Meta-learning-based cross-domain RFID human behavior recognition method and device

Abstract

This invention discloses a cross-domain RFID human behavior recognition method and apparatus based on meta-learning. The method includes: acquiring human behavior signals corresponding to each domain; preprocessing the human behavior signals; generating a task sample set based on the preprocessed data corresponding to each domain, and dividing the task sample set into a support set and a query set; constructing a relation network, and inputting the support set and query set into the relation network for training to obtain a trained human behavior recognition model. The relation network includes an embedding module and a relation module. The embedding module is used to acquire the time-frequency features of the query samples and support samples, and the relation module is used to identify the matching category of the query sample in the support set. The human behavior recognition model is used to identify the human behavior signal to be identified to obtain the corresponding recognition category. Thus, the improved relation network can be quickly adjusted and adapted with only a small amount of new data, thereby improving the recognition effect.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a cross-domain RFID human behavior recognition method based on meta-learning, a computer-readable storage medium, a computer device, and a cross-domain RFID human behavior recognition device based on meta-learning. Background Technology

[0002] Among related technologies, Human Activity Recognition (HAR), as a promising technology, has demonstrated great potential in real-world applications such as smart homes, contactless human-vehicle interaction, and fall detection for the elderly. Existing HAR methods can be broadly categorized into three types: vision-based methods, sensor-based methods, and wireless sensing-based methods. Vision-based methods are primarily sensitive to lighting conditions and raise privacy concerns. Sensor-based methods require users to wear sensors (such as accelerometers or gyroscopes), which can affect user comfort. Therefore, wireless sensing-based methods have attracted increasing research attention due to their robustness to lighting conditions and ease of use. Compared to WiFi-based wireless sensing systems, Radio Frequency Identification (RFID) systems are gaining popularity due to their ability to identify multiple users with unique ID tags, low-cost deployment, and long-distance communication.

[0003] In RFID systems, a challenging problem is addressing the domain transfer problem to ensure reliable sensing even when test data (target domain) differs from training data (source domain). Specifically, the domain transfer problem is primarily caused by environmental differences and subject diversity, including imbalanced tag distributions and heterogeneous signal distributions across domains. First, each user does not necessarily perform the same set of activities. For example, due to differences in etiquette, some users may only perform {bowing, standing}, while others may only perform {waving, walking}. Therefore, tag distributions exhibit imbalance across different domain datasets. Second, RFID signals are linear superpositions of signals from different paths, including direct signals and signals reflected from static furniture and moving people. Therefore, even if users perform the same activities in different environments, or different users perform the same activities in the same environment, the collected signal distributions are heterogeneous. The domain transfer problem severely impacts the robustness of RFID-based identification systems; performance can degrade drastically when a model trained in the source domain is applied to the target domain.

[0004] One solution for cross-domain recognition is through transfer learning, which involves fine-tuning a pre-trained model using labeled data from the target domain. These works assume shared general knowledge between different domains. However, negative transfer can occur when the source and target domains differ significantly, leading to the failure of transfer learning. Recently, unsupervised adversarial learning has made significant progress in cross-domain recognition, eliminating environmental influences and transferring activity-related features. However, feature discriminability can unexpectedly deteriorate when adversarial learning transfers features. Furthermore, both approaches require time-consuming and laborious collection of sufficient samples from both the source and target domains to achieve high accuracy in accurate recognition. Summary of the Invention

[0005] This invention aims to at least partially solve one of the technical problems in the aforementioned technologies. To this end, one objective of this invention is to propose a cross-domain RFID human behavior recognition method based on meta-learning. By leveraging the rapid adaptation capability of meta-learning to new domains, the behavior recognition of each user's human activity set is treated as a task, enabling the improved relationship network to quickly adjust and adapt with only a small amount of new data, thereby improving recognition performance.

[0006] A second objective of this invention is to provide a computer-readable storage medium.

[0007] The third objective of this invention is to provide a computer device.

[0008] The fourth objective of this invention is to propose a cross-domain RFID human behavior recognition device based on meta-learning.

