Training and recognition method and device of human behavior recognition model based on RFID
By performing feature extraction and domain discrimination processing on the RFID human behavior recognition model, the problem of low cross-domain recognition accuracy was solved, and higher recognition accuracy was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIAMEN UNIV
- Filing Date
- 2023-04-27
- Publication Date
- 2026-06-12
AI Technical Summary
Existing RFID-based human behavior recognition technology lacks consideration for non-transferable features when identifying across different domains, leading to a decrease in recognition accuracy.
By acquiring the target domain training dataset, preprocessing the signal received intensity indication sequence and phase value sequence, extracting feature vectors using the interconnected feature extraction module and self-attention module, and performing batch spectral penalty calculation through the domain discriminator to optimize the human behavior recognition model.
It effectively eliminates the influence of non-transferable features and domain-specific features, thus improving the accuracy of human behavior recognition.
Smart Images
Figure CN116486341B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of computer technology, and more specifically, to a training and identification method and apparatus for an RFID-based human behavior recognition model. Background Technology
[0002] Human behavior recognition has wide applications and is a hot research topic in the field of artificial intelligence, especially as a fundamental technology for many applications such as intelligent monitoring and human-computer interaction robots. Among current technical solutions, deep networks have rich feature representation capabilities and are often used for human behavior recognition. However, most RFID-based human behavior recognition research focuses on recognition in specific domains. When solving cross-domain problems, they usually match global features for domain adaptation, but lack consideration for non-transferable features, thus reducing recognition accuracy. Summary of the Invention
[0003] The embodiments of this application provide a training and identification method and apparatus for an RFID-based human behavior recognition model, which can at least to some extent eliminate the influence of non-transferable features and domain-specific features, thereby improving the accuracy of human behavior recognition.
[0004] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.
[0005] According to one aspect of the embodiments of this application, a training method for an RFID-based human behavior recognition model is provided, comprising:
[0006] Obtain a target domain training dataset, which includes a signal reception intensity indication sequence and a phase value sequence of the label signals received during different categories of human behavior;
[0007] The signal received strength indication sequence and the phase value sequence are preprocessed respectively to obtain the corresponding signal received strength indication matrix and phase value matrix;
[0008] The signal received strength indication matrix and the phase value matrix are input into the human behavior recognition model pre-trained on the source domain training dataset, so that the human behavior recognition model outputs the corresponding weighted target feature vector. The human behavior recognition model includes a connected feature extraction module and a self-attention module.
[0009] The weighted target feature vector is input into the domain discriminator, so that the domain discriminator outputs the corresponding domain discriminant label, which is used to determine whether the data comes from the source domain or the target domain.
[0010] Batch spectral penalty calculation is performed based on the target feature vectors from the target domain and the source domain, respectively. Based on the calculation results and the domain discrimination label, the human behavior recognition model is optimized to obtain the target human behavior recognition model.
[0011] According to one aspect of the embodiments of this application, an RFID-based human behavior identification method is provided, which is applied to a terminal device. The terminal device is connected to a reader, and the reader is used to receive tag signals from a plurality of target tags, which are respectively set at predetermined positions of the object to be collected.
[0012] The method includes:
[0013] Receive the signal reception strength indication sequence and phase value sequence transmitted by the reader when the tag signal is received;
[0014] Preprocessing is performed on the signal received strength indication sequence and the phase value sequence respectively to obtain the corresponding signal received strength indication matrix and phase value matrix;
[0015] The signal received strength indication matrix and the phase value matrix are input into the target human behavior recognition model so that the target human behavior recognition model outputs the corresponding behavior discrimination result. The target human behavior recognition model is trained by the training method described in the foregoing embodiment.
[0016] According to one aspect of the embodiments of this application, a training device for an RFID-based human behavior recognition model is provided, comprising:
[0017] The acquisition module is used to acquire a target domain training dataset, which includes a signal reception intensity indication sequence and a phase value sequence of the label signals received during different categories of human behavior.
[0018] The first preprocessing module is used to preprocess the signal received strength indication sequence and the phase value sequence respectively to obtain the corresponding signal received strength indication matrix and phase value matrix.
[0019] The first input module is used to input the signal received strength indication matrix and the phase value matrix into the human behavior recognition model pre-trained on the source domain training dataset, so that the human behavior recognition model outputs the corresponding weighted target feature vector. The human behavior recognition model includes a connected feature extraction module and a self-attention module.
