A store inventory management and searching system based on RFID induction

Abstract

The application belongs to the technical field of commercial inventory management, and discloses a store inventory commodity management and searching system based on RFID induction, which comprises the following steps: when a user performs a mobile operation, a high-frequency collection candidate RFID tag phase change time sequence is collected; before matching, the actual motion characteristics of the user are analyzed according to the instantaneous frequency of the sequence, and dynamic normalization processing is performed on the time axis; when there are multiple targets in the matching result, the final uniqueness arbitration is completed by further encoding modulation of the reader transmission power and analysis of the relevance of the echo response. The application converts the traditional mode of probabilistic judgment depending on the signal instantaneous quantity into a deterministic inventory management process based on dynamic process characteristics confirmation by constructing an inventory verification process of active interaction and dynamic feature comparison, thereby solving the uncertainty problem of inventory searching and counting in the commercial environment caused by the dense commodities and unstable signals.

Description

Technical Field

[0001] This invention relates to a store inventory management and retrieval system based on RFID sensing, belonging to the field of commercial inventory management technology. Background Technology

[0002] In current commercial inventory management and sales service processes, the use of ultra-high frequency radio frequency identification (UHF RFID) technology for rapid inventory counting and retrieval of goods has become a common technology for improving efficiency. Its value lies in its ability to achieve batch tag reading without line of sight, simplifying macro-management of inventory. A basic fact in this technology system is that when the electromagnetic waves emitted by the reader propagate in an indoor environment, especially in retail stores or warehouses with stacked goods on metal shelves, the signal does not just travel along a single straight path to reach or return. Instead, it undergoes complex reflection, diffraction, and scattering, ultimately forming the superposition of countless path signals at the receiving antenna. This phenomenon is known as the multipath effect.

[0003] However, when the application scenario shifts from macro-level inventory counting to the high-value service of accurately finding a specific product on densely packed shelves for customers, the aforementioned multipath effect transforms from a background fact into a fundamental challenge restricting the effectiveness of the technology. Specifically, when operators approach the target area with handheld readers, the indication logic based on signal strength (RSSI) generally fails. This is because in complex electromagnetic fields, the spatial distribution of signal strength is not a clearly directional peak, but rather a vague area of ​​similar intensity. As a result, even when operators are close to the target, their device interface often displays multiple adjacent products with extremely high signal strength. Ultimately, in the most critical last-mile delivery stage, this technology degenerates into a primitive method that still requires manual searching, leading to discrepancies between inventory data and physical inventory, low inventory counting efficiency, and thus undermining its core value as an inventory management tool.

[0004] To address this issue, those skilled in the art have attempted several improvement approaches. For instance, some have tried to enhance signal clarity by increasing the reader's transmission power or receiver sensitivity. However, engineering practice has shown that this actually exacerbates multipath interference, widening the ambiguous signal strength region and worsening the problem. Another approach involves applying more complex filtering or data processing algorithms at the receiver to perform in-depth analysis of the acquired static signal strength value. However, the fundamental limitation of these attempts lies in the inherently insufficient dimensionality and quality of the input information. A snapshot of signal strength acquired at a specific point in time, already blurred, contains extremely limited positional information, making subsequent algorithmic processing difficult. It is impossible to generate definitive information out of thin air. Specifically, existing technologies have the following shortcomings: 1. The inventory verification mechanism has inherent defects. The static signal strength physical quantity it relies on does not form a stable and unique mapping relationship with the physical location of the tag in a multipath environment, resulting in probabilistic and uncertain inventory location results; 2. Poor adaptability to inventory management application scenarios. In actual commercial scenarios with dense goods and complex environments, its search accuracy drops sharply, failing to meet the needs for precise inventory unit search from region to individual; 3. Limited improvement paths. Neither enhancing hardware performance nor optimizing backend algorithms can overcome the fundamental bottleneck caused by the insufficient dimension of static signal information. Therefore, how to establish a new working method that no longer attempts to resolve the location from a snapshot of ambiguous static signals, but actively generates a dynamic, highly discriminative, and unique signal feature that can uniquely represent the identity of the target product during the search process, and uses this feature to achieve precise locking in complex environments, is the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides a store inventory management and retrieval system based on RFID sensing. Its main purpose is to solve the problem that existing technologies rely on static signal strength and cannot uniquely and definitively identify densely packed goods in complex environments.

[0006] To achieve the above objectives, this invention provides a store inventory management and retrieval system based on RFID sensing. This system includes a mobile terminal integrating an RFID reader and a display unit, as well as a processor. The processor is configured to:

[0007] The RFID reader is controlled to continuously collect signals returned by multiple RFID tags within the search area at a frequency of not less than 50Hz during the user's handheld mobile terminal's mobile operation, thereby generating a phase change time series for each RFID tag.

[0008] Before performing pattern matching, the instantaneous frequency of each phase change time series is analyzed to extract motion parameters that characterize the actual speed and uniformity of the user's movement operation. Based on the motion parameters, the phase change time series is dynamically normalized on the time axis.

[0009] Each phase change time series after normalization is pattern matched with the reference phase interferogram of the target product stored in the database.