[0009] To achieve the above objectives, a first aspect of the present invention proposes a cross-domain RFID human behavior recognition method based on meta-learning. The method includes acquiring human behavior signals corresponding to each domain collected by an RFID device, wherein multiple human behavior signals collected under the same target and environment are defined as the same domain, and each human behavior signal includes the phase of the tag and the received signal strength indication; preprocessing the human behavior signals, and generating a task sample set based on the preprocessed data corresponding to each domain, and dividing the task sample set into a support set and a query set; constructing a relational network, and inputting the support set and the query set into the relational network for training to obtain a trained human behavior recognition model, wherein the relational network includes an embedding module and a relational module, the embedding module being used to acquire the time-frequency features of the query samples and support samples, and the relational module being used to identify the matching category corresponding to the query sample in the support set; acquiring the human behavior signal to be identified and the support set, and inputting the human behavior signal to be identified and the support set into the human behavior recognition model to obtain the human behavior recognition result corresponding to the human behavior signal to be identified.

[0010] According to an embodiment of the present invention, a cross-domain RFID human behavior recognition method based on meta-learning firstly acquires human behavior signals corresponding to each domain collected by an RFID device. Multiple human behavior signals collected under the same target and environment are defined as the same domain, and each human behavior signal includes the tag's phase and the received signal strength indication. Then, the human behavior signals are preprocessed, and a task sample set is generated based on the preprocessed data corresponding to each domain. The task sample set is then divided into a support set and a query set. Next, a relational network is constructed, and the support set and query set are input into the relational network for training to obtain a trained human behavior recognition model. The model comprises an embedding module and a relation module. The embedding module acquires the time-frequency features of the query sample and support samples, while the relation module identifies the matching category of the query sample in the support set. Finally, the model acquires the human behavior signal to be identified and the support set, and inputs them into the human behavior recognition model to obtain the human behavior recognition result corresponding to the human behavior signal to be identified. Thus, by leveraging the ability of meta-learning to quickly adapt to new domains, the behavior recognition of each user's human activity set is treated as a task, enabling the improved relation network to quickly adjust and adapt with only a small amount of new data, thereby improving the recognition effect.

[0011] In addition, the cross-domain RFID human behavior recognition method based on meta-learning proposed in the above embodiments of the present invention may also have the following additional technical features:

[0012] Optionally, the human behavior signal is preprocessed, including: performing one-dimensional phase expansion on the original phase of all tags; performing linear interpolation on the signal strength indication to obtain a resampled signal stream; using a filter to smooth the resampled signal stream; and constructing a DS stream for the signal stream corresponding to each tag after filtering and smoothing, wherein the DS stream calculates the difference between two consecutive data values ​​in the signal stream.

[0013] Optionally, generating a task sample set based on the data corresponding to each preprocessed domain includes: constructing a domain list corresponding to each human behavior signal, wherein the domain list includes multiple domains, and each domain includes a corresponding human behavior signal; when generating a task sample corresponding to each human behavior signal, randomly selecting a domain from the corresponding domain list and randomly selecting a human behavior signal from the selected domain to generate a task sample set.

[0014] Optionally, the embedding module includes a self-calibration module, a backbone network, and a cross-calibration module. The self-calibration module uses a non-local attention mechanism to capture the dependencies between the signal streams corresponding to each label in the sample, so as to retain activity-related information and mitigate environment-related interference. The backbone network extracts time-frequency features from activity-related information. The cross-calibration module uses a pairwise similarity module to capture query-support association masks and generates a set of calibrated support embeddings for each query sample.

[0015] Optionally, the backbone network includes three CNN blocks and a max pooling layer, wherein each CNN block includes 32 filters, 3×3 convolutions, followed by batch normalization and ReLU nonlinearity.

[0016] Optionally, the embedding module is trained according to the following loss function to learn the similarity of embedded features through a nonlinear subspace and to cluster features of the same category:

[0017]

[0018] Among them, T i Let θ represent the i-th task, θ represent the network parameters of the embedded module, and N represent the n-th task. B Represents the number of paired samples, and represents, II m,n This indicates whether samples i and j belong to the same activity category. Specifically, if samples i and j belong to the same activity category, the value is 1; if samples i and j do not belong to the same activity category, the value is 0. σ(η) i.j ) represents the cosine distance between the feature vectors of samples i and j obtained through the embedding network.

