]

[0016] in, Represents a deterministic state. Represents a random state. Indicates the true sample loss. Indicates the reconstruction loss weights. Represents the classification loss weights. Indicates the weight of the KL divergence loss. Indicates the reconstructed signal. Represents primitive human behavioral signals. Indicates the behavioral prediction results. Indicates the true label, This represents the Gaussian distribution generated by the prior network. This represents the Gaussian distribution generated by the posterior network.

[0017] Alternatively, the weighted fusion total loss can be obtained according to the following formula:

[0018]

[0019] in, Indicates the total weighted fusion loss. This indicates the weight of the imagined loss. This represents the loss of virtual samples.

[0020] To achieve the above objectives, a second aspect of the present invention provides a computer-readable storage medium storing a World Model-based RFID contactless human behavior recognition program, which, when executed by a processor, implements the World Model-based RFID contactless human behavior recognition method as described above.

[0021] To achieve the above objectives, a third aspect of the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the World Model-based RFID contactless human behavior recognition method as described above.

[0022] To achieve the above objectives, a fourth aspect of the present invention proposes a World-based... The RFID contactless human behavior recognition device of Model includes: an acquisition module for acquiring human behavior signals collected by an RFID device; a preprocessing module for preprocessing the human behavior signals to obtain a real sample sequence dataset; a training module for constructing an initial network model, inputting the real sample sequence dataset into the initial network model, obtaining the deterministic and random states at the current moment through a Transition dynamic module, and obtaining the real sample loss based on the deterministic and random states; obtaining the deterministic and random states of virtual samples based on the deterministic and random states at the last moment of the real sample sequence using the Transition dynamic module, and training based on the deterministic and random states of the virtual samples to obtain the virtual sample loss; performing a weighted fusion of the real sample loss and the virtual sample loss to obtain a weighted fusion total loss, and updating the model parameters of the initial network model based on the weighted fusion loss to obtain a trained human behavior recognition model; and a human behavior recognition module for using the trained human behavior recognition model to perform human behavior recognition to obtain the recognition result corresponding to the predicted behavior signal. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the RFID contactless human behavior recognition method based on the World Model according to an embodiment of the present invention.

[0024] Figure 2 This is a schematic diagram of the hardware and software structure of an RFID system according to an embodiment of the present invention;

[0025] Figure 3 This is a schematic diagram of the state transition process of the World Model according to an embodiment of the present invention;

[0026] Figure 4 This is a schematic diagram of the World Model algorithm training process according to an embodiment of the present invention;

[0027] Figure 5 This is a block diagram of an RFID contactless human behavior recognition device based on a World Model according to an embodiment of the present invention. Detailed Implementation

[0028] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0029] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the invention to those skilled in the art.

[0030] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0031] Figure 1 This is a flowchart illustrating the RFID contactless human behavior recognition method based on the World Model according to an embodiment of the present invention. Figure 1 As shown, this World Model-based RFID contactless human behavior identification method includes the following steps:

[0032] S101, acquire human behavior signals collected by the radio frequency identification device.

[0033] It should be noted that the data information required for training the model is collected using Radio Frequency Identification (RFID) technology as the core technology carrier; such as... Figure 2 As shown, an RFID system can be divided into two aspects: hardware components and software support. At the hardware level, the core of an RFID system consists of three main modules: tags, antennas, and readers. These three components interact via radio frequency signals to transmit data. Tags, as data carriers, parse reader commands and return information through a logic control unit. Antennas are responsible for receiving reader signals and reflecting response signals. The reader is the core of the system control, simultaneously generating, receiving, and parsing signals, and communicating with external systems. At the software level, RFID middleware is crucial for connecting the hardware and application systems, undertaking the responsibilities of data processing, equipment management, and security.

[0034] S102, preprocess the human behavior signals to obtain a real sample sequence dataset.

[0035] As an example, the human behavior signal is preprocessed to obtain a real sample sequence dataset, including: dividing the human behavior signal into frames, that is, extracting action segments according to a fixed time window; filling missing values ​​by linear interpolation; and standardizing the intra-frame features to obtain a real sample sequence.