[0010] When more than one RFID tag is identified as the target product based on the pattern matching result, the RFID reader is controlled to dynamically modulate its transmission power according to the internally generated binary encoding sequence, and the echo signal strength response of each of the more than one RFID tag is monitored simultaneously. Based on the cross-correlation calculation result of the echo signal strength response and the binary encoding sequence, the final target product is uniquely determined from the more than one RFID tag.

[0011] Preferably, the mobile operation includes the user holding a mobile terminal and making it perform a linear translation within a range of 20 cm to 40 cm from multiple RFID tags; and the processor uses a dynamic time warping algorithm to perform pattern matching between the normalized phase change time series and the reference phase interference spectrum, and determines the target product based on the matching score.

[0012] Preferably, the movement operation includes continuous movement along at least two mutually orthogonal trajectories; and the processor specifically: while moving along each of the at least two mutually orthogonal trajectories, generates a corresponding phase change time series for the RFID tag; performs pattern matching between the normalized phase change time series generated along each trajectory and the reference phase interference spectrum corresponding to the target product and associated with that trajectory; and finally determines the target product if and only if the phase change time series generated along the at least two mutually orthogonal trajectories are successfully matched with the corresponding reference phase interference spectrum.

[0013] Preferably, the processor is further configured to perform the following operations: evaluate the spectral entropy of the phase change time series before performing dynamic normalization processing on the time axis; and when the spectral entropy is lower than the complexity threshold stored in the database, issue an instruction to the user on the display unit of the mobile terminal to place the other hand behind the mobile terminal and repeat the movement operation.

[0014] Preferably, the processor determines the final target product based on the cross-correlation calculation results of the echo signal strength response and the binary coded sequence, specifically: for more than one RFID tag... Each RFID tag is used to calculate its cross-correlation peak value. , ,in, For the first The echo signal strength sequence of an RFID tag It is a binary encoded sequence. The length of the binary encoded sequence. For time delay; and, select the peak with the maximum cross-correlation. The RFID tags are used as the final target products.

[0015] Preferably, the processor extracts motion parameters and performs dynamic normalization processing on the phase change time series along the time axis. Specifically, it performs a short-time Fourier transform on the phase change time series to obtain its instantaneous frequency time series; based on the instantaneous frequency time series, it calculates the velocity variance characterizing the uniformity of the user's movement speed and identifies the pause time periods during the user's movement where the speed is lower than the speed threshold set by the system; and applies a nonlinear time axis scaling function to compress the data points within the pause time periods and extend the data points within the time periods where the speed is higher than the speed threshold set by the system, thereby generating a normalized phase change time series.

[0016] Preferably, the processor is also configured to perform the following operations: before the user performs a movement operation, control the RFID reader to synchronously modulate a low-frequency identification signal uniquely corresponding to the mobile terminal on the ultra-high frequency carrier signal it transmits; and generate a phase change time series for each RFID tag, specifically by extracting the phase value only from the received echo signal that contains the low-frequency identification signal after demodulation, through a digital phase-locked loop circuit.

[0017] Preferably, the processor is further configured to perform the following operations: during the user's movement operation, in the background, acquire signals from one or more non-target RFID tags (excluding the RFID tag identified as the target product by pattern matching) in parallel, and generate real-time phase change time series for each of the one or more non-target RFID tags; for each real-time phase change time series, extract a spectral feature summary consisting of the total path length of the phase change and the main peak frequency of the spectrum; compare the spectral feature summary with the theoretical spectral feature summary corresponding to each non-target RFID tag and stored in the database using Euclidean distance; and when the Euclidean distance exceeds the difference threshold stored in the database, mark the non-target RFID tag as having an abnormal location.

[0018] Preferably, the reference phase interference spectrum is generated during the goods warehousing stage by the control system operator performing the same movement operation on the target goods as the movement operation, and is stored in the database along with the identity information of the target goods.

[0019] Preferably, the display unit of the mobile terminal is configured to: play animation instructions to guide the user's movement trajectory and speed before the user performs a movement operation, and highlight the identity information of the target product and issue a prompt sound in conjunction with the audio module after the final target product is uniquely identified.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] 1. This invention establishes a method for confirming product search. It begins with a user performing a standardized mobile operation, which causes a continuous change in the relative position between the mobile terminal and each candidate tag. During this physical process, the system synchronously and continuously collects the phase changes of the signals of each candidate tag, generating a phase change time series for each tag. Due to the differences in the local electromagnetic environment of each tag, even for the same type of product, the generated phase change time series have unique characteristics. The system performs pattern matching between this on-site generated sequence and the pre-stored target product reference phase interference spectrum. The consistency of the matching result directly confirms the identity of the target product. This complete process transforms the traditional probabilistic judgment that relies on the instantaneous strength of the signal into a deterministic inventory verification based on dynamic process feature comparison. Signal instability caused by factors such as dense product traffic and personnel movement in the existing commercial environment is no longer interference in this method, but rather constitutes part of the identification features.

