Intelligent warehouse management method and system

Abstract

The application relates to the field of intelligent warehousing, and specifically discloses an intelligent warehousing management method and system, the method comprising the following steps: collecting radio frequency phase fingerprints, barycenter spatial distribution data and appearance panoramic images when goods are first stored in a warehouse through an AI intelligent fusion terminal, generating multi-modal identity anchoring data; comparing and calculating the current data collected in the operation stage with the anchoring data to bind a deviation index, triggering multi-modal cross verification to generate a binding integrity index; determining a trusted level based on the index, starting multi-terminal collaborative verification to update the index when the trusted level is lower than a threshold, generating a differentiated control instruction and recording; the system comprises an AI intelligent fusion terminal, a multi-terminal collaborative network, a warehousing master station system and a goods transmission control mechanism. The application realizes trusted binding of labels and goods bodies through radio frequency phase fingerprints, combines multi-terminal collaborative verification and block chain storage, effectively prevents cheating behaviors such as goods switching, and improves the safety and intelligent level of warehousing management.

Description

Technical Field

[0001] This invention relates to the field of intelligent warehousing technology, specifically to an intelligent warehousing management method and system. Background Technology

[0002] In the power industry's warehousing management, the types of materials are diverse, including a large number of irregular metal devices such as insulators, surge arresters, instrument transformers, and transformer bushings. These materials are characterized by high value, variety, and different shapes, placing extremely high demands on the accuracy and security of warehousing management. Existing intelligent warehousing systems generally use RFID technology for material identification, which involves attaching electronic tags to the surface of materials and using readers to read the EPC codes on the tags to achieve inbound and outbound management.

[0003] However, existing technologies have the following drawbacks: First, RFID tags only provide identification and lack a means to verify the physical binding relationship between the tag and the material itself. Since most power materials are made of metal, they are subject to complex reflection interference from radio frequency signals. Unscrupulous individuals can exploit this characteristic to transfer tags from high-value materials to low-value materials through "tag peeling and re-attaching," thus committing fraud. Existing systems only verify the existence of the tag ID and cannot detect such cheating. Second, existing systems' judgment of material status is limited to binary logic such as the presence or absence of tags and weight matching, failing to detect subtle anomalies such as minor tag displacement, slight material deformation, and multiple items being handled at once, making it difficult to detect and trace numerous violations. Third, warehousing decisions are made using a single-point model. Individual AI fusion terminals are prone to misjudgments due to environmental electromagnetic noise, sensor malfunctions, and other factors. Furthermore, there is a lack of coordination mechanisms between terminals, making it impossible to cross-verify information from adjacent nodes. Finally, operation logs are stored centrally, posing a risk of data tampering and audit blind spots, making it difficult to meet the stringent requirements of the power industry for full-process traceability of high-value materials.

[0004] Therefore, how to achieve a reliable binding between tags and the actual materials, build a dynamic integrity verification mechanism, and improve the collaborative decision-making capabilities of multiple terminals and the data anti-tampering capabilities have become technical problems that urgently need to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent warehouse management method and system to solve the technical problem in existing RFID warehouse management where tags are easily peeled off and replaced, leading to falsification of material identities.

[0006] To solve the above-mentioned technical problems, the present invention specifically provides the following technical solution: An intelligent warehouse management method includes the following steps: S1. Collect radio frequency phase fingerprints, center of gravity spatial distribution data and panoramic images of the materials when they first enter the warehouse through AI intelligent fusion terminals, generate multimodal identity anchoring data and store it; S2. During the operation phase, the current radio frequency phase fingerprint, current center of gravity spatial distribution data and current appearance image of the materials are collected in real time, and compared with the multimodal identity anchoring data to calculate the binding deviation index, trigger multimodal cross-validation, and generate the binding integrity index. S3. Determine the trust level based on the binding integrity index. If it is lower than the preset threshold, initiate multi-terminal collaborative verification to update the binding integrity index. Generate differentiated control instructions based on the updated index and record them.

[0007] As a preferred embodiment of the present invention, S1 specifically includes: S11. Multimodal identity anchoring is performed by an AI intelligent fusion terminal deployed at the warehouse site. The AI ​​intelligent fusion terminal integrates an RFID reading and writing module, a visual acquisition module, and a gravity sensing module. S12. During the initial material entry and calibration stage, the RFID reading and writing module is controlled to scan the electronic tags attached to the surface of the target material at multiple preset frequency points, extract the signal phase information and backscatter intensity at each frequency point, and construct a radio frequency phase fingerprint that characterizes the electromagnetic coupling characteristics between the electronic tag and the specific metal material. S13. Synchronously control the gravity sensing module to collect static weight and center of gravity spatial distribution data of the material, and control the vision acquisition module to collect panoramic images of the material's appearance. S14. The radio frequency phase fingerprint, the spatial distribution data of the center of gravity, and the panoramic image of the appearance are spatiotemporally aligned to generate multimodal identity anchoring data of the material and stored in the local trusted storage area.

[0008] As a preferred embodiment of the present invention, S12 specifically includes: S121. Control the RFID reading and writing module and its connected multi-angle antenna array to perform continuous frequency sweeping in the ultra-high frequency band of 840~960MHz in preset frequency steps; S122. At each frequency sweep point, excite the electronic tag attached to the surface of the target material to backscatter, and record the phase angle and signal strength of the backscatter signal corresponding to the frequency point; S123. After traversing all preset frequency points, the obtained phase angle sequence and signal strength sequence are combined to form a frequency domain response curve characterizing the near-field coupling effect between the tag antenna and the surface of the metal material. S124. Normalize the frequency domain response curve and extract feature points to generate a unique radio frequency phase fingerprint that can reflect the electromagnetic coupling characteristics and relative spatial position relationship between the tag and a specific metal material.

[0009] As a preferred embodiment of the present invention, S2 specifically includes: S21. During the material entry and exit operation phase, the AI ​​intelligent fusion terminal acquires the current radio frequency phase fingerprint, current center of gravity spatial distribution data and current appearance image of the electronic tag in real time; S22. Compare the current radio frequency phase fingerprint with the radio frequency phase fingerprint stored in step S1 to determine the phase consistency and calculate the binding deviation index, which reflects the integrity of the physical binding relationship between the tag and the material. S23. Match and analyze the current centroid spatial distribution data with the centroid spatial distribution data stored in step S1. If there is an anomaly, trigger the multimodal cross-validation process, perform visual feature verification based on the current appearance image, and comprehensively generate the binding integrity index.

[0010] As a preferred embodiment of the present invention, S22 specifically includes: S221. Align the current radio frequency phase fingerprint acquired in real time with the radio frequency phase fingerprint stored in step S1 to ensure that the two sets of data are compared at the same frequency point; S222. Calculate the phase difference at each corresponding frequency point, and calculate the root mean square error of the phase difference at all frequency points to obtain the phase deviation value; simultaneously calculate the signal strength change rate at each corresponding frequency point to obtain the signal strength fluctuation value. S223. The phase deviation value and the signal strength fluctuation value are weighted and fused to generate a binding deviation index; the value of the binding deviation index is positively correlated with the degree of tag peeling, replacement or deformation of metal material, and is used to quantitatively characterize the integrity of the physical binding relationship between the tag and the material body.

[0011] As a preferred embodiment of the present invention, S23 specifically includes: S231. Match and analyze the current center of gravity spatial distribution data with the center of gravity spatial distribution data stored in step S1, calculate the offset of the center of gravity projection position and the change in vertical height, and generate the center of gravity deviation index. S232. Determine whether the center of gravity deviation index exceeds the preset center of gravity anomaly threshold or the binding deviation index exceeds the preset phase anomaly threshold. If either condition is met, trigger the multimodal cross-validation process. S233. In the cross-validation process, the visual acquisition module is controlled to perform multi-angle imaging of the material, extract surface feature points in the current appearance image and perform feature matching with the stored panoramic appearance image, and calculate the visual similarity score. S234. The center of gravity deviation index, binding deviation index and visual similarity score are weighted and fused to generate a binding integrity index; the binding integrity index is used to comprehensively characterize the overall credibility of materials in the three dimensions of identity, posture and appearance.

[0012] As a preferred embodiment of the present invention, S3 specifically includes: S31. Determine the trust level of the inbound / outbound operation based on the binding integrity index. When the trust level is higher than a preset threshold, generate a standard inbound / outbound control instruction. S32. When the trust level is lower than the preset threshold but higher than the minimum admission threshold, the multi-terminal collaborative verification mode is activated, and a collaborative verification request is broadcast to other AI intelligent fusion terminals deployed in the adjacent area to obtain the historical operation records of materials and the radio frequency environmental noise characteristics of the adjacent area. S33. Update the binding integrity index through a distributed consensus protocol, and generate differentiated control instructions based on the updated binding integrity index. The differentiated control instructions include allowing passage, restricting passage, or triggering security isolation. S34. The differentiated control instructions and verification process data are encrypted and recorded in the local operation log of the blockchain structure.