[0036] Specifically, the reader continuously sends multiple query requests at a set frequency. The RFID middleware saves the timestamp, EPC code, RSSI (signal strength), and Phase data returned by the query requests. After data preprocessing, these data are used as features X for training. The actions performed under the timestamp (encoded as 0-9, a total of 10 actions) are used as training labels y. The resulting data is used as a dataset for model training and testing.

[0037] S103, Construct the initial network model by inputting the real sample sequence dataset into the initial network model, so as to obtain the deterministic state and random state at the current time through the Transition dynamic module, and obtain the real sample loss based on the deterministic state and random state.

[0038] As an example, the deterministic and random states at the current moment are obtained through the Transition dynamic module, including: iterative updates are performed by GRU units, updating the inputs to include the random state at the previous moment and the encoded features of the human behavior signal at the current moment; the random state at the previous moment and the encoded features of the human behavior signal at the current moment are concatenated and fused, and then input into GRUCell, and processed with the deterministic state at the previous moment to obtain the deterministic state at the current moment; a prior network generates a prior distribution of the random state based on the deterministic state at the current moment; a posterior network combines the deterministic state at the current moment with the encoded features of the RFID signal at the current moment to generate a posterior distribution of the random state; the current random state is obtained by sampling from the prior distribution and the posterior distribution of the random state to characterize random fluctuations.

[0039] In other words, such as Figure 3 As shown, state transitions are primarily achieved through the newly added Transition dynamic module based on the Cyclic State-Space Model (RSSM). The states in the RSSM include deterministic states. and random states The combination of these two methods is used to accurately capture the temporal evolution of RFID signals. Deterministic state. Iterative updates are performed using GRU units, with the update input including the random state from the previous time step. Encoded characteristics of the RFID signal at the current moment The two are then merged and input into GRUCell, along with the deterministic state from the previous time step. Perform calculations to obtain the deterministic state at the current moment. Random state Then, the model is constructed using a Gaussian distribution generated by a prior network and a posterior network: the prior network is based on the current time step... generate prior distribution The posterior network, on the other hand, incorporates the current moment. and generate posterior distribution The current random state is obtained by sampling from these two distributions. To characterize random fluctuations. And so on, based on the encoded characteristics of the RFID signal at the next moment. and the random state at the current moment To obtain the deterministic state at the next moment. .

[0040] It should be noted that, to address the problem of the inability to model long-term dynamic dependencies, this application adds a Transition dynamic module based on the Cyclic State-Space Model (RSSM). This module uses deterministic states... and random states The combination captures the temporal evolution of RFID signals, where the deterministic state is updated through GRU units, i.e.

[0041]

[0042] Used to capture stable dynamic trends; stochastic states are modeled using Gaussian distributions generated by prior and posterior networks, i.e.

[0043]

[0044]

[0045] This is used to characterize the random fluctuations of signals. To adapt to the complex dynamics of RFID signals, a multi-layer MLP structure is also adopted to enhance the fitting ability of the prior and posterior networks, enabling the model to effectively capture the dynamic dependencies of long-term continuous actions and overcome the limitation of traditional sliding windows that can only handle local time series.

[0046] Secondly, to ensure that the features learned by the model truly reflect the essential laws of the signal, a Decoder reconstruction constraint and a KL divergence loss are added to the traditional classification loss. The Decoder converts the RSSM state... Decoding back to the original signal dimension, i.e.

[0047]

[0048] By reconstructing the loss Ensure feature validity; KL divergence loss This constrains the difference between the posterior and prior distributions, preventing random states from deviating excessively from dynamic patterns. In the implementation, a "free energy lower bound (free nats)" technique is introduced, where the KL divergence is considered zero when it falls below a certain threshold, thus preventing the KL term from becoming too small and causing model training instability. This approach improves model convergence while ensuring the effectiveness of the constraints. After the above steps, the total loss function is expanded to:

[0049]

[0050] in, , and These represent the weights of each loss term. This multi-constraint mechanism effectively reduces the feature overfitting problem that is prone to occur in existing methods.

[0051] S104. Based on the deterministic and random states of the last moment of the real sample sequence, the deterministic and random states of the virtual sample are obtained using the Transition dynamic module. Training is then performed based on the deterministic and random states of the virtual sample to obtain the virtual sample loss.

[0052] In other words, to address the pain point of high RFID sample collection costs, this application introduces an Imagination mechanism; this mechanism starts from the end state of real samples and uses the RSSM dynamic model to generate a virtual state sequence:

[0053]

[0054] in, The actual sequence length. Imagine the number of steps.