[0022] 2. While performing the above-mentioned search and confirmation process, this invention also provides an instant auditing mechanism for the location of goods on shelves. While the user executes the mobile operating system to collect phase information of candidate tags, the background will collect the phase change time series of other non-target tags covered by the energy field of the mobile terminal in parallel. The system does not perform complete pattern matching on these massive sequences, but extracts a low-dimensional spectral feature summary for each sequence in real time and quickly compares it with the theoretical spectral feature summary corresponding to the theoretical location of the non-target tag recorded in the database. If the difference between the two exceeds the preset range, the system will silently mark a location anomaly in the background. In this way, a regular product search operation is completed by incidentally performing an imperceptible inventory of the shelves along the way, integrating two business processes that originally needed to be executed independently and incurred different management costs into a single physical operation.

[0023] 3. This invention possesses a self-adjusting capability to cope with changes in the actual operating environment. Before performing pattern matching between the phase change time series collected on-site and the reference spectrum, the system first performs a pre-analysis of the sequence itself to extract parameters that characterize the actual motion characteristics of the user's movement, such as speed and pauses. Subsequently, based on these parameters, the system dynamically normalizes the time axis of the sequence to correct for time axis stretching or compression caused by the user's non-uniform movement. Furthermore, the system also assesses the complexity of the sequence. When it determines that the phase characteristics are degraded due to the current environment being too open, the system issues an instruction to the user to introduce a standardized passive scatterer and repeat the movement operation. This series of pre-processing and interactions before matching enables the system to proactively adapt to the randomness of the operator's behavior and the diversity of the physical environment, ensuring the applicability of the core inventory verification mechanism.

[0024] 4. When identifying multiple highly similar target products, this invention can initiate a final arbitration process. At this time, the system guides the user to statically point the mobile terminal at the area where these similar targets are located, and controls its RFID reader to dynamically modulate the transmission power according to a preset coding sequence. Since these similar targets have slight differences in their spatial positions, the changes in field strength they receive and the resulting echo signal responses will have a time and amplitude correlation with the power coding sequence transmitted by the reader. By synchronously monitoring and calculating this correlation, the system can uniquely determine the one with the best response correlation and indicate it as the final target. This process avoids the limitation that the main search method cannot make a unique designation when the products are completely identical and placed adjacent to each other. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the overall process of the store product accurate search and guidance system of the present invention;

[0026] Figure 2 This is a schematic diagram showing the effect comparison of the dynamic normalization processing of the time axis in this invention;

[0027] Figure 3 This is a schematic diagram illustrating the functional architecture and user interaction of the store product accurate search and guidance system of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0029] This invention discloses an RFID-based store inventory management and retrieval system. Technically, it is a data processing method deployed within a mobile terminal. This method is executed by a processor and integrates with an RFID reader and display unit to form a closed-loop inventory management tool. The entire data processing procedure logically includes an initial candidate set establishment stage, a dynamic feature acquisition and correction stage, a pattern matching and recognition stage, and an optional arbitration or auditing stage. These stages work together to solve the inventory management problem of uniquely verifying and locating specific goods on shelves in complex, high-density electromagnetic environments. In a typical business application scenario, for example, when store staff need to find a specific item for a customer from a row of clothing racks... In practice, existing management tools based on Signal Strength Indication (RSSI) may fail due to signal ambiguity caused by multipath effects, resulting in multiple adjacent products exhibiting high signal strength. This degrades the search task to manual searching. In view of this, the system of this invention first performs a preliminary area screening using traditional RSSI signal strength. That is, the operator approaches the target area with a mobile terminal that integrates an RFID reader and a display unit. The processor controls the RFID reader to identify all RFID tags with signal strength higher than a preset threshold, such as -50dBm, and the identity information of these tags constitutes a temporary candidate set. The purpose of this step is to focus the search range from a large area to a limited local shelf, providing data input for the subsequent deterministic identification process.

[0030] Subsequently, the system's data processing flow enters the dynamic feature acquisition and correction stage. At this point, an objective application-layer obstacle arises: the movement behavior of operators during actual operations varies non-uniformly, such as pauses due to uncertainty or excessively rapid movement. This variation directly causes the acquired dynamic signal to be non-linearly stretched or compressed on the time axis, leading to a mismatch with the baseline template generated in the database based on standard movements. To address this challenge, the system of this invention employs a data correction procedure that analyzes the signal sequence's own variation characteristics without requiring external sensors. Specifically, during the process where the user performs a preset movement operation—for example, a linear translation within a range of 20 to 40 centimeters from a shelf—based on the animation instructions played by the display unit, the processor not only controls the RFID reader to continuously acquire the signal phase value of each RFID tag in the candidate set at a frequency of no less than 50 Hz, thereby generating a phase change time series for each tag, but also performs a pre-analysis on each of these sequences before pattern matching. This pre-analysis involves analyzing the phase change time series of each RFID tag. The phase change time series is subjected to a Short-Time Fourier Transform (STFT) to obtain its instantaneous frequency time series. Since the rate of phase change is proportional to the user's movement speed, this instantaneous frequency series directly represents the actual motion characteristics of the user's movement operation. Based on this, the processor calculates the speed variance, which represents the speed uniformity, and identifies pause periods when the speed is below the system-set speed threshold, such as 0.05 m / s. Subsequently, the processor applies a non-linear time axis scaling function to compress the data points within the pause periods and extend the data points within the periods when the speed is above the speed threshold. For example, if a 2-second original sequence contains a 0.5-second pause, the data points in that segment will be compressed. Finally, a normalized phase change time series is generated, which is equivalent to a uniform movement of 1.5 seconds on the time axis. This procedure actively adapts to the randomness of the operator's behavior by reversing the motion characteristics of the acquisition process from the signal stream itself and using this as a benchmark to correct the signal stream itself, ensuring the data consistency required for subsequent pattern matching.