[0013] As a preferred embodiment of the present invention, S32 specifically includes: S321. When the trust level is lower than a preset threshold but higher than the minimum admission threshold, the current AI intelligent fusion terminal automatically starts the multi-terminal collaborative verification mode. S322. Send a collaborative verification request to other adjacent AI intelligent fusion terminals via a broadcast protocol, the request including the material identification and the current collection timestamp; S323. Receive response data returned by each adjacent terminal, the response data including the trend of radio frequency phase fingerprint change of materials in the historical operation records of each terminal, the trend of center of gravity distribution change, and the real-time radio frequency environmental noise characteristics of the location of each terminal. S324. Perform timestamp alignment and validity verification on the data returned by each adjacent terminal, remove timed-out or abnormal response data, package the filtered multi-source collaborative data, and prepare to input it into the distributed consensus protocol for updating the binding integrity index.

[0014] As a preferred embodiment of the present invention, S33 specifically includes: S331. Input the filtered multi-source collaborative data into the distributed consensus protocol engine, and perform environmental interference compensation on the radio frequency phase fingerprint of the current terminal based on the radio frequency environmental noise characteristics to eliminate the influence of electromagnetic noise in the adjacent area on the fingerprint acquisition accuracy. S332. Perform time-series correlation analysis between the compensated current radio frequency phase fingerprint and the fingerprint change trend in historical operation records to identify the evolution pattern of the fingerprint in the time dimension; S333. Combining the trend of the center of gravity distribution change, a weighted consensus algorithm is used to cross-validate and evaluate the confidence of the data returned by each terminal, and an updated binding integrity index is generated; S334. Based on the numerical range of the updated binding integrity index, generate differentiated control instructions, including green instructions that allow passage, yellow instructions that restrict passage and trigger manual review, or red instructions that trigger security isolation.

[0015] An intelligent warehouse management system, used to implement an intelligent warehouse management method, includes: an AI intelligent fusion terminal, a multi-terminal collaborative network, a warehouse master station system, and a material transfer control mechanism, specifically: At least one AI-powered intelligent fusion terminal is deployed at a warehouse operation node, the AI-powered intelligent fusion terminal comprising: The RFID read / write module and multi-angle antenna array operate in the ultra-high frequency band of 840~960MHz, support multi-frequency scanning, and are used to collect the radio frequency phase information and backscatter intensity of electronic tags on the surface of materials to construct radio frequency phase fingerprints. A six-axis gravity sensor array, consisting of multiple pressure sensors arranged in a grid, is used to collect data on the static weight of materials and the three-dimensional spatial distribution of the center of gravity. The visual acquisition module includes a high-definition camera and a supplementary lighting unit, used to acquire panoramic images and multi-angle imaging of the material's appearance; The edge trusted computing module has a built-in material fingerprint database and behavior pattern recognition model, which is used to perform multimodal identity anchoring, binding deviation index calculation, binding integrity index generation and trust level decision in steps S1 to S3. The blockchain storage module is used to encrypt and record differentiated control instructions and verification process data in the local operation log of the blockchain structure, ensuring the immutability and traceability of the data; The multi-terminal collaborative network consists of multiple AI intelligent fusion terminals deployed at different work nodes interconnected through an industrial local area network. It is used to execute the distributed consistency protocol in the multi-terminal collaborative verification mode. Each terminal exchanges material historical operation records and radio frequency environmental noise characteristics through a broadcast mechanism and jointly updates the binding integrity index. The warehouse master station system, which is connected to the multi-terminal collaborative network, includes: The data audit module is used to periodically collect blockchain operation logs from various terminals to conduct cross-terminal data auditing and trace abnormal behavior. The model optimization module is used to globally optimize the weight parameters of the radio frequency environment interference model and the distributed consensus protocol based on the radio frequency environment noise characteristics and historical operation records uploaded by each terminal. The personnel access control module is used to dynamically adjust the access control permissions of corresponding personnel based on the cumulative number of violations in the violation database. The material transfer control mechanism, connected to the AI ​​intelligent fusion terminal, includes intelligent access control, cargo location status indicator lights, and an AGV scheduling module, used to receive and execute the differentiated control commands.

[0016] Compared with the prior art, the present invention has the following advantages: 1. This invention breaks through the limitation of existing RFID systems that rely solely on EPC codes. By extracting the phase information of the electromagnetic coupling between the tag and the metal material through multi-frequency scanning, it constructs a unique and unclonable radio frequency phase fingerprint. This fingerprint is then fused with the center of gravity distribution data and appearance image to generate a digital twin ID card with electromagnetic-mechanical-visual triple binding, thus physically preventing cheating behaviors such as tag replacement and material substitution.

[0017] 2. This invention compares the phase consistency of real-time radio frequency phase fingerprints and anchor fingerprints, transforming tag integrity into a quantifiable binding deviation index. When the center of gravity is abnormal or the deviation index exceeds the limit, a visual review is triggered, and a binding integrity index is generated by fusion. This achieves a leap from static identity recognition to dynamic integrity verification, and can keenly capture subtle anomalies such as tag displacement and material deformation.

[0018] 3. When the trust level is ambiguous, this invention initiates collaborative verification between adjacent terminals, and upgrades single-point decision-making to networked collaborative decision-making by cross-correcting the binding integrity index through a distributed consensus protocol; at the same time, the verification data is encrypted and recorded in the blockchain log, and the immutability is used to achieve full-process traceability, which significantly improves the robustness of the system and the trustworthiness of the data. Attached Figure Description

[0019] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the method described in Embodiment 1 of the present invention.

[0021] Figure 2 This is a framework diagram of the system described in Embodiment 2 of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The concepts involved in this application will first be described with reference to the accompanying drawings. It should be noted that the following descriptions of various concepts are only for the purpose of making the content of this application easier to understand and do not constitute a limitation on the scope of protection of this application; furthermore, the embodiments and features in the embodiments of this application can be combined with each other unless otherwise specified. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0024] Example 1: As Figure 1 As shown, the present invention provides an intelligent warehouse management method, comprising the following steps: S1. Collect radio frequency phase fingerprints, center of gravity spatial distribution data, and panoramic images of the materials upon their initial entry into the warehouse using an AI-powered intelligent fusion terminal; generate and store multimodal identity anchoring data; specifically including: S11. The deployment and hardware integration configuration of the execution entity for the multimodal identity anchoring step are as follows: The AI-powered intelligent fusion terminals deployed at key operational nodes in the warehouse, such as inbound and outbound ports and transfer areas within the warehouse, act as the main execution entities, performing multimodal identity anchoring operations. These AI-powered intelligent fusion terminals are vertical or desktop integrated edge computing devices, equipped with a support platform at their base. Internally, they integrate edge computing units, a local trusted storage area, and multi-protocol communication interfaces, including: The RFID reader module is embedded inside the terminal and uses an ultra-high frequency reader chip that supports the ISO18000-6C protocol. The operating frequency covers the adjustable frequency band of 840~960MHz. It is connected to a multi-angle antenna array consisting of three antenna arrays arranged on the top, side and diagonal of the support platform through an RF switching switch. Each antenna array is used to radiate radio frequency energy from the vertical, horizontal and diagonal directions to excite the electronic tag.

[0025] The vision acquisition module includes a high-resolution industrial camera and an LED fill light array mounted on a fixed bracket directly above the support platform. The camera's optical axis is vertically aligned with the geometric center of the support platform to capture top-down and side-view images of the materials.

[0026] The gravity sensing module includes a six-axis pressure sensing array embedded in the structural layer of the support platform. The array consists of six strain gauge pressure sensors distributed at the four corners and the geometric center of the support platform. Each sensor converts the pressure signal into an electrical signal through a Wheatstone bridge circuit, and then connects to the data acquisition unit inside the terminal through a signal conditioning circuit.

[0027] The edge computing unit includes an embedded processor with an integrated hardware security module (HSM) or trusted platform module (TPM) for performing multimodal data processing and trusted storage control; the local trusted storage area is an encrypted flash memory area physically bound to the hardware security module for storing multimodal identity anchoring data.