[0055] S105, the real sample loss and virtual sample loss are weighted and fused to obtain the weighted fusion total loss, and the model parameters of the initial network model are updated according to the weighted fusion loss to obtain the trained human behavior recognition model.

[0056] As an example, the weighted fusion total loss is obtained according to the following formula:

[0057]

[0058] in, Indicates the total weighted fusion loss. This indicates the weight of the imagined loss. This represents the loss of virtual samples.

[0059] In other words, the virtual state sequence obtained above is used for training, through... The weighted loss integrates information from real and virtual samples. (For imagined loss weights). Among them, virtual sample loss... By including only classification loss (cross-entropy loss), this approach significantly amplifies the value of real samples, generating a large number of virtual samples with a small amount of real data, which significantly improves the model's sample efficiency. At the same time, virtual samples can cover action variants that do not appear in real data, enhancing the model's generalization ability.

[0060] In addition, such as Figure 4 As shown, training stops when the model triggers the early stopping mechanism or reaches the maximum number of training iterations.

[0061] In summary, this application not only improves recognition accuracy while maintaining the simplicity and trainability of the model structure, but also achieves a good balance in terms of robustness.

[0062] S106, Use a trained human behavior recognition model to perform human behavior recognition in order to obtain the recognition result corresponding to the behavior signal to be predicted.

[0063] In other words, the same process is used to collect and preprocess the signals of the behavior to be predicted, and then input the preprocessed signal segments into the trained human behavior recognition model. The Transition module updates the state in real time. The fusion state at the current moment is input into the classifier to obtain the behavior prediction probability distribution. The category with the highest probability is selected as the recognition result (such as "sitting", "walking-standing-squatting", etc.), and the behavior label is output. For long-term behaviors lasting several seconds (such as "sitting-standing-walking"), the model captures the temporal dependencies of each stage through the RSSM module to avoid the existing misclassification of different stages of the same behavior as independent behaviors, and finally outputs a complete behavior sequence label.

[0064] To verify the effectiveness of the method proposed in this invention, the following experimental design was conducted:

[0065] In the experiment, various types of actions were preset, including basic actions such as sitting, standing, and lying down, as well as actions performed continuously in time, such as sitting-standing-walking and walking-standing-squatting. Participants were asked to complete the corresponding actions according to a predetermined procedure in an environment with RFID tag arrays, and corresponding experimental data were collected. In the data preprocessing stage, the same process was used for the models being compared: first, the raw collected data was divided into frames, and action segments were extracted according to fixed time windows; then, missing values ​​in the data were filled using linear interpolation to ensure signal continuity; finally, intra-frame features were standardized to eliminate the impact of dimensional differences on model training. The preprocessed action data were then input into the comparison models to compare the performance of the models before and after WorldModel optimization in the action recognition task.

[0066] Table 1 shows the comprehensive performance evaluation results of the World Model-based optimization method in behavioral modeling tasks at different time scales.

[0067]

[0068] As shown in the table above, the optimized model built with the World Model exhibits significant advantages in both performance and training practicality: the model effectively overcomes the limitations of traditional methods, improving accuracy in short, medium, and long sequence tasks, with a particularly outstanding effect in capturing dynamic dependencies in medium and long sequences; at the same time, by constructing multiple constraints on the total loss function, the model avoids the feature overfitting problem that is prone to occur in traditional methods, improving feature effectiveness and convergence stability, thus reducing the number of training iterations required for the network structure with relevant optimizations to reach the best accuracy, and significantly improving training efficiency; in addition, the imagination mechanism enhances the model's generalization ability while improving sample efficiency, achieving overall synergistic optimization of recognition accuracy, training efficiency, and robustness.

[0069] To implement the above embodiments, this invention also proposes a computer-readable storage medium storing a World Model-based RFID contactless human behavior recognition program. When executed by a processor, the World Model-based RFID contactless human behavior recognition program implements the World Model-based RFID contactless human behavior recognition method as described above.

[0070] According to embodiments of the present invention, a computer-readable storage medium stores a World Model-based RFID contactless human behavior recognition program, enabling a processor to implement the World Model-based RFID contactless human behavior recognition method as described above when executing the World Model-based RFID contactless human behavior recognition program. This significantly improves sample utilization efficiency, enhances long-term action modeling capabilities, and improves robustness in noisy environments. It also has a stronger generalization ability for unseen action variants, thereby balancing recognition accuracy and robustness.

