Data co-processing method and device based on dual-frequency RFID and electronic equipment

Abstract

The invention provides a data co-processing method and device based on double-frequency RFID and electronic equipment, and relates to the technical field of clothing inventory management.The method comprises the steps that a pre-stored association mapping table of the double-frequency RFID is obtained; acquiring radio frequency signal data of the dual-frequency RFID acquired by the reader-writer; generating a differential feature vector according to the continuous time sequence signal; monitoring a statistical fluctuation value of the differential feature vector, inputting the differential feature vector in the corresponding time window into the neural network model when the statistical fluctuation value exceeds a preset wake-up threshold, and outputting a spatial behavior category; and when the discrete trigger signal is detected, searching a first frequency band identification code corresponding to the target object in the association mapping table, matching a spatial behavior category in a preset time neighborhood according to the trigger time of the discrete trigger signal, generating a data processing strategy based on the matched spatial behavior category and the first frequency band identification code, and sending the data processing strategy to the target object. Therefore, cooperative intelligent data processing of double-frequency detection is realized.

Description

Technical Field

[0001] This application relates to the field of apparel inventory management technology, specifically to a data collaborative processing method, apparatus, and electronic device based on dual-frequency RFID. Background Technology

[0002] RFID (Radio Frequency Identification) technology is a technology that uses radio frequency signals to automatically identify target objects and obtain relevant information. The apparel industry commonly uses RFID technology for store management, saving manpower, material resources, and financial resources. Although RFID technology has significantly improved inventory accuracy and accelerated inventory counting, most existing RFID applications rely on a single frequency band. High-frequency bands have long reading distances and are suitable for batch inventory counting, but are easily affected by environmental multipath interference. In complex multipath environments, it is impossible to distinguish whether goods are statically displayed or being interacted with by users, leading to missing heat signature data. Summary of the Invention

[0003] To address the aforementioned issues, this application provides a data collaborative processing method, apparatus, and electronic device based on dual-frequency RFID, which achieves high-precision behavior recognition and intelligent business decision-making through dual-frequency collaboration, differential signal processing, and adaptive control.

[0004] The technical solution of this application embodiment is as follows: In a first aspect, embodiments of this application provide a data collaborative processing method based on dual-frequency RFID, applied to an edge computing device. The edge computing device is communicatively connected to a reader / writer, which is connected to an antenna disposed in a target area for collecting signals from the dual-frequency RFID. The dual-frequency RFID is disposed on the packaging of the target object. The method includes: Obtain the pre-stored association mapping table of the dual-frequency RFID, the association mapping table including the unique mapping relationship between the first frequency band identifier code and the second frequency band identifier code; The reader collects radio frequency signal data from dual-frequency RFID, the radio frequency signal data including continuous timing signals of a first frequency band and discrete trigger signals of a second frequency band, wherein the first frequency band is an ultra-high frequency band and the second frequency band is a high frequency band; Generate a differential feature vector based on the continuous time-series signal; The statistical fluctuation value of the differential feature vector is monitored. When the statistical fluctuation value exceeds the preset wake-up threshold, the differential feature vector within the corresponding time window is input into the preset neural network model, the spatial behavior category of the target object is output, and the spatial behavior category and the corresponding timestamp are cached. When the discrete trigger signal is detected, the first frequency band identifier code corresponding to the target object is searched in the association mapping table. Based on the trigger time of the discrete trigger signal, the spatial behavior category within the preset time neighborhood is matched. A data processing strategy is generated based on the matched spatial behavior category and the first frequency band identifier code, and the data processing strategy is transmitted to the cloud.

[0005] In the above technical solution, the edge computing device is connected to the reader, and the reader is connected to the antenna set in the target area to collect the dual-frequency RFID signal. The dual-frequency RFID is set on the packaging of the target object. By using dual-frequency data, the limitations of a single frequency band are solved. The dual-frequency RFID is set on the packaging of the target object to uniquely identify the target object.

[0006] Based on the above settings, a pre-stored association mapping table for dual-frequency RFID is obtained. This table includes a unique mapping relationship between the first-band identifier and the second-band identifier, establishing a connection between identifications of different frequency bands to ensure that the analyzed object is the same entity. Radio frequency signal data from the dual-frequency RFID reader is acquired. This data includes continuous time-series signals from the first frequency band and discrete trigger signals from the second frequency band. The first frequency band is an ultra-high frequency band, and the second is a high frequency band. By acquiring signals from different frequency bands, the limitations of a single frequency band are avoided. Differential feature vectors are generated based on the continuous time-series signals. These vectors accurately identify operations on the target object and filter out interference. The overall monitoring of the differential feature vectors... The system calculates fluctuation values. When the statistical fluctuation value exceeds a preset wake-up threshold, the differential feature vector within the corresponding time window is input into a preset neural network model. The model outputs the spatial behavior category of the target object and caches the spatial behavior category and its corresponding timestamp. By monitoring the spatial behavior category, high-precision business identification is achieved, which is beneficial for providing a foundation for subsequent management. When a discrete trigger signal is detected, the system searches for the first frequency band identifier code corresponding to the target object in the association mapping table. Based on the trigger time of the discrete trigger signal, the system matches the spatial behavior category within a preset time neighborhood. Based on the matched spatial behavior category and the first frequency band identifier code, a data processing strategy is generated to achieve intelligent decision-making. The data processing strategy is then transmitted to the cloud, and through cloud-edge collaboration, end-to-end intelligent management is achieved.

[0007] In some embodiments of this application, generating a differential feature vector based on the continuous time-series signal includes: Identify the target tag signal and the preset reference tag signal in the continuous time-series signal; The change in radio frequency characteristics of the target tag signal relative to the reference tag signal is calculated to generate a differential feature vector.

[0008] In some embodiments of this application, the continuous time-series signal includes a signal strength sequence and a phase sequence; The step of calculating the change in radio frequency characteristics of the target tag signal relative to the reference tag signal to generate a differential feature vector includes: The phase sequence of the target tag signal is unwrapped to obtain the unwrapped target phase sequence, and the phase sequence of the reference tag signal is unwrapped to obtain the unwrapped reference phase sequence. Subtracting the unwound target phase sequence from the unwound reference phase sequence at the same time yields the relative phase sequence. The relative phase sequence is smoothed and filtered to obtain the differential feature vector.