[0019] Optionally, the relation network can be trained according to the following loss function:

[0020]

[0021] Among them, T i Let θ represent the i-th task, and let θ represent the network parameters of the embedded module. Represents the network parameters of the relational module, II m,n This indicates whether the support sample n and the query sample m belong to the same sample. Specifically, if the two samples have the same label, the value is 1; otherwise, the value is 0. This represents the probability that the query sample m belongs to the category to which the supporting sample n belongs.

[0022] To achieve the above objectives, a second aspect of the present invention provides a computer-readable storage medium storing a meta-learning-based cross-domain RFID human behavior recognition program, which, when executed by a processor, implements the meta-learning-based cross-domain RFID human behavior recognition method as described above.

[0023] 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 meta-learning-based cross-domain RFID human behavior recognition method as described above.

[0024] To achieve the above objectives, a fourth aspect of the present invention proposes a cross-domain RFID human behavior recognition device based on meta-learning, comprising: an acquisition module for acquiring human behavior signals corresponding to each domain collected by an RFID device, wherein multiple human behavior signals collected under the same target and environment are defined as the same domain, and each human behavior signal includes the phase of the tag and the received signal strength indication; a preprocessing module for preprocessing the human behavior signals and generating a task sample set based on the data corresponding to each domain after preprocessing, and dividing the task sample set into a support set and a query set; a training module for constructing a relational network and inputting the support set and the query set into the relational network for training to obtain a trained human behavior recognition model, wherein the relational network includes an embedding module and a relational module, the embedding module for acquiring the time-frequency features of the query samples and support samples, and the relational module for identifying the matching category corresponding to the query sample in the support set; and a human behavior recognition module for acquiring the human behavior signal to be recognized and the support set, and inputting the human behavior signal to be recognized and the support set into the human behavior recognition model to obtain the human behavior recognition result corresponding to the human behavior signal to be recognized.

[0025] According to an embodiment of the present invention, a cross-domain RFID human behavior recognition device based on meta-learning acquires human behavior signals corresponding to each domain collected by an RFID device through an acquisition module. Multiple human behavior signals collected under the same target and environment are defined as the same domain, and each human behavior signal includes the tag's phase and the received signal strength indication. A preprocessing module preprocesses the human behavior signals and generates a task sample set based on the preprocessed data corresponding to each domain, dividing the task sample set into a support set and a query set. A training module constructs a relational network and inputs the support set and query set into the relational network for training to obtain a trained human behavior recognition model. The model includes an embedding module and a relation module. The embedding module is used to acquire the time-frequency features of query samples and support samples, and the relation module is used to identify the matching category of the query sample in the support set. The human behavior recognition module acquires the human behavior signal to be recognized and the support set, and inputs the human behavior signal to be recognized and the support set into the human behavior recognition model to obtain the human behavior recognition result corresponding to the human behavior signal to be recognized. Thus, through the ability of meta-learning to quickly adapt to new domains, the behavior recognition of each user's human activity set is regarded as a task, so that the improved relation network can be quickly adjusted and adapted with only a small amount of new data, thereby improving the recognition effect. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the cross-domain RFID human behavior recognition method based on meta-learning according to an embodiment of the present invention.

[0027] Figure 2 This is a schematic diagram of the data acquisition process of a radio frequency identification device according to an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram of a task generation strategy according to an embodiment of the present invention;

[0029] Figure 4 This is a schematic diagram of the overall framework of the relationship network according to an embodiment of the present invention;

[0030] Figure 5 This is a schematic diagram of the frame of the embedded module according to an embodiment of the present invention;

[0031] Figure 6 This is a block diagram of a cross-domain RFID human behavior recognition device based on meta-learning according to an embodiment of the present invention. Detailed Implementation

[0032] 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.

[0033] 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.

[0034] 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.

[0035] Figure 1 This is a flowchart illustrating a cross-domain RFID human behavior recognition method based on meta-learning according to an embodiment of the present invention, as shown below. Figure 1 As shown, this meta-learning-based cross-domain RFID human behavior recognition method includes the following steps:

[0036] S101, acquire the human behavior signal corresponding to each domain collected by the radio frequency identification device, wherein multiple human behavior signals collected under the same target and environment are defined as the same domain, and each human behavior signal includes the phase of the tag and the received signal strength indication.