[0020] The second input module is used to input the weighted target feature vector into the domain discriminator, so that the domain discriminator outputs the corresponding domain discriminant label, which is used to determine whether the data comes from the source domain or the target domain.
[0021] The processing module is used to perform batch spectral penalty calculation based on the target feature vectors from the target domain and the source domain, respectively, and to optimize the human behavior recognition model based on the calculation results and the domain discrimination label to obtain the target human behavior recognition model.
[0022] According to one aspect of the embodiments of this application, an RFID-based human behavior identification device is provided, which is applied to a terminal device. The terminal device is connected to a reader, and the reader is used to receive tag signals from a plurality of target tags, which are respectively set at predetermined positions of the object to be collected.
[0023] The device includes:
[0024] The receiving module is used to receive the signal reception strength indication sequence and phase value sequence transmitted by the reader when the tag signal is received;
[0025] The second preprocessing module is used to preprocess the signal received strength indication sequence and the phase value sequence respectively to obtain the corresponding signal received strength indication matrix and phase value matrix.
[0026] The recognition module is used to input the signal received strength indication matrix and the phase value matrix into the target human behavior recognition model, so that the target human behavior recognition model outputs the corresponding behavior discrimination result. The target human behavior recognition model is trained by the training method described in the foregoing embodiments.
[0027] According to one aspect of the embodiments of this application, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described in the above embodiments.
[0028] According to one aspect of the embodiments of this application, an electronic device is provided, including: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to perform the method described in the above embodiments.
[0029] According to one aspect of the embodiments of this application, a computer program product or computer program is provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method provided in the above embodiments.
[0030] In some embodiments of this application, the technical solutions involve obtaining a target domain training dataset, which includes signal reception intensity indication sequences and phase value sequences of labeled signals received during different categories of human behavior. The signal reception intensity indication sequences and phase value sequences are preprocessed to obtain corresponding signal reception intensity indication matrices and phase value matrices. These matrices are then input into a human behavior recognition model pre-trained on the source domain training dataset, causing the model to output a corresponding weighted target feature vector. This model includes a connected feature extraction module and a self-attention module. The weighted target feature vector is input into a domain discriminator, causing it to output a corresponding domain discrimination label. This label determines whether the data originates from the source or target domain. Batch spectral penalty calculations are performed based on the target feature vectors from the target and source domains. Finally, the human behavior recognition model is optimized based on the calculation results and the domain discrimination label to obtain the target human behavior recognition model. Therefore, this target human behavior recognition model can adapt to transferable attention and adversarial learning, thereby eliminating the influence of non-transferable features and domain-specific features, and improving the recognition accuracy of human behavior.
[0031] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description
[0032] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings:
[0033] Figure 1 A flowchart illustrating a training method for an RFID-based human behavior recognition model according to an embodiment of this application is shown.
[0034] Figure 2 A block diagram of a training apparatus for an RFID-based human behavior recognition model according to an embodiment of this application is shown;
[0035] Figure 3 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown. Detailed Implementation
[0036] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.
[0037] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.
[0038] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.
[0039] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.
[0040] Figure 1 A flowchart illustrating a training method for an RFID-based human behavior recognition model according to an embodiment of this application is shown. This method can be applied to a terminal device or a server, which may include, but is not limited to, one or more of smartphones, tablets, laptops, and desktop computers. The following description uses the application of this training method to a terminal device as an example.
[0041] like Figure 1 As shown, this training method includes at least steps S110 to S150, which are described in detail below:
[0042] In step S110, a target domain training dataset is obtained, which includes the signal reception intensity indication sequence and phase value sequence of the received label signals during different categories of human behavior.
[0043] The target domain training dataset can be the dataset to be transferred from the deep model. It may differ from the source domain training dataset due to differences in data attributes, data distribution, or collection methods. Unlike the source domain training dataset, the target domain training dataset does not have corresponding behavioral category labels, meaning it does not require manual annotation, thus reducing data acquisition costs.
[0044] This application also provides a data acquisition system, which includes a terminal device, an RFID reader, and several RFID tags. Taking a desktop computer as an example, the desktop computer is communicatively connected to the RFID reader. The RFID reader can receive tag signals from several RFID tags. The several RFID tags can be respectively set at various key points of the object being collected, such as the head, shoulders, hands, knees, and feet, etc.