[0028] S12. Construction process of radio frequency phase fingerprint: During the initial material warehousing calibration stage, the RFID reading and writing module is controlled to scan the electronic tags attached to the surface of the target material at multiple preset frequency points, extract the signal phase information and backscatter intensity at each frequency point, and construct a radio frequency phase fingerprint characterizing the electromagnetic coupling characteristics between the electronic tag and the specific metallic material; this construction process is specifically implemented through the following sub-steps: S121. Multi-angle frequency sweep excitation control: Controls the RFID read / write module and its connected multi-angle antenna array to continuously sweep frequencies in preset frequency steps within the 840~960MHz ultra-high frequency band; specifically: The frequency synthesizer inside the RFID read / write module is based on direct digital frequency synthesis technology or phase-locked loop frequency synthesis technology. It sequentially switches the local oscillator frequency according to a preset frequency point list. This list includes at least three frequency points selected in an equal or non-equal interval manner in the 840~960MHz ultra-high frequency range, such as 865MHz, 915MHz, and 956MHz, to generate a radio frequency carrier signal at a specific frequency point. After the radio frequency carrier signal is amplified by a power amplifier, it is distributed to a multi-angle antenna array through a radio frequency switching switch. The multi-angle antenna array includes a vertically polarized antenna array positioned directly above the support platform, a horizontally polarized antenna array positioned laterally, and a circularly polarized antenna array positioned obliquely. By controlling the on / off timing of the radio frequency switching switch, the three antenna arrays are sequentially turned on, radiating radio frequency energy to the electronic tag attached to the surface of the target material from the vertical, horizontal, and oblique directions, respectively, thus forming multi-angle excitation. At each frequency, each antenna element maintains a preset dwell time to ensure that the electronic tag fully acquires energy and establishes a stable backscatter link. After completing the excitation at that frequency, the frequency synthesizer switches to the next frequency in preset frequency steps, repeating the above multi-angle excitation process until all preset frequency points are traversed.

[0029] S122. Backscattering signal parameter extraction: At each frequency sweep point, the electronic tag attached to the surface of the target material is excited to backscatter, and the phase angle and signal strength of the backscattering signal corresponding to the frequency point are recorded; specifically: When the radio frequency carrier radiated by the multi-angle antenna array illuminates the electronic tag, the chip inside the electronic tag switches the impedance state of its antenna port to reflect and modulate the incident radio frequency signal, forming a backscattered signal carrying the tag's identification information. After the backscattered signal is received by the multi-angle antenna array, the transmitted signal and the received signal are separated by a circulator or directional coupler. The received signal is amplified by a low-noise amplifier to improve the signal-to-noise ratio. The amplified signal is input to a quadrature demodulator and mixed with the local carrier provided by the frequency synthesizer to generate two orthogonal baseband I / Q signals, namely the in-phase component I and the quadrature component Q.

[0030] The baseband processing unit of the RFID read / write module samples the I / Q signals and calculates the arctangent of the I / Q signal ratio using arctangent operation to obtain the phase delay of the backscattered signal relative to the local carrier, i.e., the backscattered signal phase angle. At the same time, the backscattered signal strength is obtained by calculating the square root of the I / Q signal magnitude or by measuring the signal power level through a logarithmic amplifier. For each swept frequency point and each antenna element, the backscattered signal phase angle and signal strength values ​​corresponding to that frequency point and angle are independently recorded to form a signal parameter pair for that frequency point.

[0031] S123. Frequency Domain Response Curve Construction: After traversing all preset frequency points, the acquired phase angle sequence and signal intensity sequence are combined to form a frequency domain response curve characterizing the near-field coupling effect between the tag antenna and the metal material surface; specifically: For each antenna element in the multi-angle antenna array, a continuous phase-frequency response curve is plotted with the frequency value of the swept frequency point on the x-axis and the phase angle of the backscattered signal recorded by the antenna element at each frequency point on the y-axis; a continuous intensity-frequency response curve is plotted with the frequency value of the swept frequency point on the x-axis and the intensity of the backscattered signal recorded by the antenna element at each frequency point on the y-axis.

[0032] Since the target material is made of metal, its surface conductivity will change the radiation impedance of the electronic tag antenna and the near-field electromagnetic field distribution, causing the phase and intensity of the backscattered signal to fluctuate with frequency. Materials with different metal materials, different geometries and different tag attachment positions will produce phase-frequency curves and intensity-frequency curves with unique waveform characteristics.

[0033] By summarizing multiple sets of phase-frequency and intensity-frequency curves obtained from the three antenna arrays, a complete frequency domain response dataset is formed. This dataset constitutes a frequency domain response curve that characterizes the electromagnetic coupling characteristics and relative spatial position relationship between the electronic tag and a specific metallic material, reflecting the unique electromagnetic scattering fingerprint of the material in the ultra-high frequency band.

[0034] S124. Radio Frequency Phase Fingerprint Generation: The frequency domain response curve is normalized and feature points are extracted to generate a unique radio frequency phase fingerprint. This fingerprint reflects the electromagnetic coupling characteristics and relative spatial relationship between the tag and a specific metallic material; specifically: The signal strength values ​​at each frequency point are normalized to eliminate amplitude fluctuations caused by differences in read / write distance and antenna gain, mapping the signal strength to a value range of 0 to 1. Secondly, the phase angle values ​​at each frequency point are de-wrapped. By detecting the phase jump between adjacent frequency points, i.e., when the phase difference exceeds π, 2π is added or subtracted, the phase wrapping caused by the periodicity of the arctangent function is eliminated, generating a continuous phase-frequency curve.

[0035] Feature points are extracted from the normalized frequency domain response curve, including: extreme points of the phase-frequency curve, zero-crossing frequencies, points with the maximum slope of the curve, and peak points and -3dB bandwidth boundary points of the intensity-frequency curve. The frequency coordinates, phase values, and intensity values ​​of these feature points are encoded into feature vectors in a preset order. Combined with the spatial location identifiers of the multi-angle antenna array, a unique radio frequency (RF) phase fingerprint is generated. This RF phase fingerprint essentially transforms the near-field characteristics of the electromagnetic coupling between the tag and the metallic material into a quantifiable frequency domain feature identifier. Even if the electronic tag's identification code EPC is copied, its RF phase fingerprint will show significant differences because the copied tag cannot accurately replicate the relative position and coupling state between the original tag and the specific metallic material.

[0036] S13. Synchronous Acquisition and Execution of Gravity and Visual Data: While constructing the RF phase fingerprint in step S12, simultaneously control the gravity sensing module and the visual acquisition module to acquire data; specifically: S131. For the gravity sensing module, after the material is placed on the bearing platform and stabilized, the six-axis pressure sensing array converts the mechanical deformation into an analog voltage signal through a Wheatstone bridge circuit. After being filtered and amplified by the signal conditioning circuit, it is sampled into a digital signal by an analog-to-digital converter. The edge computing unit reads the digital signals from the six sensors, sums them to calculate the static weight value of the material, and calculates the projection coordinates of the material's center of gravity in the two-dimensional coordinate system of the bearing plane, i.e., the spatial distribution data of the center of gravity, based on the force difference of the four corner pressure sensors.

[0037] S132. For the vision acquisition module, its high-resolution industrial camera, under the uniform illumination of the LED fill light array, captures a top-view panoramic image of the material from directly above to obtain the top surface geometric features, and captures a side image from the side to obtain the height direction structural features; the acquired images are processed by the image signal processor inside the camera for noise reduction, white balance and distortion correction to generate a panoramic image of the material's appearance.

[0038] The above acquisition process is controlled by a hardware synchronous trigger signal to ensure that the completion time of gravity data latching, image exposure and radio frequency phase fingerprint acquisition in step S12 has microsecond-level time consistency, and achieves strict timing alignment of multimodal data.

[0039] S14. Spatiotemporal Alignment and Trusted Storage: Radio frequency phase fingerprints, centroid spatial distribution data, and panoramic images of the exterior are spatiotemporally aligned to generate multimodal identity anchoring data for the materials and stored in a local trusted storage area; specifically: Spatiotemporal alignment includes two dimensions: temporal alignment and spatial alignment. Time alignment is achieved through hardware synchronization trigger signals. After completing the feature extraction in step S124, the RFID reading and writing module outputs a synchronization trigger pulse, which triggers the gravity sensing module and the vision acquisition module to simultaneously latch the current data, ensuring that the three types of data correspond to the physical state of the materials at the same moment. Spatial alignment is achieved by establishing a local three-dimensional spatial coordinate system. The geometric center of the platform is taken as the origin, the horizontal direction is the XY plane, and the vertical upward direction is the positive Z-axis. The phase center coordinates of each antenna in the multi-angle antenna array of the RFID read / write module and the camera optical center coordinates of the visual acquisition module are mapped to this coordinate system to record the spatial geometric relationship during radio frequency phase fingerprint acquisition.

[0040] After completing spatiotemporal alignment, the radio frequency phase fingerprint, centroid spatial distribution data, panoramic image of the appearance, and corresponding timestamps and spatial coordinate parameters are packaged to generate multimodal identity anchoring data. This data is then written to the local trusted storage area within the AI ​​intelligent fusion terminal. This storage area is encrypted and protected by the Hardware Security Module (HSM) or Trusted Platform Module (TPM) to ensure the confidentiality and integrity of the anchoring data and prevent subsequent unauthorized tampering.