[0037] It should be noted that the RFID device includes nine Monza AZ-9654 RFID passive tags, one Impinj Speedway R420 reader, one HAUHF2599 antenna (902-928MHz), and a laptop computer. The reader collects RF signals at a fixed frequency of 920.625Hz with an antenna power of 20dBm. The default mode of the reader is set to MaxThroughput mode.

[0038] In other words, such as Figure 2As shown, the aforementioned radio frequency identification (RFID) device collects human behavior signals corresponding to each domain. In a specific embodiment, the dataset collected by the aforementioned RFID device includes eight human activities of five subjects in a classroom, conference room, and multi-purpose room. Based on experience, it takes about 3 seconds to complete an activity, so the RF signals collected within 5 seconds are considered as one activity sample. For each activity, each subject repeats it 30 times in each environment according to its own understanding, without guidance. Therefore, all subjects have the same set of activities, which means that the label distribution of the collected dataset is balanced. To simulate the case of an imbalanced label distribution, some activities are randomly removed from each domain.

[0039] S102, preprocess the human behavior signal, generate a task sample set based on the data corresponding to each preprocessed domain, and divide the task sample set into a support set and a query set.

[0040] As an example, the preprocessing of the human behavior signal includes: performing one-dimensional phase expansion on the original phase of all tags; performing linear interpolation on the signal strength indication to obtain a resampled signal stream; using a filter to smooth the resampled signal stream; and constructing a DS stream for the signal stream corresponding to each tag after filtering and smoothing, wherein the DS stream calculates the difference between two consecutive data values ​​in the signal stream.

[0041] In other words, due to environmental noise caused by multipath effects, the original phase and Received Signal Strength Indicator (RSSI) values ​​cannot be used directly and require preprocessing. First, the original phase of all tags is expanded using a one-dimensional phase expansion method. Second, to address the inconsistency in tag readings caused by tag collisions, linear interpolation is used to resample the signal stream to a frequency of 4.4 Hz. After resampling, a Savitzky-Colay filter is used to mitigate static interference caused by static objects and smooth the signal stream. Finally, a DS stream is constructed for the signal stream corresponding to each tag, and the DS stream calculates the difference between two consecutive data values ​​in the signal stream.

[0042] As one embodiment, generating a task sample set based on the preprocessed data corresponding to each domain includes: constructing a domain list corresponding to each human behavior signal, wherein the domain list includes multiple domains, and each domain includes a corresponding human behavior signal; when generating a task sample corresponding to each human behavior signal, randomly selecting a domain from the corresponding domain list, and randomly selecting a human behavior signal from the selected domain to generate a task sample set.

[0043] In other words, such as Figure 3As shown, for the (n-th)th activity, the corresponding domain list (DL) is first constructed. n This includes all domains related to the (n-th)th activity, and then, in generating task T... i When sampling the (n-th)th activity, a domain is randomly selected from the corresponding domain list, and an activity is randomly selected from the selected domain. Finally, the support set and query set are divided based on the generated sample set.

[0044] S103, construct a relation network and input the support set and query set into the relation network for training to obtain a trained human behavior recognition model. The relation network includes an embedding module and a relation module. The embedding module is used to obtain the time-frequency features of the query samples and support samples, and the relation module is used to identify the matching category of the query sample in the support set.

[0045] It should be noted that, as Figure 4-5 As shown, an improved relational network was carefully designed to enable it to identify radio frequency signals caused by different activities; this network includes a suitable embedded module. Used to extract time-frequency features of query samples and support samples, and a relational module. Used to compare their embeddings and determine whether a query sample comes from a matching category in the support set.

[0046] As one embodiment, the embedding module includes a self-calibration module, a backbone network, and a cross-calibration module. The self-calibration module uses a non-local attention mechanism to capture the dependencies between the signal streams corresponding to each label in the sample, so as to retain activity-related information and mitigate environment-related interference. The backbone network extracts time-frequency features from activity-related information. The cross-calibration module uses a pairwise similarity module to capture query-support association masks and generates a set of calibrated support embeddings for each query sample.

[0047] In addition, the backbone network consists of three CNN blocks and a max pooling layer, where each CNN block consists of 32 filters, 3×3 convolutions, followed by batch normalization and ReLU nonlinearity.