[0009] In some embodiments of this application, the target tag signal comprises a phase sequence based on a period of 0 to 2π; The step of unwinding the phase sequence of the target tag signal to obtain the unwound target phase sequence includes: Calculate the phase difference between the current sampling point and the previous sampling point in the phase sequence in chronological order; Monitor whether the phase difference value has a periodic jump: If the phase difference is greater than π, a compensation value of -2π is added to the phase values ​​of the current sampling point and subsequent sampling points; If the phase difference is less than -π, a compensation value of +2π is added to the phase values ​​of the current sampling point and subsequent sampling points; Based on the phase values ​​after superimposed compensation, the phase sequence is converted into a spatially continuous linear phase sequence, which serves as the target phase sequence after unwinding.

[0010] In some embodiments of this application, the data processing strategy generated based on the matched spatial behavior category and the first frequency band identifier includes: Determine whether the matched spatial behavior category contains a preset strongly associated action; If the discrete trigger signal is detected and no spatial behavior category indicating a strongly correlated action is matched within the preset time neighborhood, a model calibration instruction is generated. The model calibration command is used to extract the continuous time-series signal within the preset time neighborhood as positive samples, and to perform online incremental training or parameter fine-tuning on the preset neural network model.

[0011] In some embodiments of this application, the generated model calibration instructions specifically include: Using the timestamp of the discrete trigger signal as the endpoint, backtrack the preset behavior-related delay to determine the potential action time window; Traverse the continuous temporal signals within the potential action time window and calculate the local fluctuation extrema of different sub-time slices; The sub-time slices with the largest local fluctuation extremes that have not triggered the preset wake-up threshold are selected, and their corresponding differential feature vectors are marked as difficult positive samples. The difficult positive samples are input into the preset neural network model to calculate the loss function, and the weight parameters of the neural network model are adjusted based on the calculation results.

[0012] In some embodiments of this application, after monitoring the statistical fluctuation value of the differential eigenvector, the method further includes: The read collision rate corresponding to the continuous time sequence signal of the first frequency band is obtained, wherein the read collision rate is calculated by the reader based on the number of collisions of the collected dual-frequency RFID signals and the total number of reads; When the statistical fluctuation value is close to zero and the collision rate is less than a preset collision threshold, the radio frequency transmission power of the reader to the first frequency band is adjusted.

[0013] Secondly, embodiments of this application provide a data collaborative processing device based on dual-frequency RFID, applied to edge computing devices, the device comprising: The table data acquisition module is used to acquire the pre-stored association mapping table of the dual-frequency RFID, the association mapping table including the unique mapping relationship between the first frequency band identifier code and the second frequency band identifier code; The signal acquisition module is used to acquire radio frequency signal data of the dual-frequency RFID collected by the reader. The radio frequency signal data includes a continuous timing signal of the first frequency band and a discrete trigger signal of the second frequency band, wherein the first frequency band is an ultra-high frequency band and the second frequency band is a high frequency band. The signal processing module is used to generate a differential feature vector based on the continuous time-series signal; The category recognition module is used to monitor the statistical fluctuation value of the differential feature vector. When the statistical fluctuation value exceeds the preset wake-up threshold, the differential feature vector within the corresponding time window is input into the preset neural network model, the spatial behavior category of the target object is output, and the spatial behavior category and the corresponding timestamp are cached. The business processing module is used to, when the discrete trigger signal is detected, look up the first frequency band identifier code corresponding to the target object in the association mapping table, match the spatial behavior category in the preset time neighborhood according to the trigger time of the discrete trigger signal, generate a data processing strategy based on the matched spatial behavior category and the first frequency band identifier code, and transmit the data processing strategy to the cloud.

[0014] Thirdly, embodiments of this application provide an electronic device including a processor, a memory, a user interface, a communication bus, and a network interface. The processor, the memory, the user interface, and the network interface are respectively connected to the communication bus. The memory is used to store instructions. The user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method described in any one of the first aspects.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing instructions that, when executed, perform the method described in any one of the methods provided in the first aspect above.

[0016] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. By first setting up dual-frequency RFID and establishing an association mapping table for dual-frequency RFID, the connection between identifications of different frequency bands is realized, ensuring that the analyzed object is the same subject. Continuous time-series signals from the first frequency band and discrete trigger signals from the second frequency band are acquired. By performing differential calculations on the continuous time-series signals and monitoring the fluctuations of the differential feature vector, interference signals can be filtered out, and wake-up spatial behavior recognition can be triggered. Combined with the detected discrete trigger signals, a data processing strategy is generated based on the matched spatial behavior category and the first frequency band identification code, improving the accuracy of identification. The data processing strategy is transmitted to the cloud, and through cloud collaboration, intelligent management of the entire chain is achieved. Therefore, this effectively solves the problems in related technologies where a single frequency band is easily affected by environmental multipath effects and it is difficult to determine the exact location of a single item or user interaction behavior, making seamless tracking impossible.

[0017] 2. By introducing reference tags and differential phase calculation, environmental noise can be removed, enabling the system to accurately identify physical behaviors of users such as "taking" and "trying on" in noisy environments, rather than simply "present / absent".

[0018] 3. By utilizing both the collision rate and signal fluctuation indicators, the device can achieve adaptive sleep and wake-up, thus solving the energy consumption and interference problems of RFID devices during large-scale deployment.

[0019] 4. Adjusting the wake-up threshold for specific environments is a targeted optimization based on feature learning. It also establishes connections between high-frequency signals, improving recognition accuracy and enabling intelligent decision-making. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a data collaborative processing method based on dual-frequency RFID provided in one embodiment of this application; Figure 2 This is a schematic diagram of the overall structure of a data collaborative processing method based on dual-frequency RFID provided in one embodiment of this application; Figure 3 This is a flowchart illustrating a data collaborative processing method based on dual-frequency RFID provided in another embodiment of this application; Figure 4 This is a flowchart illustrating a data collaborative processing method based on dual-frequency RFID provided in another embodiment of this application; Figure 5 This is a schematic diagram of the structure of a data collaborative processing device based on dual-frequency RFID provided in one embodiment of this application; Figure 6 This is a schematic diagram of the structure of an electronic device provided in one embodiment of this application. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0022] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0023] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. Furthermore, 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 indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0024] In related technologies, most existing RFID applications rely on a single frequency band. UHF RFID has a long reading distance, making it suitable for batch inventory, but it is easily affected by environmental multipath interference and has difficulty determining the exact location of individual items or user interaction behaviors (such as trying on or taking items). HF / NFC RFID has accurate near-field interaction, but requires the user to actively approach, making seamless tracking impossible. In addition, existing systems typically operate at full power, resulting in energy waste and spectrum interference, and lack the ability to mine deeper business data such as "trying on but not purchasing."