[0045] It should be noted that the Received Signal Strength Indicator (RSSI) of the tag signal received by the RFID reader attenuates as the distance from the RFID tag increases, and the phase value also changes with the propagation path. Therefore, when a person performs a specific action, the RSSI and phase value received by the RFID reader will exhibit specific patterns. Thus, the RSSI and phase value can provide useful location and action information with low cost and complexity.
[0046] Specifically, to prevent signal interference from multiple RFID readers that could cause collisions and reduce communication efficiency, this data acquisition system uses a single RFID reader. This reader records the signal reception strength (SHS) and phase values for each action and transmits them to a desktop computer for further processing. The desktop computer records the real-time SHS and phase values of each tag. When acquiring the source domain training dataset, the data acquisition personnel can manually extract the SHS and phase value sequences at the time of the action and label them to determine the corresponding behavior category label. For example, behavior category labels may include, but are not limited to, standing, walking, bowing, swinging, and waving. However, labeling is not required when acquiring the target domain training dataset.
[0047] Therefore, according to the aforementioned data acquisition system, the acquisition personnel can obtain the source domain training dataset and the target domain training dataset respectively, and store the source domain training dataset and the target domain training dataset in the local storage space for later use.
[0048] In step S110, the terminal device can obtain the target domain training dataset from its local storage space, or it can collect the dataset in real time, without any special limitation.
[0049] In step S120, the signal reception strength indication sequence and the phase value sequence are preprocessed to obtain the corresponding signal reception strength indication matrix and phase value matrix.
[0050] In one embodiment of this application, preprocessing may include, but is not limited to, at least one of resampling, phase unrolling, filtering and smoothing, and normalization.
[0051] Specifically, for resampling, the data sequence length corresponding to each action can be preset, for example, the data sequence length is 5s, and linear interpolation is performed on each data sequence to achieve a data sequence frequency of 5Hz. It should be noted that the above figures are only exemplary distances, and those skilled in the art can determine the corresponding data sequence length and frequency according to actual implementation needs, without any special limitations.
[0052] Phase expansion, which applies only to the sequence of phase values, removes the periodicity of the phase values, making them continuous and interpretable.
[0053] For filtering and smoothing, existing filters can be used to denoise and filter the data sequence after the aforementioned steps, such as the Savitzky-Golay filter.
[0054] For signal normalization processing, it can normalize the data sequence of each RFID tag.
[0055] Therefore, after the above preprocessing steps, an n×m signal reception strength indication matrix M can be obtained. rssi and phase value matrix M phase Where n is the number of tags, and m is the number of signal reception strength indicators or phase values within a predetermined time length after interpolation. For example, if the predetermined time length is 5s and the frequency is 5Hz, then m is 25, and so on.
[0056] In step S130, the signal received strength indication matrix and the phase value matrix are input to the human behavior recognition model pre-trained on the source domain training dataset, so that the human behavior recognition model outputs the corresponding weighted target feature vector. The human behavior recognition model includes a connected feature extraction module and a self-attention module.
[0057] The human behavior recognition model can be a deep model for human behavior recognition, and it can include a connected feature extraction module and a self-attention module.
[0058] In this embodiment, the terminal device can input the signal reception strength indication matrix and the phase value matrix into the source domain training dataset and the trained human behavior recognition model. The feature extraction module can convert the signal reception strength indication matrix and the phase value matrix into a spatiotemporal feature stream with higher level and richer information (i.e., the third feature vector described later). Then, the self-attention module weights the features of the spatiotemporal feature stream, which can effectively reduce the influence of non-transferable features. Furthermore, the self-attention mechanism preserves the original input tensor, which can reduce information loss and improve the robustness of the model.
[0059] In one embodiment of this application, the feature extraction module performs feature extraction on the signal received strength indication matrix and the phase value matrix, including:
[0060] The signal received strength indication matrix and the phase value matrix are respectively input into multiple convolutional layers to obtain the corresponding first feature vector;
[0061] The first feature vector corresponding to the signal received strength indication matrix and the first feature vector corresponding to the phase value matrix are concatenated to obtain the second feature vector. The second feature vector is then input into the dual-layer GRU so that the dual-layer GRU outputs the corresponding third feature vector as the input of the self-attention module.
[0062] In this embodiment, the feature extraction module can pass the signal received intensity indication matrix and the phase value matrix through their respective multi-layer convolutional layers, preferably three layers, with each layer using a 1*3 convolutional kernel. Furthermore, batch normalization can be employed during the convolution process to prevent overfitting during training. It should be understood that multiple convolutional layers can effectively extract spatial information from the signal.