[0071] To implement the above embodiments, this invention proposes a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the World Model-based RFID contactless human behavior recognition method as described above.

[0072] According to the computer device of the present invention, a World Model-based RFID contactless human behavior recognition program is stored in a memory, so that when the processor executes the World Model-based RFID contactless human behavior recognition program, the World Model-based RFID contactless human behavior recognition method described above is implemented. This significantly improves sample utilization efficiency, enhances long-term action modeling capabilities, and improves robustness in noisy environments. It also has a stronger generalization ability for unseen action variants, thereby balancing recognition accuracy and robustness.

[0073] To achieve the above embodiments, this invention proposes an RFID contactless human behavior recognition device based on a World Model, such as... Figure 5 As shown, the World Model-based RFID contactless human behavior recognition device includes: an acquisition module 10, a preprocessing module 20, a training module 30, and a human behavior recognition module 40.

[0074] The system comprises the following modules: Acquisition module 10 acquires human behavior signals collected by an RFID device; Preprocessing module 20 preprocesses the human behavior signals to obtain a real sample sequence dataset; Training module 30 constructs an initial network model by inputting the real sample sequence dataset into the initial network model, using the Transition dynamic module to obtain the deterministic and random states at the current moment, and obtaining the real sample loss based on the deterministic and random states; Based on the deterministic and random states at the last moment of the real sample sequence, the Transition dynamic module obtains the deterministic and random states of virtual samples, and obtains the virtual sample loss based on the deterministic and random states of the virtual samples; Weighted fusion of the real sample loss and the virtual sample loss is performed to obtain the weighted fusion total loss, and the model parameters of the initial network model are updated based on the weighted fusion loss to obtain a trained human behavior recognition model; Human behavior recognition module 40 uses the trained human behavior recognition model to perform human behavior recognition to obtain the recognition result corresponding to the predicted behavior signal.

[0075] It should be noted that the above regarding Figure 1 The description of the RFID contactless human behavior recognition method based on World Model is also applicable to this RFID contactless human behavior recognition device based on World Model, and will not be repeated here.

[0076] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0077] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0078] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0079] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0080] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0081] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0082] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0083] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0084] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0085] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0086] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0087] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A contactless human behavior recognition method based on World Model RFID, characterized in that, Includes the following steps: Acquire human behavioral signals collected by radio frequency identification (RFID) devices; The human behavior signals are preprocessed to obtain a dataset of real sample sequences; An initial network model is constructed, and the real sample sequence dataset is input into the initial network model to obtain the deterministic state and random state at the current time through the Transition dynamic module, and the real sample loss is obtained based on the deterministic state and random state. Based on the deterministic and random states of the last moment of the real sample sequence, the deterministic and random states of the virtual sample are obtained using the Transition dynamic module, and training is performed based on the deterministic and random states of the virtual sample to obtain the virtual sample loss. The real sample loss and the virtual sample loss are weighted and fused to obtain a weighted fusion total loss. The model parameters of the initial network model are updated according to the weighted fusion total loss to obtain a trained human behavior recognition model. The trained human behavior recognition model is used to perform human behavior recognition in order to obtain the recognition result corresponding to the behavior signal to be predicted; Specifically, the deterministic and stochastic states at the current moment are obtained through the Transition dynamic module, including: Iterative updates are performed using GRU units, with the update input including the random state of the previous time step and the encoded features of the human behavior signal at the current time step. The random state of the previous moment and the encoded features of the human behavior signal at the current moment are concatenated and fused, and then input into GRUCell. This is then used to perform calculations with the deterministic state of the previous moment to obtain the deterministic state at the current moment. The prior network generates a prior distribution of random states based on the deterministic state at the current moment; The posterior network combines the deterministic state at the current moment with the coding characteristics of the RFID signal at the current moment to generate the posterior distribution of the random state; The current random state is obtained by sampling from the prior distribution and the posterior distribution of the random state. The real sample loss includes reconstruction loss, classification loss, and KL divergence loss; The loss of the true sample is obtained using the following formula: in, Represents a deterministic state. Represents a random state. Indicates the true sample loss. Indicates the reconstruction loss weights. Represents the classification loss weights. Indicates the weight of the KL divergence loss. Indicates the reconstructed signal. Represents primitive human behavioral signals. Indicates the behavioral prediction result. Indicates the true label, This represents the Gaussian distribution generated by the prior network. This represents the Gaussian distribution generated by the posterior network. This represents a multilayer perceptron decoder, which will , express Mean Squared Error between Represents cross-entropy loss, Indicates the KL divergence loss; Specifically, the Imagination mechanism is used to generate a virtual state sequence starting from the end state of real samples and utilizing the RSSM dynamic model: in, The actual sequence length. Imagine the number of steps.