[0048] In other words, embedded module It mainly consists of three parts: self-calibration module backbone network and cross-calibration modules Self-calibration module Non-local attention mechanisms are used to capture the dependencies between signal streams corresponding to each label in a sample, thus preserving activity-related information and mitigating environment-related interference. Backbone network Further extraction of time-frequency features from activity-related information. The backbone network consists of three CNN blocks, each of which is a 3×3 convolution with 32 filters, followed by batch normalization and ReLU nonlinearity. The first CNN block includes a 2×2 max-pooling layer, while the latter two do not. (Cross-calibration module) A pairwise similarity module was used to capture query-support association masks and generate a set of calibrated support embeddings for each query sample.

[0049] As an example, the embedding module is trained according to the following loss function to learn the similarity of embedded features through a nonlinear subspace and to cluster features of the same category:

[0050]

[0051] Among them, T i Let θ represent the i-th task, θ represent the network parameters of the embedded module, and N represent the n-th task. B Represents the number of paired samples, and represents, II m,n This indicates whether samples i and j belong to the same activity category. Specifically, if samples i and j belong to the same activity category, the value is 1; if samples i and j do not belong to the same activity category, the value is 0. σ(η) i.j ) represents the cosine distance between the feature vectors of samples i and j obtained through the embedding network.

[0052] As an example, the relation network is trained according to the following loss function:

[0053]

[0054] Among them, T i Let θ represent the i-th task, and let θ represent the network parameters of the embedded module. Represents the network parameters of the relational module, II m,n This indicates whether the support sample n and the query sample m belong to the same sample. Specifically, if the two samples have the same label, the value is 1; otherwise, the value is 0. This represents the probability that the query sample m belongs to the category to which the supporting sample n belongs.

[0055] In other words, in order to minimize the intra-class distance of the embedded feature vectors and maximize the inter-class distance of the embedded feature vectors, the loss function of this application includes the cross-entropy loss function and the pairwise loss function.

[0056] S104, acquire the human behavior signal to be identified and the support set, and input the human behavior signal to be identified and the support set into the human behavior recognition model to obtain the human behavior recognition result corresponding to the human behavior signal to be identified.

[0057] Therefore, this application utilizes an improved Relation Network (RN) to identify eight common activities, each requiring only a small number of samples. First, the RN is a metric-based few-shot learning network, originally designed for few-shot image classification. The RN learns the relationships between samples of different categories through a feature embedding network and performs category matching through a learnable nonlinear comparator. However, due to the environmental sensitivity of RFID time series, RFID signal identification differs significantly from image identification. To adapt to the characteristics of RFID time series, the original signal is preprocessed, and a suitable feature embedding network and loss function are designed. Second, how to train a robust model with a limited number of source domains is a problem to consider. A task (also called an episode) is defined as a multi-class classification problem to predict activities in signals collected from a specific domain. To address this challenge, an episode-based task training strategy is adopted, and a task generation strategy is proposed. By simulating virtual new tasks, the model experiences various tasks, enabling rapid adaptation to new domains through several fine-tunings and a small number of samples. Specifically, a combination of an environment and an agent is considered a domain, and a real-world dataset comprising 15 domains, each containing eight activities, is collected. The proposed system was evaluated on the aforementioned dataset using leave-one-field cross-validation. Results show that, compared to transfer learning under the same settings, the improved relational network of this application improves the average accuracy from 77.34% and 80.04% to 90.96% and 94.32% for each activity with one and five samples, respectively. Compared to several existing state-of-the-art few-shot learning methods, the improved relational network achieves accuracy improvements of at least 7.37% and 4.04%. To simulate a dataset with an imbalanced label distribution, several activities in each domain were randomly removed. On the newly generated dataset, the improved relational network achieved recognition accuracies of 90.83% and 94.27% for each activity with one and five samples, respectively, while the existing baseline methods achieved accuracies of 68.64% and 86.55%, respectively.