[0025] Based on this, this application provides a data collaborative processing method, apparatus, electronic device, and readable storage medium based on dual-frequency RFID. The data collaborative processing method first sets up dual-frequency RFID and establishes an association mapping table for dual-frequency RFID to realize the connection between identifications of different frequency bands, ensuring that the analysis object is the same subject. It acquires radio frequency signal data from the dual-frequency RFID collected by the reader / writer. The radio frequency signal data includes continuous time-series signals of the first frequency band and discrete trigger signals of the second frequency band. The first frequency band is an ultra-high frequency band, and the second frequency band is a high frequency band. By acquiring signals from different frequency bands, the limitations of a single frequency band are avoided. Based on the continuous time-series signals, a differential feature vector is generated. This differential feature vector accurately identifies operations on the target object and filters out interference. The statistical fluctuation value of the differential feature vector is monitored. When the statistical fluctuation value exceeds a preset wake-up threshold, the differential feature vector within the corresponding time window is input into a preset neural network. The network model outputs the spatial behavior category of the target object and caches the spatial behavior category and its corresponding timestamp. By monitoring the spatial behavior category, high-precision business identification is achieved, which is beneficial for providing a foundation for subsequent management. When a discrete trigger signal is detected, the first frequency band identifier code corresponding to the target object is searched in the association mapping table. Based on the trigger time of the discrete trigger signal, the spatial behavior category in the preset time neighborhood is matched. Based on the matched spatial behavior category and the first frequency band identifier code, a data processing strategy is generated to achieve intelligent decision-making. The data processing strategy is then transmitted to the cloud, and through cloud-edge collaboration, end-to-end intelligent management is achieved.

[0026] It should be noted that this data collaborative processing method based on dual-frequency RFID, used for apparel store management, can also be applied to the management of the retail industry and the jewelry industry. Through dual-frequency RFID technology, not only can end-to-end management be achieved, but it can also be expanded to include integrated online and offline management, as well as accurate early warning of theft.

[0027] The technical solutions provided in the embodiments of this application will be further described below with reference to the accompanying drawings.

[0028] Reference Figure 1 , Figure 1This is a flowchart illustrating a data collaborative processing method based on dual-frequency RFID provided in this application embodiment. The data collaborative processing method based on dual-frequency RFID is applied to an edge computing device. The data collaborative processing device based on dual-frequency RFID executes the data collaborative processing method based on dual-frequency RFID through a processor in an electronic device or a readable storage medium. The data collaborative processing method based on dual-frequency RFID includes steps S100, S200, S300, S400, and S500.

[0029] In one embodiment, the edge computing device is communicatively connected to a reader, which is connected to an antenna located in the target area for collecting dual-frequency RFID signals. The dual-frequency RFID is installed on the packaging of the target object. For example, the reader can be installed on the ceiling of a clothing store to collect signals from various locations, achieving full signal coverage. The target area includes a shelf area and a fitting room area. The shelf area is equipped with an ultra-high frequency antenna for collecting dual-frequency RFID signals, with antennas installed on each shelf layer to accurately collect ultra-high frequency signals. The fitting room area is equipped with a high-frequency antenna or sensing device for collecting dual-frequency RFID signals. The connection structure is as follows... Figure 2 As shown, based on the above structural setup, the following steps will be explained.

[0030] Step S100: Obtain the pre-stored association mapping table of dual-frequency RFID, which includes the unique mapping relationship between the first frequency band identifier code and the second frequency band identifier code.

[0031] In one embodiment, the association mapping table includes a unique mapping relationship between a first frequency band identifier and a second frequency band identifier. The first frequency band identifier is the electronic product code of an ultra-high frequency (UHF, 860-960MHz) tag, possessing long-range (3-10 meters) group reading capability; the second frequency band identifier is the unique identifier of a high frequency (HF / NFC, 13.56MHz) tag, possessing near-field (<10cm) triggering capability. The above codes and identifiers can refer to a single identifier style; in this embodiment, they are set to correspond to different frequency band identifiers for dual-frequency RFID. The unique mapping relationship is that the same target object has both UHF and HF tags (or a two-in-one composite tag) affixed simultaneously, and their numbers are physically bound. The target object is a piece of clothing, which can be a shirt, pants, or other different items. Dual-frequency RFID is used on a piece of clothing, with the dual-frequency RFID corresponding to the first and second frequency band identifiers. During the goods warehousing process, the tag is read by a dual-frequency card issuer to establish the mapping relationship between the first and second frequency band identifiers, and the data is transmitted to the edge device for storage. A pre-stored association mapping table of frequency RFID is obtained using a preset data reading function, breaking down the data barrier between long-range tracking (UHF) and near-field user interaction (HF), ensuring that the system tracks the same physical object and providing a foundation for subsequent calculations. The preset data reading function can be either the `read()` function or the `open()` function.

[0032] Step S200: Acquire radio frequency signal data of dual-frequency RFID collected by the reader. The radio frequency signal data includes continuous timing signals of the first frequency band and discrete trigger signals of the second frequency band. The first frequency band is an ultra-high frequency band and the second frequency band is a high frequency band.

[0033] In one embodiment, the first frequency band is ultra-high frequency (UHF) for sensing spatial changes, and the second frequency band is high frequency (HF) for accurately identifying nodes. The continuous time-series signal is the time-series data returned by the reader when polling the UHF tag at high frequency, including signal strength, phase, and timestamp. High-frequency polling uses sampling at a frequency such as 20Hz. The discrete trigger signal is a single interruption event signal generated by the reader only when the high-frequency tag enters the fitting room, triggered by a sensor on the fitting room mirror or when the user brings their mobile phone close to the tag.

[0034] The edge device is equipped with an LLRP protocol or SDK interface. Once the reader is activated, it continuously communicates with the antenna to collect dual-frequency RFID signals. The reader processes the signals and outputs signal data containing signal strength, phase, and timestamps. Information is exchanged with the reader via the LLRP protocol or SDK interface to obtain the dual-frequency RFID radio frequency signal data collected and output by the reader. By simultaneously acquiring the wide coverage capability of ultra-high frequency and the accurate identification capability of high frequency, preparation is made for subsequent processing.