[0063] Next, the first feature vectors output by the two multi-layer convolutional layers can be concatenated to obtain the concatenated second feature vector. This second feature vector is then input into a two-layer GRU (gated recurrent unit) to extract temporal information and output a third feature vector as the input to the self-attention module.
[0064] In one embodiment, the aforementioned third feature vector is input into a self-attention module. The self-attention module can pass the third feature vector through a linear layer to obtain three terms: Q, K, and V, and then obtain an attention map, as shown in the following formula:
[0065]
[0066] Where, d k Let K be the dimension.
[0067] Then, V is multiplied by the attention map above, and the original input x (i.e., the third feature vector) is added to obtain the final output of the self-attention module, Output (i.e., the target feature vector), as shown in the following formula:
[0068] Output(x, Q, K, V)=x+AttentionMap(Q, K).
[0069] Therefore, by using a self-attention module to weight the features extracted by the feature extraction module, the influence of non-transferable features can be effectively mitigated.
[0070] Please continue to refer to this. Figure 1 In step S140, the weighted target feature vector is input to the domain discriminator so that the domain discriminator outputs the corresponding domain discriminant label, which is used to determine whether the data comes from the source domain or the target domain.
[0071] In this embodiment, the terminal device can input the weighted target feature vector into the domain discriminator so that the domain discriminator can output a domain discrimination label, which is used to determine whether the data comes from the source domain or the target domain.
[0072] Understandably, the source domain training dataset is a training dataset with a large number of behavior category labels. A human behavior recognition model can be trained on the source domain training dataset to produce parameters that perform well on this training dataset. However, directly using a human behavior recognition model trained on the source domain training dataset for prediction on the target domain training dataset will not yield good results. This is because the human behavior recognition model may overfit the data in the source domain and fail to extract features from the data in the target domain well. However, since there are no behavior category labels in the target domain, the human behavior recognition model cannot be trained directly on the target domain.
[0073] To address this, a domain discriminator is introduced. In one example, this discriminator includes a user domain discriminator and an environment domain discriminator. The user domain discriminator determines whether the current input target feature vector and the source domain training dataset belong to the same data collection object, while the environment domain discriminator determines whether the current input target feature vector and the source domain training dataset belong to the same data collection environment. Thus, even when the target domain lacks a corresponding behavior category label, the user domain discriminator and environment domain discriminator can simultaneously assign additional labels—user domain labels and environment domain labels—to the data in both the source and target domains. At this point, each data point, including those from both the source and target domains, has two additional, distinct labels. Therefore, through this training process, the human behavior recognition model can gradually master the features of data from both the source and target domains, helping to achieve better results in classification tasks.
[0074] In step S150, batch spectral penalty calculation is performed based on the target feature vectors from the target domain and the source domain, respectively. Based on the calculation results and the domain discrimination label, the human behavior recognition model is optimized to obtain the target human behavior recognition model.
[0075] In this embodiment, a batch spectral penalty can be calculated based on the target feature vectors from the target domain and the source domain. This batch spectral penalty consists of singular value decomposition, which can obtain the maximum singular value from the source feature matrix and the target feature matrix to determine the loss of the corresponding batch spectral penalty. By adjusting the amplitude of the feature matrix, the overfitting of the human behavior recognition model can be mitigated, thereby improving the generalization ability of the human behavior recognition model.
[0076] Furthermore, based on the domain-discriminative label, the cross-entropy loss between it and the true label can be determined. Based on this loss, the gradient of each parameter in the model can be calculated using the backpropagation algorithm for updating.
[0077] Therefore, based on the calculation results of batch spectral penalty and domain discrimination labels, the human behavior recognition model can be optimized. That is, by adjusting the parameters of the human behavior recognition model, the accuracy of the recognition results of the human behavior recognition model can be improved, so as to obtain the target human behavior recognition model.
[0078] In another embodiment of this application, a method for training a human behavior recognition model is provided, the method comprising the following steps:
[0079] S1: Set the data collected in a certain environment and by volunteers as the source domain data, and set the data collected in other environments or by other volunteers as the target domain data.
[0080] S2: Source domain data or target domain data are processed by a feature extractor and a self-attention module to obtain f′. source or f′ target .