[0058] In summary, the cross-domain RFID human behavior recognition method based on meta-learning according to embodiments of the present invention firstly acquires human behavior signals corresponding to each domain collected by the RFID device, wherein multiple human behavior signals collected under the same target and environment are defined as the same domain, and each human behavior signal includes the tag's phase and the received signal strength indication; then, the human behavior signals are preprocessed, and a task sample set is generated based on the preprocessed data corresponding to each domain, and the task sample set is divided into a support set and a query set; next, a relational network is constructed, and the support set and query set are input into the relational network for training to obtain a trained human behavior recognition method. The improved model comprises an embedding module and a relation module. The embedding module acquires the time-frequency features of the query sample and support samples, while the relation module identifies the matching category of the query sample in the support set. Finally, the human behavior signal to be identified and the support set are acquired and input into the human behavior recognition model to obtain the human behavior recognition result corresponding to the human behavior signal to be identified. Thus, through the ability of meta-learning to quickly adapt to new domains, the behavior recognition of each user's human activity set is regarded as a task, enabling the improved relation network to quickly adjust and adapt with only a small amount of new data, thereby improving the recognition effect.

[0059] To implement the above embodiments, this invention also proposes a computer-readable storage medium storing a meta-learning-based cross-domain RFID human behavior recognition program. When executed by a processor, this meta-learning-based cross-domain RFID human behavior recognition program implements the meta-learning-based cross-domain RFID human behavior recognition method as described above.

[0060] According to an embodiment of the present invention, a computer-readable storage medium stores a meta-learning-based cross-domain RFID human behavior recognition program, so that when a processor executes the meta-learning-based cross-domain RFID human behavior recognition program, it implements the meta-learning-based cross-domain RFID human behavior recognition method as described above. Thus, by leveraging the ability of meta-learning to quickly adapt to new domains, the behavior recognition of each user's human activity set is treated as a task, enabling the improved relationship network to quickly adjust and adapt with only a small amount of new data, thereby improving the recognition effect.

[0061] 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 cross-domain RFID human behavior recognition method based on meta-learning as described above.

[0062] According to the computer device of the present invention, a meta-learning-based cross-domain RFID human behavior recognition program is stored in a memory, so that when the processor executes the meta-learning-based cross-domain RFID human behavior recognition program, the meta-learning-based cross-domain RFID human behavior recognition method described above is implemented. Thus, by leveraging the ability of meta-learning to quickly adapt to new domains, the behavior recognition of each user's human activity set is treated as a task, enabling the improved relationship network to be quickly adjusted and adapted with only a small amount of new data, thereby improving the recognition effect.

[0063] To implement the above embodiments, this invention proposes a cross-domain RFID human behavior recognition device based on meta-learning, such as... Figure 6 As shown, the cross-domain RFID human behavior recognition device based on meta-learning includes: an acquisition module 10, a preprocessing module 20, a training module 30, and a human behavior recognition module 40.

[0064] The system comprises the following modules: an acquisition module 10, which acquires human behavior signals corresponding to each domain collected by the RFID device, wherein multiple human behavior signals collected under the same target and environment are defined as the same domain, and each human behavior signal includes the phase of the tag and the received signal strength indication; a preprocessing module 20, which preprocesses the human behavior signals and generates a task sample set based on the data corresponding to each domain after preprocessing, and divides the task sample set into a support set and a query set; a training module 30, which constructs a relational network and inputs the support set and query set into the relational network for training to obtain a trained human behavior recognition model, wherein the relational network includes an embedding module and a relational module, the embedding module is used to acquire the time-frequency features of the query samples and support samples, and the relational module is used to identify the matching category corresponding to the query sample in the support set; and a human behavior recognition module 40, which acquires the human behavior signal to be recognized and the support set, and inputs the human behavior signal to be recognized and the support set into the human behavior recognition model to obtain the human behavior recognition result corresponding to the human behavior signal to be recognized.

[0065] Optionally, the preprocessing module 20 is also used to perform one-dimensional phase expansion on the original phase of all tags; perform linear interpolation on the signal strength indication to obtain a resampled signal stream; perform filtering and smoothing on the resampled signal stream using a filter; and construct a DS stream for the signal stream corresponding to each tag after filtering and smoothing, wherein the DS stream calculates the difference between two consecutive data values ​​in the signal stream.

[0066] Optionally, the preprocessing module 20 is also used to construct a domain list corresponding to each human behavior signal, wherein the domain list includes multiple domains, and each domain includes a corresponding human behavior signal; when generating a task sample corresponding to each human behavior signal, a domain is randomly selected from the corresponding domain list, and a human behavior signal is randomly selected from the selected domain to generate a task sample set.