[0035] Step S300: Generate differential feature vectors based on continuous time-series signals.

[0036] In one embodiment, a differential feature vector is generated based on a continuous time-series signal, including but not limited to the following steps: Step S310: Identify the target tag signal and the preset reference tag signal in the continuous time sequence signal.

[0037] In one embodiment, the original signal is greatly affected by the environment. For example, in store management, fluctuations caused by human movement may be misidentified as fluctuations in clothing handling, leading to significant errors in recognition accuracy. This error caused by interference is resolved by identifying the target tag signal and a preset reference tag signal within a continuous time-series signal. The preset reference tag can be the tag of a garment adjacent to the target tag. For example, if movement causes a garment to fluctuate, adjacent garments will also fluctuate. The reliability of the fluctuation is determined based on the adjacent garments. In this case, the reader reads the tag of the fluctuating item, including the reference tag. Alternatively, a reference tag with a known ID can be fixedly affixed to a shelf pillar or antenna. All clothing on the shelf uses this tag as a reference point. In this case, the reader will simultaneously read the moving item (target tag) and the stationary reference tag during scanning.

[0038] When using a reference tag at a fixed location, the specific identification process is as follows: First, the electronic product code (EPC) in the received radio frequency data packet is read; the EPC is then compared with the reference tag identifier list for fixed locations (recorded and stored when setting up the reference tag at the fixed location); if the EPC exists in the reference tag identifier list, the signal strength and phase data carried by the data packet are determined to belong to the reference tag signal; thus, the reference tag signal and the target tag signal are identified. If the EPC does not exist in the reference tag identifier list, the data packet is determined to belong to the target tag signal, which reflects the spatial state of the product to be identified.

[0039] When dealing with adjacent target objects, the specific identification process is as follows: The location information of the signal collected by the transmission antenna is obtained. One of the adjacent shaking target items is used as the reference tag signal, and the other as the target tag signal. That is, each target item can serve as both a reference signal and a target tag signal. This method saves the cost associated with fixed-location setups. The above method can distinguish between shaking caused by picking up the object and fluctuations caused by human movement. For example, in the case of adjacent target objects, human movement can cause multiple target objects to shake. By monitoring the positional changes of adjacent items, if the shaking is caused by human movement, the relative positions of the target tag and the reference tag will not change, thus effectively identifying the picking operation.

[0040] Step S320: Calculate the change in radio frequency characteristics of the target tag signal relative to the reference tag signal to generate a differential feature vector.

[0041] like Figure 3 As shown, the continuous time-series signal includes a signal strength sequence and a phase sequence. Specifically, the change in radio frequency characteristics of the target tag signal relative to the reference tag signal is calculated to generate a differential feature vector, including but not limited to the following steps: Step S321: Dewrap the phase sequence of the target tag signal to obtain the dewrapped target phase sequence, and dewrap the phase sequence of the reference tag signal to obtain the dewrapped reference phase sequence.

[0042] In some possible embodiments of this application, the target tag signal includes a phase sequence based on a period from 0 to 2π. The phase sequence of the target tag signal is unwrapped to obtain the unwrapped target phase sequence, including but not limited to: calculating the phase difference between the current sampling point and the previous sampling point in the phase sequence in chronological order; monitoring whether there is a periodic jump in the phase difference: if the phase difference is greater than π, a compensation value of -2π is added to the phase value of the current sampling point and subsequent sampling points; if the phase difference is less than -π, a compensation value of +2π is added to the phase value of the current sampling point and subsequent sampling points; based on the phase value after the superimposed compensation value, the phase sequence is converted into a spatially continuous linear phase sequence, which is used as the unwrapped target phase sequence.

[0043] Specifically, the RFID phase value cycles between 0 and 2π, meaning the target tag signal is a phase sequence based on a 0-2π period, and the phase change is continuous across physical distances. Based on the time sequence of the acquired signals, the phase difference between the current sampling point and the previous sampling point is calculated, i.e., the phase difference between adjacent sampling points. The time sequence of the acquired signals is formed by timestamps corresponding to the signal acquisition time, and the signals acquired over a period of time are arranged in chronological order according to their timestamps. The process of monitoring whether there are periodic jumps in the phase difference is as follows: if the phase difference is greater than π, a positive truncation is determined, and 2π is subtracted from each phase sequence for compensation, resulting in a superimposed phase value. If the phase difference is less than -π, a negative truncation is determined, and 2π is added for compensation, resulting in a superimposed phase value. The superimposed phase value is a sawtooth wave. Converting the phase sequence into a spatially continuous linear phase sequence, i.e., converting the sawtooth wave into a smooth linear waveform, serves as the unwound target phase sequence, thereby reducing calculation errors caused by periodic jumps. Similarly, the reference tag signal also suffers from the same problem. Unwinding the phase sequence of the reference tag signal yields an unwound reference phase sequence, reducing calculation errors caused by periodic jumps. Through unwinding, the true distance / position change trend is restored, eliminating calculation errors caused by periodic jumps.

[0044] Step S322: Subtract the corresponding unwound reference phase sequence from the unwound target phase sequence at the same time to obtain the relative phase sequence.

[0045] In some possible embodiments of this application, the unwrapped target phase sequence is subtracted from the unwrapped reference phase sequence at the same time, expressed by the formula: Φrelative(t) = Φtarget(t) − Φreference(t), where Φrelative(t) is the relative phase sequence, Φtarget(t) is the unwrapped target phase sequence, and Φreference(t) is the unwrapped reference phase sequence at the same time. Subtraction of signals at the same time ensures that they are signals triggered simultaneously. For example, assuming the environment causes all signal phases to drift by one phase, the drifted phases are canceled out after subtraction, and the resulting relative phase only reflects the actual movement of the goods relative to the shelf. This accurately reflects whether the goods have been picked up or whether the jitter error is caused by human activity, effectively improving the accuracy of identification.

[0046] Step S323: Perform smoothing filtering on the relative phase sequence to obtain the differential feature vector.

[0047] In some possible embodiments of this application, moving average filtering or Kalman filtering can be used to smooth the relative phase sequence, removing high-frequency random noise and obtaining a differential feature vector. Generating a smooth differential feature vector prepares the system for subsequent spatial behavior recognition, improving the accuracy of subsequent neural network recognition.