[0081] S3: f′ source The predicted behavior labels are obtained through a behavior predictor, and the loss L is calculated by comparing them with the actual behavior labels. y The loss is calculated using cross-entropy loss.
[0082] S4: Place a pair of f′ source and f′ target User domain labels and environment domain labels are obtained separately using a user domain discriminator and an environment domain discriminator, and then compared with the actual user domain labels or environment domain labels to obtain the loss. and Summing them again gives L d ,
[0083] S5: Place a pair of f′ source and f′ target Used to calculate batch spectral penalty
[0084] Specifically, the singular value matrix ∑ is calculated for both. source and ∑ target Then, extract the i-th largest value β from each of them. s，i and β t，i， Summing their k largest values yields the BSP loss.
[0085]
[0086] S6: The total loss of this model is
[0087] L = L y -αL d +βL bsp
[0088] α and β are hyperparameters used in the loss of the balancing discriminator and attention transfer.
[0089] Based on this total loss, the gradients of each parameter in the model can be calculated using the backpropagation algorithm to update the gradients and thus complete the training.
[0090] Therefore, the human behavior recognition model trained by the above training method can eliminate the influence of non-transferable features and domain-specific features, thereby improving the accuracy of human behavior recognition.
[0091] In one embodiment of this application, a human behavior recognition method based on RFID is also provided. This method can be applied to a terminal device, which is connected to a reader. The reader is used to receive tag signals from a plurality of target tags, which are respectively set at predetermined positions of the object to be collected.
[0092] The method includes:
[0093] Receive the signal reception strength indication sequence and phase value sequence transmitted by the reader when the tag signal is received;
[0094] Preprocessing is performed on the signal received strength indication sequence and the phase value sequence respectively to obtain the corresponding signal received strength indication matrix and phase value matrix;
[0095] The signal received strength indication matrix and the phase value matrix are input into the target human behavior recognition model so that the target human behavior recognition model outputs the corresponding behavior discrimination result. The target human behavior recognition model is trained by the training method described in the foregoing embodiment.
[0096] Therefore, by using the human behavior recognition model obtained through the training method described in the foregoing embodiments, the influence of non-transferable features and domain-specific features can be eliminated, thereby improving the accuracy of human behavior recognition. It should be noted that for details not disclosed in the embodiments of the RFID-based human behavior recognition method of this application, please refer to the embodiments of the RFID-based human behavior recognition model training method described above.
[0097] The following describes an embodiment of the apparatus described in this application, which can be used to execute the training method for the RFID-based human behavior recognition model or the RFID-based human behavior recognition method described in the above embodiments of this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the training method for the RFID-based human behavior recognition model or the RFID-based human behavior recognition method described above in this application.
[0098] Figure 2 A block diagram of a training apparatus for an RFID-based human behavior recognition model according to an embodiment of this application is shown.
[0099] Reference Figure 2 As shown, a training device for an RFID-based human behavior recognition model according to an embodiment of this application includes:
[0100] The acquisition module is used to acquire a target domain training dataset, which includes a signal reception intensity indication sequence and a phase value sequence of the label signals received during different categories of human behavior.
[0101] The first preprocessing module is used to preprocess the signal received strength indication sequence and the phase value sequence respectively to obtain the corresponding signal received strength indication matrix and phase value matrix.
[0102] The first input module is used to input the signal received strength indication matrix and the phase value matrix into the human behavior recognition model pre-trained on the source domain training dataset, so that the human behavior recognition model outputs the corresponding weighted target feature vector. The human behavior recognition model includes a connected feature extraction module and a self-attention module.
[0103] The second input module is used to input the weighted target feature vector into the domain discriminator, so that the domain discriminator outputs the corresponding domain discriminant label, which is used to determine whether the data comes from the source domain or the target domain.
[0104] The processing module is used to perform batch spectral penalty calculation based on the target feature vectors from the target domain and the source domain, respectively, and to optimize the human behavior recognition model based on the calculation results and the domain discrimination label to obtain the target human behavior recognition model.
[0105] In one embodiment of this application, the feature extraction module performs feature extraction on the signal received strength indication matrix and the phase value matrix, including: inputting the signal received strength indication matrix and the phase value matrix into multiple convolutional layers respectively to obtain corresponding first feature vectors; concatenating the first feature vector corresponding to the signal received strength indication matrix and the first feature vector corresponding to the phase value matrix to obtain a second feature vector, and inputting the second feature vector into a two-layer GRU so that the two-layer GRU outputs a corresponding third feature vector as the input of the self-attention module.