[0031] After obtaining the normalized phase change time series, the system enters the pattern matching and recognition stage. The processor retrieves the pre-stored reference phase interference spectrum of the target product from the database. This reference spectrum is generated by the operator performing the same movement operation on the target product during the product warehousing stage. The Dynamic Time Warping (DTW) algorithm is used to calculate the matching score between the on-site acquired sequence and the reference spectrum. Because in a specific local electromagnetic environment, even identical products will have unique phase interference spectra due to slight differences in their placement and orientation, resulting from controlled micro-movements. Therefore, the RFID tag corresponding to the sequence with a matching score much higher than other candidate tags is identified as the target product by the system. Its identification information is then highlighted on the display unit and the audio module is activated. The system uses a prompt tone to transform a probabilistic judgment into a deterministic identification process based on dynamic process feature comparison. However, in certain specific business management situations, such as when two identical items are placed adjacent to each other, the main identification process may output two or more high-probability matching results. In this case, the system cannot make a final unique arbitration. To solve this decision-making problem caused by the symmetry of the physical world, the system of this invention is configured to automatically trigger a final arbitration process when this scenario is determined to be entered. In this process, the system first guides the user to statically point the mobile terminal at the area where these similar targets are located through the display unit. Subsequently, the processor controls the RFID reader to stop emitting a stable power electromagnetic field and instead emit a binary encoded sequence generated internally, such as... =101101 dynamically modulates the transmitted power, where 1 represents standard power and 0 represents a 5% reduction in power; during this period, the system synchronously monitors and records the echo signal intensity response sequences of these multiple similar targets. Because of their slight spatial differences, the changes in field strength received by each candidate tag, and the resulting echo signal intensity response, will differ in timing and amplitude from the power-coded sequence transmitted by the reader. The processor calculates the cross-correlation peak value for each candidate tag. Its calculation formula is ,in, The length of the binary encoded sequence. For time delay; ultimately, the peak value with the largest cross-correlation is selected. The RFID tag is used as the final target product. This arbitration mechanism based on active channel interrogation completes the process from fuzzy inventory to locking a single inventory unit by analyzing the differences in the responses of different beacons to coded field strength disturbances.

[0032] Furthermore, to improve the operational efficiency of the entire business data management system, this invention also provides a real-time shelf location auditing mechanism that accompanies the core search process. Specifically, during the user's movement to find the target product, the processor, in an independent background thread, concurrently collects signals from one or more other non-target RFID tags covered by the mobile terminal's energy field and generates real-time phase change time series for each. Considering that performing complete pattern matching on all non-target tags would consume excessive computational resources, the system does not store these original sequences. Instead, it extracts a low-dimensional spectral feature summary for each sequence in real time, consisting of the total path length of the phase change and the main peak frequency of the spectrum. Subsequently, the system processes... The device compares the real-time extracted spectral feature summary with the theoretical spectral feature summary corresponding to the non-target RFID tag retrieved from the database using Euclidean distance. The theoretical spectral feature summary is generated when the product is put on the shelf or the system is initialized at the location where the product should theoretically be. When the calculated Euclidean distance exceeds the difference threshold stored in the database, for example, this threshold can be determined by three times the statistical standard deviation of multiple measurements at the same location, the system marks the non-target RFID tag as having an abnormal location in the background and reports the relevant information to the management system. In this way, a routine product search operation is incidentally completed with a seamless inventory check of the shelves along the way, integrating two originally independent business management processes into a single physical operation.

[0033] It should be noted that this system is designed with mechanisms to address various boundary conditions and potential risks. For example, to address co-channel interference caused by multiple RFID readers operating simultaneously in the same commercial space, the system can be configured to first execute a private beacon establishment procedure before the user performs a movement operation. This involves controlling the RFID reader to synchronously modulate a low-frequency identification signal, such as a 1kHz square wave, uniquely corresponding to the mobile terminal onto its transmitted UHF carrier signal. Correspondingly, at the receiving end, the system uses a digital phase-locked loop circuit to extract the phase value only from the received echo signal that contains the low-frequency identification signal after demodulation. This mechanism filters out signals from other interfering readers at the source, ensuring the purity of the data required for the core identification process. Furthermore, to address potential missed detections due to the anisotropic orientation of goods, the system's guidance command can be upgraded from a single unidirectional linear translation to a continuous movement (such as an L-shape) containing at least two mutually orthogonal trajectories. At this point, the processor generates a corresponding phase change time series for each trajectory. Only when the sequences generated along all trajectories successfully match the corresponding reference spectrum in the database can the target product be finally determined. This approach avoids the risk of recognition failure caused by a single scanning path being in the blind zone of the tag antenna through multi-dimensional information collection. Furthermore, to address scenarios where phase features degrade due to insufficient multipath effect in an overly open electromagnetic environment, the processor is configured to evaluate the spectral entropy of the phase change time series before performing time axis normalization. When the spectral entropy is lower than the complexity threshold stored in the database, the system determines that it is currently in a feature degradation state and issues a command to the user on the mobile terminal's display unit to place their other hand about 10 centimeters behind the mobile terminal and repeat the movement operation. The user's hand here plays the role of a standardized passive scatterer, artificially enriching the local electromagnetic environment, thereby ensuring the applicability of the core recognition mechanism in different environments.