[0048] In one embodiment, after monitoring the statistical fluctuation value of the differential feature vector, the data collaborative processing method based on dual-frequency RFID further includes, but is not limited to: obtaining the read collision rate corresponding to the continuous time sequence signal of the first frequency band, wherein the read collision rate is calculated by the reader based on the number of collisions and the total number of reads of the collected dual-frequency RFID signals; and adjusting the radio frequency transmission power of the reader to the first frequency band when the statistical fluctuation value is close to zero and the collision rate is less than a preset collision threshold.

[0049] Specifically, the RFID reader follows the EPC C1G2 protocol for anti-collision processing. During each inventory cycle, the reader issues a query command and allocates several communication time slots. The reader counts the total number of time slots and the number of time slots where signal collisions occur within that cycle. The collision rate is the ratio of the number of conflicting time slots to the total number of time slots. The collision threshold indicates whether the current area has been fully identified; a collision rate less than the preset threshold indicates that no new tags have been identified in the current area. When the statistical fluctuation value is close to zero and the collision rate is less than the preset threshold, it indicates that no one is affecting the movement of goods in the current space, and the number of tags in the current space is stable. To reduce power consumption, the reader's RF duty cycle is adjusted, or the reader's RF transmission power for the first frequency band is adjusted to put the antenna corresponding to the first frequency band into standby mode, thereby reducing the frequency of the antenna corresponding to the first frequency band and achieving energy saving.

[0050] Step S400: Monitor the statistical fluctuation value of the differential feature vector. When the statistical fluctuation value exceeds the preset wake-up threshold, input the differential feature vector within the corresponding time window into the preset neural network model, output the spatial behavior category of the target object, and cache the spatial behavior category and the corresponding timestamp.

[0051] In one embodiment, the standard deviation of the differential feature vector within a preset time window is calculated as the fluctuation value. The standard deviation reflects the degree of data deviation; a larger deviation indicates that the product has been moved. Variance can also be used for calculation. By monitoring the statistical fluctuation value of the differential feature vector, spatial behavior category recognition can be triggered subsequently. The wake-up threshold measures the amount of physical displacement of the product, representing the boundary of the degree of data deviation, rather than a static display, thereby determining whether the target object has been taken within the time window. The time window can be 2 seconds, 5 seconds, or other values; the wake-up threshold can be 0.5 radians, approximately 1.3 centimeters, an empirical value derived from the physical wavelength of RFID and the background noise of the environment. This value can filter out signal jitter caused by electronic thermal noise or weak environmental vibration when the tag is stationary (usually less than 0.2 radians), while ensuring that it can sensitively capture significant displacement (usually greater than 5 centimeters, or greater than 1.9 radians) when a customer takes the product, thus achieving a balance between system sensitivity and anti-false triggering rate.

[0052] In one embodiment, the calculated standard deviation is compared with a wake-up threshold. If the statistical fluctuation value exceeds the preset wake-up threshold, it indicates that the target object has moved physically. The differential feature vector within the corresponding time window is then input into a preset neural network model, which outputs the spatial behavior category of the target object. The preset neural network model can be a lightweight convolutional neural network (CNN) model or a recurrent neural network (RNN) model, both of which are pre-trained models. For example, a CNN model includes an input layer, convolutional layers, nonlinear mapping layers, and fully connected layers. The training process is existing technology and will not be elaborated here. The specific processing of the neural network model is as follows: First, the differential feature vector within the time window is transformed into a two-dimensional vector. The two-dimensional feature vector represents two dimensions: time and corresponding features. The CNN model is then used to perform feature extraction, activation function processing, and feature linear mapping on the two-dimensional feature vector to output the spatial behavior category. Spatial behavior categories include stationary, picking / browsing, walking / taking away, etc. Based on the time in the two-dimensional feature vector, the spatial behavior category and its corresponding timestamp are cached for subsequent matching of spatial behavior categories within the time neighborhood based on the timestamp, ensuring the consistency of actions over time. Through a two-tiered mechanism of initial screening based on fluctuation values ​​and recognition by neural networks, both the computing power of edge devices is saved, and accurate recognition of complex actions is achieved.

[0053] In another embodiment, if the statistical fluctuation value does not exceed the preset wake-up threshold, it indicates that the target object has not moved physically and is considered an error fluctuation. No further identification calculation is required, thus saving computing resources.

[0054] Step S500: When a discrete trigger signal is detected, the first frequency band identifier code corresponding to the target object is searched in the association mapping table. Based on the trigger time of the discrete trigger signal, the spatial behavior category in the preset time neighborhood is matched. A data processing strategy is generated based on the matched spatial behavior category and the first frequency band identifier code, and the data processing strategy is transmitted to the cloud.

[0055] In one embodiment, the discrete trigger signal is an HF / NFC read event. Data collected by the sensor is transmitted to the reader. This data signal is labeled with a signal identifier to indicate the data collected by the sensor and reflects information such as which fitting room it belongs to. When the edge computing device interacts with the reader, detecting a discrete trigger signal indicates that the target object is in the fitting room and has left the shelf. This is combined with spatial behavior categories to determine the act of taking and trying on clothes. The trigger searches for the first frequency band identifier code corresponding to the target object in the association mapping table. This discrete trigger signal is also identified by a unique second frequency band identifier code. By matching the second frequency band identifier code corresponding to the discrete signal with the corresponding first frequency band identifier code in the association mapping table, the same target object can be identified, avoiding target object identification errors and enabling target object tracking. Based on the trigger time of the discrete trigger signal, the spatial behavior category within a preset time neighborhood is queried. The preset time neighborhood can be 30 seconds, 1 minute, etc. The timestamps corresponding to the trigger time and the spatial behavior category within the time neighborhood are subtracted after time conversion. If the result is within the time neighborhood, the corresponding spatial behavior category is matched. This ensures the continuity of actions, enables accurate tracking of the same target object, and guarantees the accuracy of subsequent data processing strategies.

[0056] In one embodiment, a data processing strategy is generated based on the matched spatial behavior category and the first frequency band identifier code, including but not limited to the following steps: Step S510: Determine whether the matched spatial behavior category contains a preset strongly associated action.