[0106] In one embodiment of this application, the preprocessing includes at least one of resampling, phase unwrapping, filtering and smoothing, and normalization.
[0107] In one embodiment of this application, the domain discriminator includes a user domain discriminator and an environment domain discriminator. The user domain discriminator is used to determine whether it is the same collection object as the source domain training dataset, and the environment domain discriminator is used to determine whether it is the same collection environment as the source domain training data.
[0108] In one embodiment of this application, an RFID-based human behavior recognition device is also provided. The device is applied to a terminal device, which is connected to a reader. The reader is used to receive tag signals from a plurality of target tags, which are respectively set at predetermined positions of the object to be collected.
[0109] The device includes:
[0110] The receiving module is used to receive the signal reception strength indication sequence and phase value sequence transmitted by the reader when the tag signal is received;
[0111] The second preprocessing module is used to preprocess the signal received strength indication sequence and the phase value sequence respectively to obtain the corresponding signal received strength indication matrix and phase value matrix.
[0112] The recognition module is used to input the signal received strength indication matrix and the phase value matrix into the target human behavior recognition model, so that the target human behavior recognition model outputs the corresponding behavior discrimination result, wherein the target human behavior recognition model is trained by the training method of any one of claims 1-4.
[0113] Figure 3 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown.
[0114] It should be noted that, Figure 3The computer system of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.
[0115] like Figure 3 As shown, the computer system includes a Central Processing Unit (CPU) 301, which can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 302 or programs loaded from storage portion 308 into Random Access Memory (RAM) 303, such as performing the methods described in the above embodiments. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An Input / Output (I / O) interface 305 is also connected to the bus 304.
[0116] The following components are connected to I / O interface 305: an input section 306 including a keyboard, mouse, etc.; an output section 307 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.
[0117] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by central processing unit (CPU) 301, it performs various functions defined in the system of this application.
[0118] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. The transmitted data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.
[0119] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0120] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.
[0121] In another aspect, this application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the methods described in the above embodiments.
[0122] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.
[0123] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the method according to the embodiments of this application.
[0124] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.
[0125] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.
Claims
1. A training method for a human behavior recognition model based on RFID, characterized in that, include: Obtain a target domain training dataset, which includes a signal reception intensity indication sequence and a phase value sequence of the label signals received during different categories of human behavior; The signal received strength indication sequence and the phase value sequence are preprocessed respectively to obtain the corresponding signal received strength indication matrix and phase value matrix; The signal received strength indication matrix and the phase value matrix are input into the human behavior recognition model pre-trained on the source domain training dataset, so that the human behavior recognition model outputs the corresponding weighted target feature vector. The human behavior recognition model includes a connected feature extraction module and a self-attention module. The weighted target feature vector is input into a domain discriminator so that the domain discriminator outputs a corresponding domain discriminant label, which is used to determine whether the data comes from the source domain or the target domain. Batch spectral penalty calculation is performed based on the target feature vectors from the target domain and the source domain respectively, and the human behavior recognition model is optimized based on the calculation results and the domain discrimination label to obtain the target human behavior recognition model. Batch spectral penalty is composed of singular value decomposition, which can obtain the maximum singular value from the source feature matrix and the target feature matrix to determine the loss of the corresponding batch spectral penalty. By adjusting the amplitude of the feature matrix, the overfitting of the human behavior recognition model can be mitigated, thereby improving the generalization ability of the human behavior recognition model. Furthermore, the cross-entropy loss between the domain-discriminative label and the true label is determined, and the gradient of each parameter in the model is calculated using the backpropagation algorithm based on this loss for updating. Therefore, based on the calculation results of batch spectral penalty and domain discrimination labels, the human behavior recognition model is optimized. That is, by adjusting the parameters of the human behavior recognition model, the accuracy of the recognition results of the human behavior recognition model is improved, so as to obtain the target human behavior recognition model.
2. The method according to claim 1, characterized in that, The feature extraction module performs feature extraction on the signal received strength indication matrix and the phase value matrix, including: The signal received strength indication matrix and the phase value matrix are respectively input into multiple convolutional layers to obtain the corresponding first feature vector; The first feature vector corresponding to the signal received strength indication matrix and the first feature vector corresponding to the phase value matrix are concatenated to obtain the second feature vector. The second feature vector is then input into the dual-layer GRU so that the dual-layer GRU outputs the corresponding third feature vector as the input of the self-attention module.