[0034] Example 1: In the daily inventory management of a high-density warehouse-style apparel store, the WMS system in the background recorded that a blue cashmere sweater with a specific ID should be stored on the shelf in section C on the third floor. However, when the store staff conducted a routine check, they found that the product was missing from the designated location. At the same time, a handheld RFID device based on RSSI signal strength detected dozens of tags with similar signal strength in the area. These responses came from other clothing hanging closely on the shelves and the complex electromagnetic reflections generated by the surrounding metal pillars and shelves. This made the signal strength-based search method unable to make an effective judgment in this scenario, constituting a management anomaly where the inventory data did not match the physical entity.

[0035] To address this inventory anomaly, the staff activated the RFID-based store inventory management and retrieval system of this invention. Using a handheld mobile terminal, they entered Zone C on the third floor. The system first performed a preliminary screening process based on RSSI signal strength, adding 15 RFID tags with signal strengths higher than -50dBm to a candidate set. Subsequently, the system switched to the dynamic feature acquisition and correction stage. The display unit prompted the staff to use the mobile terminal to move it at approximately a constant speed from left to right across the clothing rack area in front of them, maintaining a distance of about 30 centimeters. During this movement, the processor continuously acquired the phase change time series of all 15 tags in the candidate set at a frequency of 50Hz. After acquisition... After completion, the system did not immediately perform pattern matching. Instead, it first initiated the time axis dynamic normalization processing procedure described in the previous implementation. For example, for one of the candidate labels, the system analyzed its original phase change time series through short-time Fourier transform, identified a pause period of 0.4 seconds in the middle of the movement caused by the operator observing the shelf number, during which the speed was below 0.05 m / s, and compressed the data points on the time axis to generate a normalized phase change time series that eliminated the random interference of the operator's behavior. This step provides a corrected and standardized data input for subsequent pattern matching, enabling the recognition mechanism to operate under a unified benchmark.

[0036] After normalizing the phase change time series of all candidate tags, the system performs pattern matching using a dynamic time warping algorithm on these 15 sequences against the reference phase interferometry spectrum of the target blue cashmere sweater stored in the database. The matching results show that the matching score of one tag's sequence with the reference spectrum is much higher than all other tags. The system then highlights and locks the ID information of the target product on the display unit and indicates that it is located in a turnover box of clothes waiting to be sorted under the current shelf, under several other garments. The system's processing method is to generate a dynamic signature that can represent the physical location of the target through a dynamic human-machine collaborative interaction, and transform a localization problem that depends on the spatial distribution of the signal into a pattern recognition problem at the information theory level, thereby avoiding the inherent limitations of static signal strength in such scenarios. Following this guidance, the item was located in the turnover box, completing the closed-loop processing of this inventory anomaly event. The completion of this search also triggered the system's real-time shelf location audit mechanism in the background. At the same time that the operator was performing the movement operation, the system simultaneously collected the phase change time series of another 35 non-target tags within the energy field range, and extracted a spectral feature summary for each sequence in real time, consisting of the total path length of the phase change and the main peak frequency of the spectrum. By comparing these summaries with the theoretical spectral feature summaries of the corresponding tags in the database using Euclidean distance, the system marked two items in the background as having location anomalies, with the difference in their summaries exceeding a preset threshold. A report containing the IDs of these two items and their current location information was automatically generated and pushed to the store management backend, providing data support for subsequent inventory and shelf organization operations.

[0037] Example 2: To verify the recognition performance of the technical solution of the present invention under different tag densities, this comparative experiment was conducted. The results show that in a high-density tag environment, the recognition method based on phase change time series has higher accuracy and stability compared to the method based on Received Signal Strength Indication (RSSI). The experiment was conducted in a standardized electromagnetic shielded room. The platform consisted of a 2m x 1m x 0.5m metal shelf, a mobile terminal integrating an ultra-high frequency RFID reader module with an operating frequency of 920MHz to 925MHz and capable of outputting phase values ​​at a frequency of not less than 50Hz, and a set of cardboard sheets affixed with the same type of passive RFID tags. To ensure the reproducibility of the operation, the mobile terminal was fixed on a three-axis robotic arm, with a uniform rotation speed of 0.3m / s. The device performs a linear translational movement at a speed of 30 cm in front of the shelf. Two scenarios were set up: medium density (20 labels / layer) and high density (50 labels / layer). Tests were conducted on a control group using the RSSI recognition method and an experimental group using the method of this invention. In the control group using the RSSI recognition method, under medium density conditions, 13 out of 20 independent tests were successfully identified, with an accuracy rate of 65%. When the density increased to high density, the accuracy rate dropped to 25%, with only 5 successful identifications. In contrast, the experimental group using the method of this invention, under the same medium density conditions, successfully identified labels in all 20 tests, with an accuracy rate of 100%. Under high density conditions, 19 out of 20 tests were successfully identified, maintaining an accuracy rate of 95%.