[0057] In some possible embodiments of this application, the preset strongly correlated action is the "taking and trying on" action. The system determines whether the matched spatial behavior category includes the "taking and trying on" action. This action is associated with the UHF signal detected in the first frequency band, verifying the accuracy of the UHF signal detection and facilitating subsequent calibration instructions based on the generated model. For example, if the spatial behavior category includes the "taking and trying on" action and the high-frequency signal is triggered, it is considered a normal state; otherwise, it is considered an abnormal state. By utilizing the short-range, high-determinism characteristics of high-frequency signals as a "supervisory signal" for UHF signal identification results, automated anomaly detection without manual annotation is achieved.

[0058] Step S520: If a discrete trigger signal is detected and no spatial behavior category indicating a strongly correlated action is matched within a preset time neighborhood, a model calibration instruction is generated.

[0059] In some possible embodiments of this application, if a discrete trigger signal is detected, it indicates that a high-frequency signal in the second frequency band has been detected. If no spatial behavior category indicating a strongly correlated action is matched within a preset time neighborhood, it is considered an abnormal state. This means that the high-frequency signal was triggered, but the ultra-high frequency historical records are all "stationary" or "micro-movements," without any "strongly correlated action." For example, a user took the item, but the ultra-high frequency signal was too weak or obstructed to detect it, or the fluctuation did not reach the wake-up threshold. This leads to detection errors, and a model calibration instruction is generated to correct the abnormal situation.

[0060] Step S530: The model calibration command is used to extract continuous time-series signals within a preset time neighborhood as positive samples, and to perform online incremental training or parameter fine-tuning on the preset neural network model.

[0061] In some possible embodiments of this application, the model calibration instruction is used to extract continuous time-series signals within a preset time neighborhood as positive samples. Since the preset time neighborhood reflects the continuity of ultra-high frequency and high frequency detection signals, extracting continuous time-series signals within this time neighborhood can enable the detection of the same target object. Using these signals as positive samples, the neural network model can be trained online incrementally or its parameters can be fine-tuned to improve the sensitivity of weak signal features. This can improve the classification accuracy of the neural network model, thereby effectively identifying fluctuations and improving accuracy.

[0062] like Figure 4 As shown, the process of generating model calibration instructions includes, but is not limited to, the following steps: Step S521: Using the timestamp of the discrete trigger signal as the endpoint, backtrack the preset behavior-related delay to determine the potential action time window.

[0063] In some possible embodiments of this application, the behavior-related delay is the average time taken from when a user picks up the target product to when they try it on, such as 2 minutes. The related delay is calculated by pushing forward from the timestamp of the discrete trigger signal, thus locking in the potential action time window. The potential action time window indicates the detection signals (continuous timing signals) stored in a circular buffer within that time window, preparing for subsequent adjustments.

[0064] Step S522: Traverse the continuous time-series signals within the potential action time window and calculate the local fluctuation extrema of different sub-time slices.

[0065] In some possible embodiments of this application, a sub-window of preset width slides within a potential action time window, with the sub-window being 10 seconds and the step size set to 5 seconds, generating N sub-time slices. The phase sequence within each sub-time slice is differentially processed to calculate local fluctuation extrema, which can be a maximum value, minimum value, phase variance, etc.

[0066] Step S523: Select the sub-time slices with the largest local fluctuation extreme values ​​that have not triggered the preset wake-up threshold, and mark their corresponding differential feature vectors as difficult positive samples.

[0067] In some possible embodiments of this application, according to step S522, the local fluctuation extreme value is obtained. The local fluctuation extreme value is the maximum value. The sub-time slice with the largest local fluctuation extreme value that has not triggered the preset wake-up threshold is selected. The fluctuation value of this sub-time slice is less than the wake-up threshold, indicating that it has been missed. The differential feature vector corresponding to the sub-time slice is marked as a difficult positive sample and labeled (picking action) so that the neural network model can be adjusted later based on the difficult positive sample. Alternatively, if the fluctuation value of all sub-time slices is less than the wake-up threshold, the differential feature vector corresponding to the largest fluctuation value among the sub-time slices is selected and marked as a difficult positive sample and labeled (picking action) so that the neural network model can be adjusted later based on the difficult positive sample. This solves the problem of "threshold setting dilemma" in traditional methods (high threshold leads to missed detection, low threshold leads to false alarms). Through this backtracking mechanism, the system accurately finds the real action samples "in a critical state".

[0068] Step S524: Input the difficult positive samples into the preset neural network model to calculate the loss function, and adjust the weight parameters of the neural network model based on the calculation results.

[0069] In some possible embodiments of this application, labeled difficult positive samples are input into a preset neural network model. Since these samples were previously missed, the model's prediction of their probability as "actions" may be very low. For example, the predicted value may be as low as 0.3, and the label may be 1. The error between the predicted value and the label can be calculated using the cross-entropy loss function. In this case, the loss function value will be relatively large, so a small step size is used for single or small-batch gradient descent. Only the weights of the fully connected layers of the model are fine-tuned, or the feature weights of specific dimensions are updated, to avoid destroying the model's existing ability to recognize normal signals, thereby adjusting the weight parameters of the neural network model. After the update, the model remembers this weak waveform feature. The next time it encounters a similar "weak fluctuation," the confidence level of the model output may increase from 0.3 to 0.6, thereby indirectly lowering the effective wake-up threshold for this type of specific action, effectively identifying weak fluctuations, and realizing a closed loop in model adjustment.

[0070] It should be noted that fine-tuning based on the aforementioned data processing instructions enables the tracking of product data. Through the collaborative processing described above, the "take-and-try-on" process is statistically analyzed to indicate the popularity of target products. Spatial behavior categories also include taking and trying on products before checkout and leaving, or taking products and then checking out and leaving. By generating data processing instructions, the accuracy of signal collection is ensured. Based on the aforementioned dual-frequency collaborative detection, the tracking of the same target product is achieved, facilitating store management. Data processing instructions can also be transmitted to the cloud for aggregation and analysis, facilitating subsequent model adjustments.

[0071] In one embodiment, a three-tiered mapping relationship between stores, inventory, and transactions can be implemented to achieve coordinated allocation of stores. Specifically, this is achieved through the following process: generating business allocation instructions for the target object based on the first frequency band identifier code and spatial behavior category to allocate the target goods.

[0072] In one embodiment, each target object also includes a product code for the transaction, which reflects the category and brand of the target object. The product number and the first frequency band identifier correspond one-to-one, indicating the category and brand type of a target object. It also includes location information on the shelf, which corresponds to the first frequency band identifier. The location of the target object can be determined through the first frequency band identifier, and the target object marked with the corresponding first frequency band identifier can be found through the target location. Obtaining the product code and target location of the target object provides a foundation for subsequent tracking of the product object.