3. The method according to claim 1, characterized in that, The preprocessing includes at least one of resampling, phase unrolling, filtering and smoothing, and normalization.
4. The method according to any one of claims 1-3, characterized in that, The domain discriminator includes a user domain discriminator and an environment domain discriminator. The user domain discriminator is used to determine whether it is the same collection object as the source domain training dataset, and the environment domain discriminator is used to determine whether it is the same collection environment as the source domain training data.
5. A human behavior recognition method based on RFID, characterized in that, The method is applied to a terminal device, which is connected to a reader. The reader is used to receive tag signals from several target tags, which are respectively set at predetermined positions of the object to be collected. The method includes: Receive the signal reception strength indication sequence and phase value sequence transmitted by the reader when the tag signal is received; Preprocessing is performed on the signal received strength indication sequence and the phase value sequence respectively to obtain the corresponding signal received strength indication matrix and phase value matrix; The signal received strength indication matrix and the phase value matrix are input into the target human behavior recognition model so that the target human behavior recognition model outputs the corresponding behavior discrimination result. The target human behavior recognition model is trained by the training method of any one of claims 1-4.
6. A training device for a human behavior recognition model based on RFID, characterized in that, include: The acquisition module is used to acquire a target domain training dataset, which includes a signal reception intensity indication sequence and a phase value sequence of the label signals received during different categories of human behavior. The first preprocessing module is used to preprocess the signal received strength indication sequence and the phase value sequence respectively to obtain the corresponding signal received strength indication matrix and phase value matrix. The first input module is used to input the signal received strength indication matrix and the phase value matrix into the human behavior recognition model pre-trained on the source domain training dataset, so that the human behavior recognition model outputs the corresponding weighted target feature vector. The human behavior recognition model includes a connected feature extraction module and a self-attention module. The second input module is used to input the weighted target feature vector into the domain discriminator, so that the domain discriminator outputs the corresponding domain discriminant label, which is used to determine whether the data comes from the source domain or the target domain. The processing module is used to perform batch spectral penalty calculation based on the target feature vectors from the target domain and the source domain respectively, and to optimize the human behavior recognition model based on the calculation results and the domain discrimination label to obtain the target human behavior recognition model. Batch spectral penalty is composed of singular value decomposition, which can obtain the maximum singular value from the source feature matrix and the target feature matrix to determine the loss of the corresponding batch spectral penalty. By adjusting the amplitude of the feature matrix, the overfitting of the human behavior recognition model can be mitigated, thereby improving the generalization ability of the human behavior recognition model. Furthermore, the cross-entropy loss between the domain-discriminative label and the true label is determined, and the gradient of each parameter in the model is calculated using the backpropagation algorithm based on this loss for updating. Therefore, based on the calculation results of batch spectral penalty and domain discrimination labels, the human behavior recognition model is optimized. That is, by adjusting the parameters of the human behavior recognition model, the accuracy of the recognition results of the human behavior recognition model is improved, so as to obtain the target human behavior recognition model.
7. The apparatus according to claim 6, characterized in that, The feature extraction module performs feature extraction on the signal received strength indication matrix and the phase value matrix, including: The signal received strength indication matrix and the phase value matrix are respectively input into multiple convolutional layers to obtain the corresponding first feature vector; The first feature vector corresponding to the signal received strength indication matrix and the first feature vector corresponding to the phase value matrix are concatenated to obtain the second feature vector. The second feature vector is then input into the dual-layer GRU so that the dual-layer GRU outputs the corresponding third feature vector as the input of the self-attention module.
8. A human behavior recognition device based on RFID, characterized in that, The method is applied to a terminal device, which is connected to a reader. The reader is used to receive tag signals from several target tags, which are respectively set at predetermined positions of the object to be collected. The device includes: The receiving module is used to receive the signal reception strength indication sequence and phase value sequence transmitted by the reader when the tag signal is received; The second preprocessing module is used to preprocess the signal received strength indication sequence and the phase value sequence respectively to obtain the corresponding signal received strength indication matrix and phase value matrix. The recognition module is used to input the signal received strength indication matrix and the phase value matrix into the target human behavior recognition model, so that the target human behavior recognition model outputs the corresponding behavior discrimination result, wherein the target human behavior recognition model is trained by the training method of any one of claims 1-4.
9. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 5.
10. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 5.