[0041] S2. During the operation phase, the current radio frequency phase fingerprint, current center of gravity spatial distribution data, and current appearance image of the materials are collected in real time. These are compared with multimodal identity anchoring data to calculate the binding deviation index, trigger multimodal cross-validation, and generate a binding integrity index. Specifically, this includes: S21. Real-time multimodal data acquisition triggering and acquisition, specifically: During the material entry and exit operations, when materials are placed on the carrier platform of the AI ​​intelligent fusion terminal or in the electromagnetic field coverage area of ​​the RFID reader module, the real-time acquisition process is triggered; the automatic start is triggered by the weight increase signal detected by the gravity sensor module or the electronic tag radio frequency field strength change signal detected by the RFID reader module.

[0042] The AI-powered intelligent fusion terminal controls the RFID reading and writing module and its multi-angle antenna array. Following the same preset frequency list in step S12, including the same frequency step, the same antenna array switching sequence, and the same excitation power parameters, it performs multi-angle frequency scanning excitation on the electronic tag, and collects the current backscattered signal phase angle and current signal strength at each frequency point in real time to construct the current radio frequency phase fingerprint.

[0043] The system synchronously controls a six-axis pressure sensor array to collect the current static weight value and the force distribution of each sensor. It calculates the current center of gravity projection coordinates through a torque balance algorithm to form the current center of gravity spatial distribution data. It also controls a vision acquisition module to collect current appearance images under the illumination of an LED supplementary lighting array, including top-view panoramic images and side-view structural images.

[0044] The above real-time acquisition process ensures the consistency of the timestamps of the current radio frequency phase fingerprint, the current centroid spatial distribution data and the current appearance image through hardware synchronous triggering signals, ensuring that the three types of data characterize the physical state of the material at the same moment.

[0045] S22. Radio Frequency Phase Fingerprint Comparison and Binding Deviation Index Calculation: The current radio frequency phase fingerprint is compared with the anchored radio frequency phase fingerprint stored in the local trusted storage area in step S1 to calculate the binding deviation index, which reflects the integrity of the physical binding relationship between the tag and the material; specifically: S221. Frequency Alignment and Data Preprocessing: The real-time acquired current RF phase fingerprint is aligned with the RF phase fingerprint stored in step S1 to ensure that the two sets of data are compared at the same frequency point; specifically: Extract all recorded frequency values ​​of the swept frequency points from the current radio frequency phase fingerprint data structure, denoted as f1, f2, ..., f n This forms the current frequency sequence; the anchored radio frequency phase fingerprint stored in step S14 is read from the local trusted storage area, and the recorded sweep frequency values ​​are extracted and denoted as f'1, f'2, ..., f' nAn anchor frequency sequence is formed. Since the two sets of data are collected by the same AI intelligent fusion terminal according to the same preset frequency list in step S121, theoretically the frequency values ​​of the current frequency sequence and the anchor frequency sequence should be completely consistent. However, due to the temperature drift of the crystal oscillator or the difference in the calibration of the frequency synthesizer, there may be a slight frequency deviation. Therefore, it is necessary to establish a correspondence by accurately matching the frequency values. For each frequency value within the tolerance range of ±10kHz, the frequency point is determined as the corresponding frequency point.

[0046] For each successfully matched frequency point, extract the current phase angle θ of the antenna array corresponding to that frequency point from the current radio frequency phase fingerprint. current With current signal strength RSSI current Simultaneously, the corresponding frequency point and the anchoring phase angle θ of the corresponding antenna array are extracted from the anchored radio frequency phase fingerprint. anchor With anchor signal strength RSSI anchor If data for a certain frequency point is missing due to electromagnetic interference or tag response failure, the linear interpolation method is used to calculate the estimated value of the missing frequency point using data from two adjacent valid frequency points, or the missing frequency point is marked as invalid data and removed from subsequent calculations.

[0047] S222. Phase Deviation and Intensity Fluctuation Calculation: Calculate the phase difference at each corresponding frequency point, and perform root mean square error calculation on the phase difference at all frequency points to obtain the phase deviation value; simultaneously calculate the rate of change of signal intensity at each corresponding frequency point to obtain the signal intensity fluctuation value; specifically: For each aligned frequency point, calculate the phase difference Δθ = θ current -θ anchor Since the phase angle is periodic, ranging from -π to π or 0 to 2π, the calculation results need to be unwrapped: if Δθ>π, then let Δθ=Δθ-2π; if Δθ<-π, then let Δθ=Δθ+2π, to ensure that the phase difference reflects the true phase shift rather than a periodic jump.

[0048] The phase difference values ​​of all frequency points are arranged into a phase difference sequence {Δθ1, Δθ2, ..., Δθ}. n The root mean square error (RMSE) of the sequence is calculated by first summing the squares of the differences, dividing by the number of effective frequency points n, and then taking the square root to obtain the phase deviation value. This value quantitatively reflects the overall deviation of the phase characteristics caused by the change in the coupling relationship between the tag and the metal material. The rate of change of signal strength at each frequency point is calculated synchronously using the formula |RSSI. current - RSSI anchor | / |RSSI anchorAlternatively, the absolute value of the difference in the logarithmic domain can be used to obtain the relative intensity change rate at each frequency point. The arithmetic mean or root mean square value of the change rate at all frequency points can be calculated to obtain the signal intensity fluctuation value, which reflects the degree of change in RF link efficiency and tag coupling state.

[0049] S223. Weighted Fusion to Generate Binding Deviation Index: The phase deviation value and signal strength fluctuation value are weighted and fused to generate a binding deviation index; specifically: Based on the high sensitivity of phase characteristics in metallic environments to the physical binding state of tags and materials, and the high sensitivity of signal strength to changes in environmental distance, a first weighting coefficient w1 is set to assign a phase deviation value, typically ranging from 0.6 to 0.8. A second weighting coefficient w2 is set to assign a signal strength fluctuation value, typically ranging from 0.2 to 0.4, and w1 + w2 = 1.

[0050] Calculate the weighted fusion value Index raw = w1×(phase deviation value / π) + w2×signal strength fluctuation value, where the phase deviation value is normalized by dividing by π to map its value range to the interval between 0 and 1; for IIndex raw Amplitude limiting is performed; if the value is greater than 1, it is set to 1; if it is less than 0, it is set to 0, resulting in the final binding deviation index. The index ranges from 0 to 1. The larger the value, the greater the difference between the current radio frequency phase fingerprint and the anchored radio frequency phase fingerprint. It is positively correlated with the degree to which the tag is attached to other materials after being peeled off, the tag is replaced with a heterogeneous tag with the same ID, or the metal material is deformed. It is used to quantitatively characterize the integrity of the physical binding relationship between the electronic tag and the material body. When the binding deviation index exceeds the preset phase anomaly threshold, such as 0.3, it is determined that there is a risk of binding relationship damage.

[0051] S23. Centroid Matching Analysis and Multimodal Cross-Validation: The current centroid spatial distribution data is matched with the centroid spatial distribution data stored in step S1. If an anomaly is detected, the multimodal cross-validation process is triggered. Visual features are reviewed based on the current appearance image, and a binding integrity index is generated comprehensively. Specifically: S231. Calculation of Center of Gravity Deviation Index: Match and analyze the current spatial distribution data of the center of gravity with the spatial distribution data of the center of gravity stored in step S1, calculate the offset of the projected position of the center of gravity and the change in vertical height, and generate the center of gravity deviation index; specifically: Extract the current centroid projection coordinates (X) from the current centroid spatial distribution data. current , Y current This coordinate is calculated using a torque balance algorithm based on the current force values ​​(F1, F2, F3, F4) of the four corner pressure sensors in the six-axis pressure sensor array, i.e., X. current= (F2 + F3, - F1 - F4) × L / (F total ), Y current = (F3 + F4 - F1 - F2) × W / (F total ), where L and W are half the length and width of the bearing platform, respectively, and F total This is the total weight.

[0052] Read the anchored centroid projection coordinates (X) stored in step S1 from the local trusted storage area. anchor , Y anchor ); Calculate the Euclidean distance between the two coordinate points in the two-dimensional coordinate system of the bearing plane to obtain the offset D of the centroid projection position. xy =√[(X current - X anchor )² + (Y current - Y anchor )²).

[0053] Simultaneous calculation of vertical height change: By comparing the difference between the current force ratio of each sensor in the six-axis pressure sensor array and the anchor force ratio, if the force ratio of the central sensor increases significantly while the ratio of the four corners decreases, it indicates that the center of gravity of the material has risen, such as being raised or replaced with a high-density material; conversely, it indicates that the center of gravity has decreased. Alternatively, the difference between the current height value of the top of the material and the anchor height value can be obtained through the structured light ranging function of the vision acquisition module, which is used as the vertical height change ΔH.