[0067] Optionally, the embedding module includes a self-calibration module, a backbone network, and a cross-calibration module. The self-calibration module uses a non-local attention mechanism to capture the dependencies between the signal streams corresponding to each label in the samples, so as to preserve activity-related information and mitigate environment-related interference. The backbone network extracts time-frequency features from activity-related information. The cross-calibration module uses a pairwise similarity module to capture query-support association masks and generates a set of calibrated support embeddings for each query sample. The backbone network includes three CNN blocks and a max-pooling layer, where each CNN block includes 32-filter 3×3 convolutions followed by batch normalization and ReLU nonlinearity.

[0068] Optionally, the embedding module is trained according to the following loss function to learn the similarity of embedded features through a nonlinear subspace and to cluster features of the same category:

[0069]

[0070] Among them, T i Let θ represent the i-th task, θ represent the network parameters of the embedded module, and N represent the n-th task. B Represents the number of paired samples, and represents, II m,n This indicates whether samples i and j belong to the same activity category. Specifically, if samples i and j belong to the same activity category, the value is 1; if samples i and j do not belong to the same activity category, the value is 0. σ(η) i.j ) represents the cosine distance between the feature vectors of samples i and j obtained through the embedding network.

[0071] Optionally, the relation network can be trained according to the following loss function:

[0072]

[0073] Among them, T i Let θ represent the i-th task, and let θ represent the network parameters of the embedded module. Represents the network parameters of the relational module, II m,n This indicates whether the support sample n and the query sample m belong to the same sample. Specifically, if the two samples have the same label, the value is 1; otherwise, the value is 0. This represents the probability that the query sample m belongs to the category to which the supporting sample n belongs.

[0074] It should be noted that the above regarding Figure 1 The description of the meta-learning-based cross-domain RFID human behavior recognition method also applies to this meta-learning-based cross-domain RFID human behavior recognition device, and will not be repeated here.

[0075] In summary, the cross-domain RFID human behavior recognition device based on meta-learning according to embodiments of the present invention acquires human behavior signals corresponding to each domain collected by the RFID device through an acquisition module. Multiple human behavior signals collected under the same target and environment are defined as the same domain, and each human behavior signal includes the tag's phase and the received signal strength indication. A preprocessing module preprocesses the human behavior signals and generates a task sample set based on the preprocessed data corresponding to each domain, dividing the task sample set into a support set and a query set. A training module constructs a relational network and inputs the support set and query set into the relational network for training to obtain a trained human behavior recognition system. The model includes an embedding module and a relation module. The embedding module is used to acquire the time-frequency features of query samples and support samples, and the relation module is used to identify the matching category of the query sample in the support set. The human behavior recognition module acquires the human behavior signal to be recognized and the support set, and inputs the human behavior signal to be recognized and the support set into the human behavior recognition model to obtain the human behavior recognition result corresponding to the human behavior signal to be recognized. Thus, through the ability of meta-learning to quickly adapt to new domains, the behavior recognition of each user's human activity set is regarded as a task, so that the improved relation network can be quickly adjusted and adapted with only a small amount of new data, thereby improving the recognition effect.

[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 the preferred embodiments as well as 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., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is 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 cross-domain RFID human behavior recognition method based on meta-learning, characterized in that, Includes the following steps: The human behavior signal corresponding to each domain collected by the radio frequency identification device is acquired. Multiple human behavior signals collected under the same target and environment are defined as the same domain, and each human behavior signal includes the phase of the tag and the received signal strength indication. The human behavior signal is preprocessed, and a task sample set is generated based on the data corresponding to each preprocessed domain. The task sample set is then divided into a support set and a query set. A relation network is constructed, and the support set and the query set are input into the relation network for training to obtain a trained human behavior recognition model. The relation network includes an embedding module and a relation module. The embedding module is used to obtain the time-frequency features of the query sample and the support sample, and the relation module is used to identify the matching category of the query sample in the support set. The human behavior signal to be identified and the support set are obtained, and the human behavior signal to be identified and the support set are input into the human behavior recognition model to obtain the human behavior recognition result corresponding to the human behavior signal to be identified. The embedding module includes a self-calibration module, a backbone network, and a cross-calibration module. The self-calibration module uses a non-local attention mechanism to capture the dependencies between the signal streams corresponding to each label in the sample, so as to retain activity-related information and mitigate environment-related interference. The backbone network extracts time-frequency features from activity-related information. The cross-calibration module uses a pairwise similarity module to capture query-support association masks and generates a set of calibrated support embeddings for each query sample.