[0073] In one embodiment, the preset transaction database is a database that records transactions based on scanned product codes to determine whether a target object has completed a transaction. When the spatial behavior category indicates that the target product has been taken away, the transaction database is queried for the product code corresponding to the first frequency band identifier to determine the transaction information of the target object corresponding to that first frequency band identifier. Specifically, the behavioral trajectories of all product objects under the product number category are first statistically analyzed.

[0074] In some possible embodiments of this application, since the product code represents the category and brand type of the target object, the product code is used to uniquely identify the transaction and inventory quantity of the same type of target object. The inventory status is queried using the product code, which can be either sufficient stock or insufficient stock. Based on the category of the product code, the behavioral trajectory of all product objects under that category is statistically analyzed. This behavioral trajectory is a log recording the events of products in that category over the past week or month, including the number of times items were picked up, the number of times they entered the fitting room, and the number of times they checked out. Based on the spatial behavioral category indication and corresponding to the product codes of the traded items, the inventory status is effectively extracted, and the product inventory of the target object is adjusted according to the number of times items were picked up and the number of times they entered the fitting room. Through the above process, the three-level mapping relationship of stores, inventory, and transactions is integrated, effectively managing the front-end and back-end of the apparel. Based on the above-generated business allocation instruction, the business status instruction is uploaded to the cloud. The cloud, based on the business status instruction, comprehensively summarizes the inventory information of other stores, allocates products, and sends the allocation information to the edge computing device.

[0075] like Figure 5 As shown in the figure, this application embodiment provides a data collaborative processing device based on dual-frequency RFID, applied to an edge computing device. The data collaborative processing device 100 based on dual-frequency RFID acquires a pre-stored association mapping table of dual-frequency RFID through a table data acquisition module 110. The association mapping table includes a unique mapping relationship between the first frequency band identifier code and the second frequency band identifier code. It acquires radio frequency signal data of dual-frequency RFID collected by the reader through a signal acquisition module 120. The radio frequency signal data includes a continuous time sequence signal of the first frequency band and a discrete trigger signal of the second frequency band, wherein the first frequency band is an ultra-high frequency band and the second frequency band is a high frequency band. The signal processing module 130 generates data based on the continuous time sequence signal. A differential feature vector is generated; then, the category recognition module 140 monitors the statistical fluctuation value of the differential feature vector. When the statistical fluctuation value exceeds the preset wake-up threshold, the differential feature vector within the corresponding time window is input into the preset neural network model, outputting the spatial behavior category of the target object, and caching the spatial behavior category and the corresponding timestamp; when the business processing module 150 detects a discrete trigger signal, it searches for the first frequency band identifier code corresponding to the target object in the association mapping table, matches the spatial behavior category within the preset time neighborhood according to the trigger time of the discrete trigger signal, generates a data processing strategy based on the matched spatial behavior category and the first frequency band identifier code, and transmits the data processing strategy to the cloud.

[0076] It should be noted that the table data acquisition module 110 is connected to the signal acquisition module 120, the signal acquisition module 120 is connected to the signal processing module 130, the signal processing module 130 is connected to the category recognition module 140, and the category recognition module 140 is connected to the business processing module 150. The above-mentioned data collaborative processing method based on dual-frequency RFID is applied to the data collaborative processing device 100 based on dual-frequency RFID. The data collaborative processing device 100, by setting up dual-frequency RFID and establishing an association mapping table for dual-frequency RFID, realizes the connection between identifications of different frequency bands, ensuring that the analysis object is the same subject. It acquires the radio frequency signal data of dual-frequency RFID collected by the reader / writer. The radio frequency signal data includes the continuous time-series signal of the first frequency band and the discrete trigger signal of the second frequency band. The first frequency band is an ultra-high frequency band, and the second frequency band is a high frequency band. By acquiring signals from different frequency bands, the limitations of a single frequency band signal are avoided. Based on the continuous time-series signal, a differential feature vector is generated. By generating the differential feature vector, the operation on the target object can be accurately identified. Filter out interference; monitor the statistical fluctuation value of the differential feature vector. When the statistical fluctuation value exceeds the preset wake-up threshold, input the differential feature vector within the corresponding time window into the preset neural network model, output the spatial behavior category of the target object, and cache the spatial behavior category and the corresponding timestamp. By monitoring the spatial behavior category, high-precision business identification is achieved, which is beneficial for providing a foundation for subsequent management. When a discrete trigger signal is detected, look up the first frequency band identifier code corresponding to the target object in the association mapping table. According to the trigger time of the discrete trigger signal, match the spatial behavior category within the preset time neighborhood. Generate a data processing strategy based on the matched spatial behavior category and the first frequency band identifier code to achieve intelligent decision-making. Transmit the data processing strategy to the cloud and achieve end-to-end intelligent management through cloud-edge collaboration.

[0077] It should also be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0078] This application also discloses an electronic device. (See reference...) Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. The electronic device 500 may include: at least one processor 501, at least one network interface 504, a user interface 503, a memory 505, and at least one communication bus 502.

[0079] The communication bus 502 is used to enable communication between these components.

[0080] The user interface 503 may include a display screen and a camera. Optionally, the user interface 503 may also include a standard wired interface and a wireless interface.

[0081] The network interface 504 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0082] The processor 501 may include one or more processing cores. The processor 501 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 505, and by calling data stored in memory 505. Optionally, the processor 501 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array. The processor 501 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and Modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor 501.

[0083] The memory 505 may include random access memory (RAM) or read-only memory. Optionally, the memory 505 may include a non-transitory computer-readable storage medium. The memory 505 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 505 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 505 may also be at least one storage device located remotely from the aforementioned processor 501. (Refer to...) Figure 6 The memory 505, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program based on a dual-frequency RFID data collaborative processing method.

[0084] exist Figure 6 In the illustrated electronic device 500, the user interface 503 is mainly used to provide an input interface for the user and acquire user input data; while the processor 501 can be used to call an application program stored in the memory 505 for a data collaborative processing method based on dual-frequency RFID. When executed by one or more processors 501, the electronic device 500 performs one or more methods as described in the above embodiments. It should be noted that, for the foregoing method embodiments, for the sake of simplicity, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0085] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0086] In the various embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between apparatuses or units may be electrical or other forms.