[0038] Data trends show that the recognition accuracy of the control group decreased significantly with increasing tag density. This is because RSSI, as a scalar value, no longer forms a reliable mapping relationship with physical location in environments with complex multipath effects. In contrast, the method used in the experimental group is based on phase change time series. The high-dimensional feature information contained in this series constructs a unique spatial process feature for each tag under a specific motion path. This feature has a higher tolerance to the signal complexity caused by multipath effects, and thus can maintain a high level of recognition accuracy in high-density interference environments. The experimental results confirm the technical solution of this invention from a data perspective, providing an effective technical path for solving the problem of accurate product search in dense commercial environments.

[0039] Example 3: This example combines Figures 1 to 3 This document describes a store inventory management and retrieval system based on RFID sensing, such as... Figure 1As shown in the diagram, the user movement operation at the top serves as the core physical input of the system, branching off into two parallel processing branches: the main search process on the left and the parallel auditing mechanism on the right. The main search process sequentially includes three core processing steps: dynamic feature acquisition and correction, pattern matching and recognition, and an optional arbitration stage, ultimately outputting the search result. The parallel auditing mechanism sequentially includes four processing steps: acquiring non-target tag signals, extracting spectral feature summaries, comparing with theoretical features, and marking abnormally positioned goods. The system database located at the center of the flowchart provides data support for the two processing branches. It internally stores the reference phase interferometry spectrum, theoretical spectral feature summaries, and various threshold parameters, and interacts with the above processing stages through data calls or data call / comparison.

[0040] like Figure 2 As shown, the horizontal axis of the graph represents time (seconds), and the vertical axis represents phase value (degrees). The graph contains two contrasting curves. The original sequence (including pauses), represented by the solid line, shows waveform distortion caused by user movement pauses around 1.8 seconds on the time axis. The normalized sequence, represented by the dashed line, shows that this distortion was corrected after being processed by the system's dynamic time axis normalization algorithm, thus presenting a smoother and more regular periodic change, restoring the ideal waveform corresponding to uniform movement operation.

[0041] like Figure 3 As shown in the diagram, two core participants are clearly identified: the store operators on the left and the system administrator in the upper right, as well as an external related system, namely the warehouse management system in the lower right. The store operators are the main initiators of the core function of accurately finding products. This core function is related to three specific sub-use cases: receiving search guidance, performing real-time auditing of product location, and handling search arbitration. The system administrator mainly interacts with the system initialization and parameter calibration functions. The warehouse management system provides basic data updates and maintenance for the entire system through interaction with the function of updating the product benchmark spectrum.

[0042] Example 4: In the initial deployment phase of a large electronic component warehousing center, the system faced an engineering problem: ensuring stable operation of its core identification and auditing functions across thousands of different storage locations and with varying operator habits. The core issue was establishing a set of quantifiable initial operating parameters adapted to the specific deployment environment for key algorithm modules such as dynamic time axis normalization and real-time shelf merchandise location auditing. To address this, a system initialization and parameter calibration process was implemented during the initial deployment. The first step of this process involved offline calibration of the core parameters in the dynamic time axis normalization module, specifically the speed threshold used to distinguish between movement and pause. During this process, a system administrator, using a mobile terminal, repeatedly performed 10 uniform linear translation operations in a designated open test area, guided by the display unit. The system recorded the entire phase change time series of a reference RFID tag during these 10 operations and performed a short-time Fourier transform on each sequence to calculate its instantaneous frequency sequence. The processor then calculated the overall average of these 10 instantaneous frequency sequences. vs. population standard deviation And the instantaneous frequency threshold corresponding to the speed threshold of the system. Set as This threshold was subsequently embedded in the system configuration and used as the criterion for all subsequent dynamic timeline normalization processing. In other words, in actual operation, when the system analyzes the phase change time series generated by user operations, if the instantaneous frequency is lower than a certain threshold... The time period will be identified as a pause and its data points will be compressed along the time axis.

[0043] The second step of this process is to establish a theoretical spectral feature summary database covering the entire warehouse area and its corresponding difference thresholds for the real-time auditing mechanism of shelf merchandise locations. The administrator, carrying a mobile terminal pre-loaded with a database of reference phase interferometric spectra for various merchandise, systematically collects features from each warehouse location, starting from the first location A-01-01. Specifically, a merchandise with a known ID is placed in each location, and the administrator performs five standardized movement operations on that location using the mobile terminal. The system collects five corresponding phase change time series and calculates a spectral feature summary for each series, consisting of the total path length of the phase change and the main peak frequency of the spectrum. Subsequently, the processor calculates the vector average of these five spectral feature summaries and stores this result as the theoretical spectral feature summary of location A-01-01 in the database. Simultaneously, the processor also calculates the pairwise Euclidean distances between these five summaries and obtains the standard deviation of these distance values. Therefore, the difference threshold used to determine the location anomaly of storage location A-01-01. Set as The data is then stored along with the theoretical spectral feature summary of the storage location. This process is repeated sequentially until the feature data of all storage locations in the warehouse have been collected and calibrated. After completing the above two stages, the system deployment of the warehousing center is complete. Its database not only contains the reference phase interference spectrum of the goods in the warehouse, but also establishes a set of unique theoretical spectral feature summaries and difference thresholds for each storage location corresponding to the physical environment of the warehouse. At the same time, the system's motion compensation algorithm also obtains the judgment basis based on the calibration of measured data. Thus, the system has the basic operating parameters required for data processing in subsequent search and inventory operations.