2. The cross-domain RFID human behavior recognition method based on meta-learning as described in claim 1, characterized in that, Preprocessing the human behavioral signals includes: Perform a one-dimensional phase unfolding on the original phase of all labels; Linear interpolation is performed on the signal strength indication to obtain a resampled signal stream; A filter is used to smooth the resampled signal stream. A DS stream is constructed for the signal stream corresponding to each tag after filtering and smoothing, where the DS stream calculates the difference between two consecutive data values ​​in the signal stream.

3. The cross-domain RFID human behavior recognition method based on meta-learning as described in claim 1, characterized in that, A task sample set is generated based on the data corresponding to each preprocessed domain, including: Construct a domain list corresponding to each human behavioral signal, wherein the domain list includes multiple domains, and each domain includes corresponding human behavioral signals; When generating task samples corresponding to each human behavior signal, a domain is randomly selected from the corresponding domain list, and a human behavior signal is randomly selected from the selected domain to generate a task sample set.

4. The cross-domain RFID human behavior recognition method based on meta-learning as described in claim 3, characterized in that, The backbone network consists of three CNN blocks and a max pooling layer. Each CNN block includes 32 filters, 3×3 convolutions, followed by batch normalization and ReLU nonlinearity.

5. The cross-domain RFID human behavior recognition method based on meta-learning as described in claim 4, characterized in that, The embedding module is trained using the following loss function to learn the similarity of embedded features through a nonlinear subspace and to cluster features of the same category: Among them, T i Let θ represent the i-th task, θ represent the network parameters of the embedded module, and N represent the n-th task. B Represents the number of paired samples, II m,n This indicates whether samples i and j belong to the same activity category. Specifically, if samples i and j belong to the same activity category, the value is 1; if samples i and j do not belong to the same activity category, the value is 0. σ(η) i.j ) represents the cosine distance between the feature vectors of samples i and j obtained through the embedding network.

6. The cross-domain RFID human behavior recognition method based on meta-learning as described in claim 1, characterized in that, The relational network is trained using the following loss function: Among them, T i Let θ represent the i-th task, and let θ represent the network parameters of the embedded module. Represents the network parameters of the relational module, II m,n This indicates whether the support sample n and the query sample m belong to the same sample. Specifically, if the two samples have the same label, the value is 1; otherwise, the value is 0. This represents the probability that the query sample m belongs to the category to which the supporting sample n belongs.

7. A computer-readable storage medium, characterized in that, It stores a meta-learning-based cross-domain RFID human behavior recognition program, which, when executed by a processor, implements the meta-learning-based cross-domain RFID human behavior recognition method as described in any one of claims 1-6.

8. 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 cross-domain RFID human behavior recognition method based on meta-learning as described in any one of claims 1-6.

9. A cross-domain RFID human behavior recognition device based on meta-learning, characterized in that, include The acquisition module is used to acquire human behavior signals corresponding to each domain collected by the radio frequency identification device. Multiple human behavior signals collected under the same target and environment are defined as the same domain, and each human behavior signal includes the phase of the tag and the received signal strength indication. The preprocessing module is used to preprocess the human behavior signal, generate a task sample set based on the data corresponding to each preprocessed domain, and divide the task sample set into a support set and a query set. A training module is used to construct a relation network and input the support set and the query set into the relation network for training to obtain a trained human behavior recognition model. The relation network includes an embedding module and a relation module. The embedding module is used to obtain the time-frequency features of the query sample and the support sample, and the relation module is used to identify the matching category of the query sample in the support set. The human behavior recognition module is used to acquire the human behavior signal to be recognized and the support set, and input the human behavior signal to be recognized and the support set into the human behavior recognition model to obtain the human behavior recognition result corresponding to the human behavior signal to be recognized. The embedding module includes a self-calibration module, a backbone network, and a cross-calibration module. The self-calibration module uses a non-local attention mechanism to capture the dependencies between the signal streams corresponding to each label in the sample, so as to retain activity-related information and mitigate environment-related interference. The backbone network extracts time-frequency features from activity-related information. The cross-calibration module uses a pairwise similarity module to capture query-support association masks and generates a set of calibrated support embeddings for each query sample.

Images