[0054] offset D of the center of gravity projection position xy The normalized horizontal deviation is obtained by normalizing the deviation from the preset allowable unit weight baseline value, such as 1 mm per kilogram. The normalized height deviation is obtained by normalizing the vertical height change ΔH with the preset height tolerance, such as 5 mm. The two normalized results are then weighted and summed according to preset weights, namely, a horizontal deviation weight of 0.7 and a height deviation weight of 0.3, to generate a center of gravity deviation index. The index ranges from 0 to 1. The larger the value, the more severe the replacement of the material body, the tilting of its posture, or the change in its stacking state.

[0055] S232. Multimodal cross-validation trigger judgment: Determine whether the centroid deviation index exceeds a preset centroid anomaly threshold or the binding deviation index exceeds a preset phase anomaly threshold. If either condition is met, the multimodal cross-validation process is triggered; specifically: A preset center of gravity anomaly threshold is set, typically ranging from 0.2 to 0.3, corresponding to an equivalent offset of the center of gravity exceeding 20% ​​to 30% of the static weight. This is compared with a preset phase anomaly threshold, typically ranging from 0.25 to 0.35, corresponding to a significant deviation of the radio frequency phase fingerprint. The edge computing unit of the AI ​​intelligent fusion terminal compares the center of gravity deviation index calculated in step S231 with the preset center of gravity anomaly threshold, and simultaneously compares the binding deviation index calculated in step S22 with the preset phase anomaly threshold.

[0056] If the center of gravity deviation index is greater than the preset center of gravity anomaly threshold, it indicates that the six-axis pressure sensor array has detected a significant change in the weight distribution of the material body, which may indicate that the material has been completely replaced or that its internal structure has been damaged. If the binding deviation index is greater than the preset phase anomaly threshold, it indicates that the RFID reader module has detected a significant change in the electromagnetic coupling characteristics between the tag and the metal material, which may indicate that the tag has been peeled off and reattached or that the tag has been transferred to other metal materials.

[0057] When either of the above two conditions is met, the edge computing unit determines that the detection result of the current single sensor dimension is abnormal, but it is necessary to exclude false alarms caused by environmental interference or operational errors. Therefore, the multimodal cross-validation process is triggered. This process introduces independent verification of the visual dimension to review and confirm the anomalies detected by radio frequency and gravity, distinguishing between malicious attack behavior, such as tag transfer attack, and normal handling operation error, so as to avoid business interruption caused by misjudgment of a single dimension.

[0058] S233. Visual Feature Verification Execution: In the cross-validation process, the visual acquisition module is controlled to perform multi-angle imaging of the materials, extract surface feature points from the current appearance image, and perform feature matching with the stored panoramic appearance image to calculate the visual similarity score; specifically: After triggering the multimodal cross-validation process, the AI ​​intelligent fusion terminal controls the vision acquisition module to adjust its electric gimbal or robotic arm to change the camera angle and perform multi-angle fine imaging of materials. This includes taking a vertical shot from directly above to obtain a panoramic image from above, taking a side view texture image from a 45-degree angle, and focusing on the nameplate area or specific texture area on the surface of the material, such as the insulator skirt or the surface of metal castings, to perform local high-definition imaging.

[0059] The scale-invariant feature transform (SIFT) algorithm or the accelerated robust feature SURF algorithm is used to perform key point detection and feature descriptor extraction on the acquired current appearance image, generating a set of 128-dimensional or 64-dimensional feature descriptor vectors with rotation, scale, and illumination invariance. The appearance panoramic image and its pre-extracted anchored feature descriptor library stored in step S1 are read from the local trusted storage area. The nearest neighbor search algorithm, such as fast approximate nearest neighbor search based on KD tree, is used to match the feature descriptors of the current image with the anchored feature descriptor library, calculate the Euclidean distance or Hamming distance between each descriptor, and filter out matching point pairs whose distance is less than a preset threshold, such as the ratio of the nearest neighbor distance to the second nearest neighbor distance being less than 0.8. The number of successfully matched feature points is counted, and the ratio of this number to the total number of feature points in the anchored feature descriptor library is calculated. Alternatively, the inverse of the average distance between all matched point pairs is calculated to obtain a visual similarity score, which is normalized to the range of 0 to 1. The closer the score is to 1, the more similar the appearance. If the visual similarity score is higher than a preset appearance threshold, such as 0.85, it is determined to be the same material, and the aforementioned radio frequency or gravity anomalies are due to environmental interference. If the score is lower than the threshold, it is determined that the appearance of the material has changed significantly or there is a risk of replacement.

[0060] S234. Binding Integrity Index Generation: The centroid deviation index, binding deviation index, and visual similarity score are weighted and fused to generate the binding integrity index; specifically: The centroid deviation index and binding deviation index are normalized and inversely mapped so that their range is mapped to 0 to 1, with smaller values ​​indicating more normality. The centroid confidence weight W is then calculated. center = 1 - min(center of gravity deviation index / preset center of gravity anomaly threshold, 1), calculate the radio frequency confidence weight W rf = 1 - min(binding deviation index / preset phase anomaly threshold, 1); for visual similarity score S visual Directly used as the visual credibility weight W visual The range is from 0 to 1.

[0061] Set the fusion weight coefficients for the three dimensions, such as W. center Weight 0.3, W rf Weight 0.4, W visual A weight of 0.3 reflects the core role of radio frequency features in physical binding verification; the weighted fusion value Integrity is calculated. Index = 0.3×W center + 0.4×W rf + 0.3×W visualThe binding integrity index is obtained; the index ranges from 0 to 1. The higher the value, the higher the overall credibility of the material in the three dimensions of identity legality verified by radio frequency phase fingerprint, posture stability verified by center of gravity distribution, and appearance consistency verified by vision, and the more complete the physical binding relationship.

[0062] S3. Determine the trust level based on the binding integrity index. If it is lower than a preset threshold, initiate multi-terminal collaborative verification to update the binding integrity index. Generate differentiated control instructions based on the updated index and record them; specifically including: S31. Trust Level Determination and Standard Instruction Generation: The trust level of inbound and outbound operations is determined based on the binding integrity index. When the trust level is higher than a preset threshold, standard inbound and outbound control instructions are generated; specifically: The edge computing unit of the AI ​​intelligent fusion terminal compares the binding integrity index generated in step S234, ranging from 0 to 1, with the preset level thresholds; sets a first preset threshold, such as 0.8, as the dividing point between high and medium trust levels, and sets a second preset threshold, such as 0.5, as the dividing point between medium and low trust levels, i.e., the minimum admission threshold.

[0063] When the binding integrity index is greater than or equal to the first preset threshold, the trust level is determined to be high trust level, indicating that the multimodal identity verification of the material is completely passed, and the physical binding relationship, appearance characteristics and weight distribution of the tag and the material itself are all in a normal state. At this time, the edge computing unit calls the preset warehouse business rule library, combines it with the current inbound and outbound business process, and generates standard inbound and outbound control instructions. The standard inbound and outbound control instructions include standard passage instructions for controlling material transmission mechanisms, such as the start of automated guided vehicles or electric roller conveyors, instructions for updating the inventory status of the material in the warehouse management database, and instructions for reporting normal inbound and outbound events to the warehouse master station system. After the standard inbound and outbound control instructions are signed and protected in the Trusted Execution Environment (TEE) or Hardware Security Module (HSM) inside the AI ​​intelligent fusion terminal, they are sent to the material transmission control mechanism for execution to ensure the security of the instruction generation and transmission process.

[0064] S32. When the trust level is lower than a preset threshold but higher than a minimum admission threshold, a multi-terminal collaborative verification mode is activated, broadcasting a collaborative verification request to other adjacent AI intelligent fusion terminals to obtain historical operation records of materials and radio frequency environmental noise characteristics of the nearby area; specifically: S321. When the binding integrity index is lower than the first preset threshold but higher than or equal to the second preset threshold, the confidence level is determined to be medium confidence level, indicating that there is a slight anomaly in the sensor detection of a single terminal, such as a slight deviation in the radio frequency phase fingerprint or a slight shift in the center of gravity distribution, but it cannot be determined whether it is environmental interference or potential attack behavior; at this time, the current AI intelligent fusion terminal automatically switches to the multi-terminal collaborative verification mode.

[0065] S322. Broadcasting of Collaborative Verification Requests: When the trust level is lower than a preset threshold but higher than the minimum admission threshold, the current AI intelligent fusion terminal automatically initiates a multi-terminal collaborative verification mode and sends collaborative verification requests to other adjacent AI intelligent fusion terminals via a broadcast protocol; specifically: The requesting node invokes its network communication interface, supporting RJ45 wired industrial Ethernet or IEEE 802.11ax Wi-Fi 6 wireless protocol, to construct a collaborative verification request message. The frame structure of the request message includes: an Ethernet frame header or Wi-Fi MAC header, a transport layer header based on the UDP protocol, and an application layer payload. The application layer payload is encoded in JSON or Protocol Buffers format and contains the following fields: material identification, current acquisition timestamp, request type identifier, unique device identifier of the requesting node, and digital signature.