[0087] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0088] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0089] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0090] The above are merely exemplary embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Other embodiments of this disclosure will readily conceive of those skilled in the art upon consideration of the specification and the disclosure of practical truths.

[0091] This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A data collaborative processing method based on dual-frequency RFID, characterized in that, The method, applied to an edge computing device, is communicatively connected to a reader / writer, which is connected to an antenna positioned in a target area for collecting signals from a dual-frequency RFID tag. The dual-frequency RFID tag is mounted on the packaging of the target object. Obtain the pre-stored association mapping table of the dual-frequency RFID, the association mapping table including the unique mapping relationship between the first frequency band identifier code and the second frequency band identifier code; The reader collects radio frequency signal data from dual-frequency RFID, the radio frequency signal data including continuous timing signals of a first frequency band and discrete trigger signals of a second frequency band, wherein the first frequency band is an ultra-high frequency band and the second frequency band is a high frequency band; Generate a differential feature vector based on the continuous time-series signal; The statistical fluctuation value of the differential feature vector is monitored. When the statistical fluctuation value exceeds the preset wake-up threshold, the differential feature vector within the corresponding time window is input into the preset neural network model, the spatial behavior category of the target object is output, and the spatial behavior category and the corresponding timestamp are cached. When the discrete trigger signal is detected, the first frequency band identifier code corresponding to the target object is searched in the association mapping table. Based on the trigger time of the discrete trigger signal, the spatial behavior category within the preset time neighborhood is matched. A data processing strategy is generated based on the matched spatial behavior category and the first frequency band identifier code, and the data processing strategy is transmitted to the cloud.

2. The method according to claim 1, characterized in that, The step of generating a differential feature vector based on the continuous time-series signal includes: Identify the target tag signal and the preset reference tag signal in the continuous time-series signal; The change in radio frequency characteristics of the target tag signal relative to the reference tag signal is calculated to generate a differential feature vector.

3. The method according to claim 2, characterized in that, The continuous time-series signal includes a signal strength sequence and a phase sequence; The step of calculating the change in radio frequency characteristics of the target tag signal relative to the reference tag signal to generate a differential feature vector includes: The phase sequence of the target tag signal is unwrapped to obtain the unwrapped target phase sequence, and the phase sequence of the reference tag signal is unwrapped to obtain the unwrapped reference phase sequence. Subtracting the unwound target phase sequence from the unwound reference phase sequence at the same time yields the relative phase sequence. The relative phase sequence is smoothed and filtered to obtain the differential feature vector.

4. The method according to claim 3, characterized in that, The target tag signal includes a phase sequence based on a period from 0 to 2π; The step of unwinding the phase sequence of the target tag signal to obtain the unwound target phase sequence includes: Calculate the phase difference between the current sampling point and the previous sampling point in the phase sequence in chronological order; Monitor whether the phase difference value has a periodic jump: If the phase difference is greater than π, a compensation value of -2π is added to the phase values ​​of the current sampling point and subsequent sampling points; If the phase difference is less than -π, a compensation value of +2π is added to the phase values ​​of the current sampling point and subsequent sampling points; Based on the phase values ​​after superimposed compensation, the phase sequence is converted into a spatially continuous linear phase sequence, which serves as the target phase sequence after unwinding.

5. The method according to claim 1, characterized in that, The data processing strategy for generating data based on the matched spatial behavior category and the first frequency band identifier includes: Determine whether the matched spatial behavior category contains a preset strongly associated action; If the discrete trigger signal is detected and no spatial behavior category indicating a strongly correlated action is matched within the preset time neighborhood, a model calibration instruction is generated. The model calibration command is used to extract the continuous time-series signal within the preset time neighborhood as positive samples, and to perform online incremental training or parameter fine-tuning on the preset neural network model.

6. The method according to claim 5, characterized in that, The generated model calibration instructions specifically include: Using the timestamp of the discrete trigger signal as the endpoint, backtrack the preset behavior-related delay to determine the potential action time window; Traverse the continuous temporal signals within the potential action time window and calculate the local fluctuation extrema of different sub-time slices; The sub-time slices with the largest local fluctuation extremes that have not triggered the preset wake-up threshold are selected, and their corresponding differential feature vectors are marked as difficult positive samples. The difficult positive samples are input into the preset neural network model to calculate the loss function, and the weight parameters of the neural network model are adjusted based on the calculation results.

7. The method according to claim 1, characterized in that, After monitoring the statistical fluctuation value of the differential eigenvector, the method further includes: The read collision rate corresponding to the continuous time sequence signal of the first frequency band is obtained, wherein the read collision rate is calculated by the reader based on the number of collisions of the collected dual-frequency RFID signals and the total number of reads; When the statistical fluctuation value is close to zero and the collision rate is less than a preset collision threshold, the radio frequency transmission power of the reader to the first frequency band is adjusted.

8. A data collaborative processing device based on dual-frequency RFID, characterized in that, Applied to edge computing devices, the device includes: The table data acquisition module is used to acquire the pre-stored association mapping table of the dual-frequency RFID, the association mapping table including the unique mapping relationship between the first frequency band identifier code and the second frequency band identifier code; The signal acquisition module is used to acquire radio frequency signal data of the dual-frequency RFID collected by the reader. The radio frequency signal data includes a continuous timing signal of the first frequency band and a discrete trigger signal of the second frequency band, wherein the first frequency band is an ultra-high frequency band and the second frequency band is a high frequency band. The signal processing module is used to generate a differential feature vector based on the continuous time-series signal; The category recognition module is used to monitor the statistical fluctuation value of the differential feature vector. When the statistical fluctuation value exceeds the preset wake-up threshold, the differential feature vector within the corresponding time window is input into the preset neural network model, the spatial behavior category of the target object is output, and the spatial behavior category and the corresponding timestamp are cached. The business processing module is used to, when the discrete trigger signal is detected, look up the first frequency band identifier code corresponding to the target object in the association mapping table, match the spatial behavior category in the preset time neighborhood according to the trigger time of the discrete trigger signal, generate a data processing strategy based on the matched spatial behavior category and the first frequency band identifier code, and transmit the data processing strategy to the cloud.

9. An electronic device, characterized in that, The device includes a processor, a memory, a user interface, a communication bus, and a network interface. The processor, the memory, the user interface, and the network interface are respectively connected to the communication bus. The memory is used to store instructions. The user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1-7.

Images