[0044] Example 5: When a new batch of goods arrives at an existing SKU, to address potential individual differences in RFID tags or variations in tag placement between different production batches, this system executes an incremental update procedure for the reference phase interference spectrum. Specifically, when the Warehouse Management System (WMS) enters information for a new batch of goods, it triggers a prompt instructing the operator to remove a sample from the new batch and perform a standardized movement operation to collect its reference phase interference spectrum. After collection, the system does not overwrite the existing reference spectrum of the old batch for that SKU. Instead, it associates the newly generated reference spectrum with the new batch number and stores them together in the database, linking it to the SKU along with the reference spectrum of the old batch. Thus, in subsequent search operations, when the processor needs to perform pattern matching for the SKU, it compares the phase change time series collected on-site with the reference phase interference spectra of all batches belonging to that SKU in the database. As long as a match is found with the reference spectrum of any one of the batches, the system determines that the target product has been found.

[0045] When the system needs to initiate the final arbitration process to distinguish between two adjacent identical target products, the depth of its transmit power dynamic modulation is configured to be adaptively adjusted based on signal strength. Before transmitting the binary coded sequence, the processor first obtains the average echo signal strength (RSSI) values ​​of the two target tags and retrieves the corresponding modulation depth parameter from a preset lookup table based on the RSSI value. This lookup table is generated through experimental calibration during the system initialization phase. Its contents divide the RSSI value into multiple intervals and assign a modulation depth to each interval. Generally, when the RSSI value is higher than -40dBm, the system selects a smaller modulation depth, such as 3%, and when the RSSI value is lower than -60dBm, the system selects a larger modulation depth, such as 10%. The processor then uses this retrieved modulation depth parameter to perform subsequent transmit power dynamic modulation and cross-correlation calculations, so that the operating parameters of the arbitration process are adapted to the current signal strength of the target.

[0046] Example 6: To establish an objective benchmark for evaluating the quality of user movement operations in actual operations, this system executes a modeling procedure for reference features during the initialization phase. This procedure utilizes the phase change time series generated by dozens of standard movement operations performed by multiple operators, collected during the calibration phase in the previous examples. The processor does not use these sequences solely for parameter calibration but further extracts the signal envelope variability and total path length of the phase change for each sequence, forming a two-dimensional feature vector. Subsequently, the system calculates the centroid and statistical distribution of all these feature vectors from standard operations, thereby constructing a clustering model representing qualified movement operations in the feature space. Based on the boundary of this model, a movement quality threshold is determined. In subsequent actual search operations, before performing pattern matching, the system first extracts the same two-dimensional feature vector from the phase change time series generated by the user's current operation and calculates its distance from the centroid of the aforementioned clustering model. Only when this distance is less than the movement quality threshold is the system considered the data acquisition valid and continues to execute subsequent steps; otherwise, the process is terminated and an instruction to correct the movement operation is issued to the user.

[0047] In the pattern matching stage, a quantifiable judgment criterion is established to distinguish between high-confidence matches and fuzzy matches caused by weak signals or environmental interference. The system uses an offline optimization method to determine the absolute threshold of matching confidence. In this optimization procedure, the system first constructs a score distribution model for successful matches based on the matching score data of a large number of confirmed successful search cases in the historical database. At the same time, it also constructs a highest score distribution model for failed matches based on a large number of cases with failed searches or scanning in areas without targets. The optimization goal of the system is to find a score threshold that can maximize the distinction between the scores of successful matches and failed matches. That is, at this threshold, the product of the true positive rate and the true negative rate reaches its maximum. The optimal split point calculated by this procedure is set as the global confidence threshold of the system. In the actual search process, even if any candidate tag achieves the highest relative matching score, if its absolute score is lower than this confidence threshold, it will still be judged as an invalid match by the system, and a prompt that no target was found will be returned to the user.

[0048] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A RFID-sensing based store inventory management and search system, the system comprising a mobile terminal integrated with a RFID reader and a display unit and a processor, characterized in that, The processor is configured to perform the following inventory verification steps: The processor is configured to perform the following inventory verification steps: The processor is configured to perform the following inventory verification steps: The processor is configured to perform the following inventory verification steps: The processor is configured to perform the following inventory verification steps:

2. The RFID-sensing based store inventory management and searching system according to claim 1, wherein, The movement operation includes continuous movement along at least two mutually orthogonal trajectories; and the processor is specifically configured to: generate a corresponding phase change time sequence for each RFID tag when moving along each of the at least two mutually orthogonal trajectories; and perform pattern matching between the phase change time sequence generated along each trajectory and a corresponding reference phase interference spectrum associated with the target product and related to the trajectory.