[0066] The requesting node sends the request message to the predefined adjacent deployment area within the warehouse LAN via a broadcast address or a publish / subscribe topic based on the MQTT / AMQP protocol. This typically covers other AI intelligent fusion terminals within a radius of 50 meters, ensuring that other terminals in the same warehouse area or adjacent work areas can receive the verification request.

[0067] S323. Reception and parsing of response data: Receive response data returned by each adjacent terminal. The response data includes the trend of radio frequency phase fingerprint changes, the trend of center of gravity distribution changes, and the real-time radio frequency environmental noise characteristics of each terminal's location in the historical operation records of each terminal; specifically: Upon receiving a collaborative verification request message, each AI intelligent fusion terminal deployed in adjacent locations first verifies the validity of the digital signature in the request message through a hardware security module. After confirming the legitimate identity of the requesting node, it initiates the local data query and collection process. Each response node queries its local trusted storage area or local database to retrieve the historical operation records corresponding to the electronic tag of the material, extracts the anchor radio frequency phase fingerprint, historical centroid spatial distribution data and corresponding historical timestamp of the material when it last passed through the terminal, and forms radio frequency phase fingerprint change trend data containing historical phase deviation sequence, and centroid distribution change trend data containing historical centroid offset sequence. Simultaneously, each response node controls its RFID reader / writer module to perform background frequency scanning according to the same preset frequency list as step S121, even when no materials are placed there. This process collects the phase and intensity base of the current environmental noise and generates real-time radio frequency environmental noise characteristics, including the background noise power spectral density and phase jitter variance at each frequency point. Each response node encapsulates the above data into a response message, which includes: the device identifier of the response node, the historical radio frequency phase fingerprint obtained, historical centroid spatial distribution data, real-time radio frequency environmental noise characteristics, data generation timestamp, and the digital signature of the response node. Each response node sends the response message back via unicast, directly to the IP address of the requesting node, or via multicast. The requesting node listens for and receives these response data through its network interface and temporarily stores them in a memory buffer for processing.

[0068] S324. Validation and Filtering of Response Data: Timestamp alignment and validity verification are performed on the data returned by each adjacent terminal. Timed-out or abnormal response data is discarded. The filtered multi-source collaborative data is packaged and prepared for input into the distributed consensus protocol for updating the binding integrity index. Specifically: After sending the collaborative verification request, the requesting node starts a timer. After receiving the response messages from each responding node, it extracts the data in the message to generate a timestamp, calculates the difference between the timestamp and the current system time, and if the difference exceeds the timeout threshold, such as 500 milliseconds, the response is determined to be a late response and is discarded. Extract the digital signature from the message, and verify the signature validity using the public key certificate of the response node in the pre-set trusted anchor list stored in the local trusted storage area. If the verification fails, it is judged as an abnormal response and is removed. The response data that passes the timeliness and identity verification are validated for content validity. The system checks whether the historical radio frequency phase fingerprint contains complete frequency point data, whether the historical centroid spatial distribution data contains valid coordinate values, and whether the real-time radio frequency environmental noise characteristics are within a reasonable range. Valid response data that passes the validation is retained, and the response latency and historical credibility weight of each response node are recorded. The requesting node packages the filtered multi-source collaborative data, including the historical radio frequency phase fingerprint set, historical centroid distribution set, real-time environmental noise feature set and corresponding node weight of each valid response node, into a structured data packet according to a unified time sequence format, and prepares it to be input into the distributed consensus protocol engine for subsequent interference compensation and consensus calculation.

[0069] S33. Distributed Consistency Protocol Execution and Differentiated Instruction Generation: The distributed consistency protocol updates the binding integrity index, and differentiated control instructions are generated based on the updated binding integrity index. These differentiated control instructions include allowing passage, restricting passage, or triggering security isolation; specifically: S331. Environmental Interference Compensation and Fingerprint Correction: The filtered multi-source collaborative data is input into the distributed consensus protocol engine. Based on the radio frequency environmental noise characteristics, environmental interference compensation is performed on the radio frequency phase fingerprint of the current terminal to eliminate the impact of electromagnetic noise in the vicinity on the fingerprint acquisition accuracy; specifically: The edge computing unit of the requesting node inputs the multi-source collaborative data packaged in step S324 into the distributed consensus protocol engine deployed locally. This engine is based on a lightweight and practical Byzantine fault-tolerant algorithm or Raft consensus algorithm, which is adapted to the limited environment of edge computing resources.

[0070] The protocol engine first analyzes the real-time RF environmental noise characteristics returned by each response node, extracts the background noise power spectral density and phase jitter variance at each location, constructs a two-dimensional spatial noise distribution map of the current storage environment, uses each terminal location as a sampling point, and calculates the local noise level at the location of the requesting node using an inverse distance weighted interpolation algorithm; identifies the specific frequency band interference intensity at the current location of the requesting node, such as whether there is narrowband interference or harmonic interference generated by nearby high-power electrical equipment, and performs environmental interference compensation on the current RF phase fingerprint based on the interference characteristics: if narrowband noise is detected at a specific frequency point, an adaptive notch filter is used to eliminate the phase jitter at that frequency point; if broadband noise floor elevation is detected, a wavelet threshold denoising algorithm is used to smooth the phase sequence; and compares the noise characteristics of adjacent nodes using a spatial interpolation algorithm to correct the systematic phase deviation caused by local multipath effects, generating a compensated current RF phase fingerprint. This fingerprint eliminates the influence of electromagnetic noise in the nearby area and environmental multipath, and is closer to the real electromagnetic coupling characteristics under ideal interference-free conditions.

[0071] S332. Temporal Correlation Analysis and Evolution Pattern Identification: A temporal correlation analysis is performed between the compensated current RF phase fingerprint and the fingerprint change trend in historical operation records to identify the evolution pattern of the fingerprint over time; specifically: The protocol engine extracts historical radio frequency phase fingerprints recorded by each response node from multi-source collaborative data, and constructs a time series of radio frequency feature evolution for the material according to the timestamp order. This includes the current compensated fingerprint of the requesting node as the latest sample, and the historical fingerprints recorded by each adjacent node as past samples. The engine uses a dynamic time warping algorithm or a long short-term memory network model to perform pattern analysis on the time series, calculates the similarity or distance metric of phase fingerprints between adjacent time points, and identifies whether the fingerprint change trend conforms to a normal slow drift pattern, such as a gradual phase change caused by tag aging or metal surface oxidation, which exhibits small-amplitude continuous fluctuation characteristics, or conforms to abnormal sudden jump characteristics, such as a step phase change caused by tag peeling and re-attaching or material replacement, which exhibits large-amplitude discrete jump characteristics.

[0072] Simultaneously, the consistency between historical fingerprints recorded by each response node is analyzed. If historical fingerprints across nodes show similar evolutionary trends, the credibility of the current compensated fingerprint is enhanced. If there are significant discrepancies, the material may have abnormal cross-regional circulation records, providing temporal correlation weights for subsequent consensus calculations.

[0073] S333. Weighted Consensus Calculation and Index Update: Combining the trend of the center of gravity distribution change, a weighted consensus algorithm is used to cross-validate and evaluate the confidence level of the data returned by each terminal, generating an updated binding integrity index; specifically: The protocol engine synchronously parses the historical centroid spatial distribution data returned by each response node, constructs the temporal trend of centroid distribution, and analyzes whether the direction and magnitude of centroid offset conform to the posture change pattern during normal material handling. The weighted consensus algorithm assigns dynamic weights to each response node: the weight allocation is based on the node's historical credibility score, dynamically adjusted based on records of accurate data provided in past collaborative verification, with initial values ​​of equal weight; response latency, the shorter the latency, the higher the weight, reflecting data freshness; and the similarity between the data and the current compensated fingerprint, the higher the similarity, the higher the weight.

[0074] The weighted average deviation between the historical RF phase fingerprint and the compensated current fingerprint of each response node is calculated. Combined with the weighted consistency assessment of the centroid distribution change trend, a comprehensive deviation value is generated. This comprehensive deviation value is transformed to the 0 to 1 interval through the sigmoid function or piecewise linear mapping, and then reversed by subtracting the normalized deviation value from 1 to obtain the updated binding integrity index. After cross-validation and consensus calculation of multi-source data, the confidence level of this index is significantly higher than that of the original index calculated by a single node based on local data, effectively reducing the risk of misjudgment caused by environmental interference and single-point sensor failure.