2. The RFID contactless human behavior recognition method based on World Model as described in claim 1, characterized in that, The human behavior signals are preprocessed to obtain a dataset of real sample sequences, including: The human behavior signal is divided into frames, that is, action segments are extracted according to fixed time windows; Missing values ​​are filled using linear interpolation. Intra-frame features are standardized to obtain the true sample sequence.

3. The RFID contactless human behavior recognition method based on World Model as described in claim 1, characterized in that, The weighted fusion total loss is obtained using the following formula: in, Indicates the total weighted fusion loss. This indicates the weight of the imagined loss. This represents the loss of virtual samples.

4. A computer-readable storage medium, characterized in that, It stores a World Model-based RFID contactless human behavior recognition program, which, when executed by the processor, implements the World Model-based RFID contactless human behavior recognition method as described in any one of claims 1-3.

5. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the RFID contactless human behavior recognition method based on the World Model as described in any one of claims 1-3.

6. A World Model-based RFID contactless human behavior recognition device, characterized in that, include The acquisition module is used to acquire human behavior signals collected by the radio frequency identification device; The preprocessing module is used to preprocess the human behavior signals to obtain a real sample sequence dataset; The training module is used to construct an initial network model. The real sample sequence dataset is input into the initial network model to obtain the deterministic and stochastic states at the current time step through the Transition dynamic module, and the real sample loss is obtained based on the deterministic and stochastic states. Based on the deterministic and stochastic states at the last time step of the real sample sequence, the Transition dynamic module is used to obtain the deterministic and stochastic states of virtual samples, and training is performed based on the deterministic and stochastic states of the virtual samples to obtain the virtual sample loss. The real sample loss and the virtual sample loss are weighted and fused to obtain a weighted fusion total loss. The model parameters of the initial network model are updated according to the weighted fusion total loss to obtain a trained human behavior recognition model. The human behavior recognition module is used to perform human behavior recognition using the trained human behavior recognition model in order to obtain the recognition result corresponding to the behavior signal to be predicted. Specifically, the deterministic and stochastic states at the current moment are obtained through the Transition dynamic module, including: Iterative updates are performed using GRU units, with the update input including the random state of the previous time step and the encoded features of the human behavior signal at the current time step. The random state of the previous moment and the encoded features of the human behavior signal at the current moment are concatenated and fused, and then input into GRUCell. This is then used to perform calculations with the deterministic state of the previous moment to obtain the deterministic state at the current moment. The prior network generates a prior distribution of random states based on the deterministic state at the current moment; The posterior network combines the deterministic state at the current moment with the coding characteristics of the RFID signal at the current moment to generate the posterior distribution of the random state; The current random state is obtained by sampling from the prior distribution and the posterior distribution of the random state. The real sample loss includes reconstruction loss, classification loss, and KL divergence loss; The loss of the true sample is obtained using the following formula: in, Represents a deterministic state. Represents a random state. Indicates the true sample loss. Indicates the reconstruction loss weights. Represents the classification loss weights. Indicates the weight of the KL divergence loss. Indicates the reconstructed signal. Represents primitive human behavioral signals. Indicates the behavioral prediction result. Indicates the true label, This represents the Gaussian distribution generated by the prior network. This represents the Gaussian distribution generated by the posterior network. This represents a multilayer perceptron decoder, which will , express Mean Squared Error between Represents cross-entropy loss, Indicates the KL divergence loss; Specifically, the Imagination mechanism is used to generate a virtual state sequence starting from the end state of real samples and utilizing the RSSM dynamic model: in, The actual sequence length. Imagine the number of steps.

Images