3. The RFID-based system for managing and searching inventory of a store according to claim 1, wherein, The movement operation includes continuous movement along at least two mutually orthogonal trajectories; and the processor is specifically configured to: generate a corresponding phase change time sequence for each RFID tag when moving along each of the at least two mutually orthogonal trajectories; and perform pattern matching between the phase change time sequence generated along each trajectory and a corresponding reference phase interference spectrum associated with the target product and related to the trajectory. The movement operation includes continuous movement along at least two mutually orthogonal trajectories; and the processor is specifically configured to: generate a corresponding phase change time sequence for each RFID tag when moving along each of the at least two mutually orthogonal trajectories; and perform pattern matching between the phase change time sequence generated along each trajectory and a corresponding reference phase interference spectrum associated with the target product and related to the trajectory.

4. The RFID-sensing based store inventory management and searching system according to claim 1, wherein, The movement operation includes continuous movement along at least two mutually orthogonal trajectories; and the processor is specifically configured to: generate a corresponding phase change time sequence for each RFID tag when moving along each of the at least two mutually orthogonal trajectories; and perform pattern matching between the phase change time sequence generated along each trajectory and a corresponding reference phase interference spectrum associated with the target product and related to the trajectory. The movement operation includes continuous movement along at least two mutually orthogonal trajectories; and the processor is specifically configured to: generate a corresponding phase change time sequence for each RFID tag when moving along each of the at least two mutually orthogonal trajectories; and perform pattern matching between the phase change time sequence generated along each trajectory and a corresponding reference phase interference spectrum associated with the target product and related to the trajectory. The movement operation includes continuous movement along at least two mutually orthogonal trajectories; and the processor is specifically configured to: generate a corresponding phase change time sequence for each RFID tag when moving along each of the at least two mutually orthogonal trajectories; and perform pattern matching between the phase change time sequence generated along each trajectory and a corresponding reference phase interference spectrum associated with the target product and related to the trajectory.

5. The RFID-sensing based store inventory management and searching system according to claim 1, wherein, The processor determines the final target product based on the cross-correlation calculation result of the echo signal strength response and the binary coded sequence, specifically: for the first RFID tag in the plurality of RFID tags, a cross-correlation peak value is calculated , , , , , , , , , wherein, is the echo signal strength sequence of the first RFID tag, is the binary coded sequence, is the sequence length of the binary coded sequence, is the time delay; and the RFID tag with the largest cross-correlation peak value is selected as the final target product.

6. The RFID-sensing based store inventory management and searching system according to claim 1, wherein, The processor extracts the motion parameter and performs dynamic normalization on the phase change time sequence in time axis, specifically: by performing short-time Fourier transform on the phase change time sequence, the time sequence of instantaneous frequency is obtained; based on the time sequence of instantaneous frequency, the speed variance representing the uniformity of user moving speed is calculated, and the pause time period in which the speed is lower than the speed threshold set by the system is identified in the user moving process; and a nonlinear time axis scaling function is applied to compress the data points in the pause time period and to extend the data points in the time period in which the speed is higher than the speed threshold set by the system, thereby generating the normalized phase change time sequence.

7. The RFID-sensing based store inventory management and searching system according to claim 1, wherein, The processor is further configured to perform the following operations: before the user performs the moving operation, the RFID reader is controlled to synchronously modulate a low-frequency identification signal corresponding to the mobile terminal on the ultra-high frequency carrier signal emitted thereby; and for each RFID tag, the phase change time sequence is generated, specifically: by the digital phase-locked loop circuit, only the echo signal received and containing the low-frequency identification signal after demodulation is subjected to phase value extraction.

8. The RFID-sensing based store inventory management and searching system according to claim 1, wherein, The processor is further configured to perform the following operations: during the user performing the moving operation, in the background, one or more non-target RFID tags other than the RFID tag of the target commodity determined by pattern matching are collected in parallel, and real-time phase change time sequences of the one or more non-target RFID tags are generated respectively; for each real-time phase change time sequence, a spectral feature abstract composed of the total path length of phase change and the main peak frequency of spectrum is extracted in real time; the spectral feature abstract is compared with the theoretical spectral feature abstract corresponding to the non-target RFID tag and stored in the database in terms of Euclidean distance; and when the Euclidean distance exceeds the difference threshold stored in the database, the non-target RFID tag is marked as position abnormal.

9. The RFID-sensing based store inventory management and searching system according to claim 1, wherein, The reference phase interference spectrum is generated by controlling the system operator to perform the same moving operation on the target commodity in the commodity warehousing stage, and is stored in the database together with the identity information of the target commodity.

10. The RFID-sensing based store inventory management and searching system according to claim 1, wherein, The display unit of the mobile terminal is configured to: before the user performs the moving operation, play an animation instruction to guide the moving trajectory and moving speed of the user, and after the final target commodity is uniquely determined, highlight the identity information of the target commodity and link the audio module to issue a prompt sound.

Images