[0075] S334. Generation and Issuance of Differentiated Control Instructions: Based on the numerical range of the updated binding integrity index, generate differentiated control instructions. Instructions include green instructions allowing passage, yellow instructions restricting passage and triggering manual review, or red instructions triggering security isolation; specifically: The edge computing unit of the requesting node compares the updated binding integrity index with the preset decision threshold: the high confidence threshold is set to 0.8 and the low confidence threshold is set to 0.5. If the updated index is greater than or equal to 0.8, it indicates that the identity and binding relationship of the materials are highly reliable after multi-terminal collaborative verification, and a green instruction is generated to allow passage. The green instruction includes a standard passage control signal to control the normal start of the material transmission mechanism, a confirmation signal to update the inventory status of the materials in the warehouse management database, and a log recording instruction to report normal inbound and outbound events to the warehouse master station system.

[0076] If the updated index is between 0.5 and 0.8, it indicates a certain degree of uncertainty, such as strong environmental interference or slight inconsistencies in historical data. This generates a yellow instruction to restrict passage and trigger manual review. The yellow instruction includes a path planning signal to control the material transport mechanism to transfer materials to the manual review buffer area, a push signal to send a review alarm notification to the handheld terminal of the on-site management personnel or the central control room, and an instruction to lock the status of the material in the database as pending review.

[0077] If the updated index is below 0.5, or if consistency analysis determines that there are replay attacks, unreasonable cyclical characteristics in the time series of historical fingerprints and current fingerprints, tag cloning, or cross-node fingerprints exhibiting non-physical high consistency but lacking normal evolutionary patterns, then a red instruction to trigger security isolation is generated. The red instruction includes an emergency braking signal to immediately lock the material transport mechanism and stop its operation, a trigger signal to activate the on-site audible and visual alarm device, an emergency notification signal to report the security incident to the security management system, and an instruction to mark all currently collected multimodal data as security evidence and encrypt and store it in the isolation area. After the differentiated control instruction is digitally signed and protected for integrity in the trusted execution environment module, it is sent to the material transport control mechanism.

[0078] S34. Blockchain Structure Operation Log Recording: Differentiated control instructions and verification process data are encrypted and recorded in the local operation log of the blockchain structure; specifically: After the AI-powered intelligent fusion terminal generates and executes differentiated control commands, the trusted execution environment module constructs a new log block. The header of the log block contains the hash value of the previous block, the current timestamp, the block number, and the unique device identifier of the AI-powered intelligent fusion terminal. The payload of the log block contains key data for this inbound / outbound verification, including the electronic tag identifier of the materials, the binding integrity index before and after the update, the type code of the differentiated control command, the list of adjacent terminal node identifiers participating in the collaborative verification, and a summary of the verification process, such as whether multimodal cross-verification was triggered and whether the collaborative verification mode was started.

[0079] The Trusted Execution Environment (TEE) module uses a private key deployed in the hardware security module to digitally sign log blocks, ensuring the non-repudiation of log content. Subsequently, the signed new block is appended to the end of the blockchain structure stored in the local trusted storage area, forming a chain storage. The local operation log of this blockchain structure is independent of the host computer warehouse management system. Even if the network connection is interrupted or the host computer is attacked, the operation records on the edge side still maintain cryptographic immutability and traceability, meeting the compliance requirements of the power industry for operation auditing and liability determination.

[0080] Example 2: Figure 2As shown, an intelligent warehouse management system is used to implement an intelligent warehouse management method, including: an AI intelligent fusion terminal, a multi-terminal collaborative network, a warehouse master station system, and a material transfer control mechanism, specifically: A. At least one AI-powered intelligent fusion terminal, deployed at a warehouse operation node, the AI-powered intelligent fusion terminal including: The RFID read / write module and multi-angle antenna array operate in the ultra-high frequency band of 840~960MHz, support multi-frequency scanning, and are used to collect the radio frequency phase information and backscatter intensity of electronic tags on the surface of materials to construct radio frequency phase fingerprints. A six-axis gravity sensor array, consisting of multiple pressure sensors arranged in a grid, is used to collect data on the static weight of materials and the three-dimensional spatial distribution of the center of gravity. The visual acquisition module includes a high-definition camera and a supplementary lighting unit, used to acquire panoramic images and multi-angle imaging of the material's appearance; The edge trusted computing module has a built-in material fingerprint database and behavior pattern recognition model, which is used to perform multimodal identity anchoring, binding deviation index calculation, binding integrity index generation and trust level decision in steps S1 to S3. The blockchain storage module is used to encrypt and record differentiated control instructions and verification process data in the local operation log of the blockchain structure, ensuring the immutability and traceability of the data; B. Multi-terminal collaborative network, which consists of multiple AI intelligent fusion terminals deployed at different work nodes interconnected through an industrial local area network, is used to execute the distributed consistency protocol in the multi-terminal collaborative verification mode. Each terminal exchanges material historical operation records and radio frequency environmental noise characteristics through a broadcast mechanism and jointly updates the binding integrity index. C. The warehouse master station system, connected to a multi-terminal collaborative network, includes: The data audit module is used to periodically collect blockchain operation logs from various terminals to conduct cross-terminal data auditing and trace abnormal behavior. The model optimization module is used to globally optimize the weight parameters of the radio frequency environment interference model and the distributed consensus protocol based on the radio frequency environment noise characteristics and historical operation records uploaded by each terminal. The personnel access control module is used to dynamically adjust the access control permissions of corresponding personnel based on the cumulative number of violations in the violation database. D. Material transfer control mechanism, connected to AI intelligent fusion terminal control, including intelligent access control, cargo location status indicator lights and AGV scheduling module, used to receive and execute differentiated control commands.

[0081] As can be seen from the above description, the embodiments of the present invention achieve the following technical effects: Existing RFID warehouse management technologies rely solely on the EPC codes of electronic tags for identification, making them highly vulnerable to unauthorized substitution of goods by peeling and reattaching tags. Furthermore, the reflection interference from the metal surface further exacerbates identification errors. This invention proposes the concept of radio frequency phase fingerprinting in step S1. By scanning at multiple frequencies, it extracts the phase information and backscattering intensity of the electromagnetic coupling between the tag and the metal surface, constructing a unique and unclonable frequency domain response curve. This fingerprint, along with the three-dimensional centroid spatial distribution data and panoramic image of the goods, is spatiotemporally aligned to generate multimodal identity anchoring data, forming a digital twin ID card with triple binding of electromagnetic, mechanical, and visual characteristics. Even if the tag is peeled and reattached, its near-field coupling characteristics with the metal surface will inevitably change, making the radio frequency phase fingerprint unmatchable. This physically eliminates tag replacement and goods substitution, providing fundamental protection for the storage security of high-value materials in the power industry.

[0082] Existing technologies for judging the status of materials are limited to binary logic such as the presence or absence of tags and whether the weight matches, failing to perceive changes in the integrity of the physical binding relationship between the tags and the materials themselves, and even less able to identify abnormal behavior patterns during operations. This invention proposes a binding deviation index as a quantitative indicator in step S2. By aligning the current radio frequency phase fingerprint with the anchor fingerprint in real time and calculating the phase difference, tag integrity is transformed into a measurable mathematical parameter. Based on this, step S23 establishes an anomaly-driven multimodal cross-validation mechanism—when the center of gravity distribution is abnormal or the binding deviation index exceeds the limit, a visual review is automatically triggered, and the center of gravity deviation, phase deviation, and visual similarity are weighted and fused to generate a binding integrity index that comprehensively characterizes the three-dimensional credibility of the materials. This mechanism achieves a leap from static identity recognition to dynamic integrity verification, capable of keenly capturing subtle anomalies such as minute tag displacement, material deformation, multiple items being picked up and placed at once, and resetting after picking up and placing, elevating the level of precision in warehouse management to a whole new dimension.

[0083] Existing warehousing systems all operate on a single-point decision-making model. Individual terminals are highly susceptible to misjudgments due to environmental electromagnetic noise, sensor malfunctions, and other factors. Furthermore, centralized storage of operation logs poses risks of data tampering and audit blind spots. This invention creates a multi-terminal collaborative verification model in step S3. When the trust level is within an ambiguous range, a collaborative request is broadcast to adjacent terminals to obtain historical operation records and radio frequency environmental noise characteristics. Cross-validation and confidence assessment are performed through a distributed consensus protocol, upgrading single-point decision-making to networked collaborative decision-making, significantly improving the system's robustness and anti-interference capabilities. Simultaneously, differentiated control commands and verification process data are encrypted and recorded in a local operation log on a blockchain structure. Leveraging the immutability and traceability of the blockchain, every verification and every decision is verifiable. The warehousing master station system can periodically perform cross-terminal data audits, achieving a qualitative leap from single-point trust to global trust, and from process agnosticity to full-process traceability, fully meeting the stringent requirements for data traceability and non-repudiation in high-security fields such as the power industry.

[0084] The embodiments and / or implementation methods described above are merely preferred embodiments and / or implementation methods for implementing the technology of the present invention, and are not intended to limit the implementation methods of the technology of the present invention in any way. Any person skilled in the art may make some modifications or alterations to other equivalent embodiments without departing from the scope of the technical means disclosed in the present invention, but these should still be regarded as the technology or embodiments that are substantially the same as the present invention.

[0085] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are only preferred embodiments of this application. It should be noted that due to the limitations of written expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. An intelligent warehouse management method, characterized in that, include: The AI-powered intelligent fusion terminal collects radio frequency phase fingerprints, center of gravity spatial distribution data, and panoramic images of the materials upon their first entry into the warehouse, generating and storing multimodal identity anchoring data. During the operation phase, the current radio frequency phase fingerprint, current center of gravity spatial distribution data and current appearance image of the materials are collected in real time, and compared with the multimodal identity anchoring data to calculate the binding deviation index, trigger multimodal cross-validation, and generate the binding integrity index. The trust level is determined based on the binding integrity index. If it is lower than the preset threshold, multi-terminal collaborative verification is initiated to update the binding integrity index. Differential control instructions are generated and recorded based on the updated index. Multimodal identity anchoring is performed by an AI intelligent fusion terminal deployed on the warehouse site. The AI ​​intelligent fusion terminal integrates an RFID reading and writing module, a visual acquisition module, and a gravity sensing module. During the initial material entry and calibration phase, the RFID reading and writing module is controlled to scan the electronic tags attached to the surface of the target material at multiple preset frequency points, extract the signal phase information and backscatter intensity at each frequency point, and construct the radio frequency phase fingerprint. The gravity sensing module is synchronously controlled to collect static weight and center of gravity spatial distribution data of the materials, and the vision acquisition module is controlled to collect panoramic images of the materials' appearance. The radio frequency phase fingerprint, the spatial distribution data of the center of gravity, and the panoramic image of the appearance are spatiotemporally aligned to generate multimodal identity anchoring data of the material and stored in a local trusted storage area. The RFID reader / writer module and its connected multi-angle antenna array are controlled to perform continuous frequency sweeping in preset frequency steps. At each frequency sweep point, the electronic tag attached to the surface of the target material is excited to backscatter, and the phase angle and signal strength of the backscatter signal corresponding to the frequency point are recorded. After traversing all preset frequency points, the acquired phase angle sequence and signal strength sequence are combined to form a frequency domain response curve characterizing the near-field coupling effect between the tag antenna and the surface of the metal material. The frequency domain response curve is normalized and feature points are extracted to generate a unique radio frequency phase fingerprint.

2. The intelligent warehouse management method according to claim 1, characterized in that, During the operational phase, the current radio frequency phase fingerprint, current center of gravity spatial distribution data, and current appearance image of the materials are collected in real time. These are compared with the multimodal identity anchoring data to calculate the binding deviation index, trigger multimodal cross-validation, and generate a binding integrity index, specifically including: During the material entry and exit operations, the AI ​​intelligent fusion terminal acquires the current radio frequency phase fingerprint, current center of gravity spatial distribution data and current appearance image of the electronic tag in real time; The current radio frequency phase fingerprint is compared with the radio frequency phase fingerprint to calculate the binding deviation index, which reflects the integrity of the physical binding relationship between the tag and the material. The current centroid spatial distribution data is matched and analyzed with the centroid spatial distribution data. If there is an anomaly, a multimodal cross-validation process is triggered. Based on the current appearance image, visual features are reviewed, and a binding integrity index is generated.

3. The intelligent warehouse management method according to claim 2, characterized in that, The current radio frequency phase fingerprint is compared with the radio frequency phase fingerprint to calculate the binding deviation index, which reflects the integrity of the physical binding relationship between the tag and the material. Specifically, this includes: The current radio frequency phase fingerprint acquired in real time is aligned with the radio frequency phase fingerprint at the same frequency point to ensure that the two sets of data are compared at the same frequency point. Calculate the phase difference at each corresponding frequency point, and calculate the root mean square error of the phase difference at all frequency points to obtain the phase deviation value; simultaneously calculate the rate of change of signal strength at each corresponding frequency point to obtain the signal strength fluctuation value. The phase deviation value and the signal strength fluctuation value are weighted and fused to generate the binding deviation index.

4. The intelligent warehouse management method according to claim 3, characterized in that, The current centroid spatial distribution data is matched and analyzed with the centroid spatial distribution data. If an anomaly is detected, a multimodal cross-validation process is triggered. Based on the current appearance image, visual feature verification is performed, and a binding integrity index is generated comprehensively, specifically including: The current center of gravity spatial distribution data is matched and analyzed with the center of gravity spatial distribution data to calculate the offset of the center of gravity projection position and the change in vertical height, and to generate the center of gravity deviation index. Determine whether the center of gravity deviation index exceeds a preset center of gravity anomaly threshold or the binding deviation index exceeds a preset phase anomaly threshold. If either condition is met, the multimodal cross-validation process is triggered. In the cross-validation process, the visual acquisition module is controlled to perform multi-angle imaging of the material, extract surface feature points in the current appearance image, and perform feature matching with the stored panoramic appearance image to calculate the visual similarity score. The center of gravity deviation index, binding deviation index, and visual similarity score are weighted and fused to generate the binding integrity index.

5. The intelligent warehouse management method according to claim 4, characterized in that, The trust level is determined based on the binding integrity index. If it is lower than a preset threshold, multi-terminal collaborative verification is initiated to update the binding integrity index. Differential control instructions are generated and recorded based on the updated index, specifically including: The trust level of inbound and outbound operations is determined based on the binding integrity index. When the trust level is higher than a preset threshold, a standard inbound and outbound control instruction is generated. When the trust level is lower than the preset threshold but higher than the minimum admission threshold, the multi-terminal collaborative verification mode is activated, and a collaborative verification request is broadcast to other AI intelligent fusion terminals deployed in the adjacent area to obtain the historical operation records of materials and the radio frequency environmental noise characteristics of the nearby area. The binding integrity index is updated through a distributed consensus protocol, and differentiated control instructions are generated based on the updated binding integrity index. The differentiated control instructions include allowing passage, restricting passage, or triggering security isolation. The differentiated control instructions and verification process data are encrypted and recorded in the local operation log of the blockchain structure.

6. The intelligent warehouse management method according to claim 5, characterized in that, When the trust level is lower than a preset threshold but higher than a minimum admission threshold, a multi-terminal collaborative verification mode is activated, specifically including: When the trust level is lower than a preset threshold but higher than the minimum admission threshold, the current AI intelligent fusion terminal will automatically start the multi-terminal collaborative verification mode. The system sends a collaborative verification request to other adjacent AI intelligent fusion terminals via a broadcast protocol. The request includes the material identification and the current collection timestamp. Receive response data returned by each adjacent terminal. The response data includes the trend of radio frequency phase fingerprint change of materials in the historical operation records of each terminal, the trend of center of gravity distribution change, and the real-time radio frequency environmental noise characteristics of the location of each terminal. The data returned by each adjacent terminal is timestamped and validated. Timed-out or abnormal response data is removed. The filtered multi-source collaborative data is packaged and prepared for input into the distributed consensus protocol for updating the binding integrity index.

7. The intelligent warehouse management method according to claim 6, characterized in that, The binding integrity index is updated through a distributed consensus protocol, and differentiated control instructions are generated based on the updated binding integrity index. These differentiated control instructions include allowing passage, restricting passage, or triggering security isolation, specifically including: The filtered multi-source collaborative data is input into the distributed consensus protocol engine, and environmental interference compensation is performed on the radio frequency phase fingerprint of the current terminal based on the radio frequency environmental noise characteristics. The current RF phase fingerprint after compensation is compared with the fingerprint change trend in historical operation records through time-series correlation analysis to identify the evolution pattern of fingerprint in the time dimension. Based on the trend of the center of gravity distribution, a weighted consensus algorithm is used to cross-validate and evaluate the confidence of the data returned by each terminal, and an updated binding integrity index is generated. Based on the numerical range of the updated binding integrity index, differentiated control instructions are generated. These instructions include green instructions that allow passage, yellow instructions that restrict passage and trigger manual review, or red instructions that trigger security isolation.

8. An intelligent warehouse management system, characterized in that, A method for implementing an intelligent warehouse management system as described in any one of claims 1-7 includes: an AI intelligent fusion terminal, a multi-terminal collaborative network, a warehouse master station system, and a material transfer control mechanism, specifically: At least one AI-powered intelligent fusion terminal is deployed at a warehouse operation node, the AI-powered intelligent fusion terminal comprising: The RFID read / write module and multi-angle antenna array support multi-frequency scanning and are used to collect the radio frequency phase information and backscatter intensity of electronic tags on the surface of materials to construct radio frequency phase fingerprints. A six-axis gravity sensor array, consisting of multiple pressure sensors arranged in a grid, is used to collect data on the static weight of materials and the three-dimensional spatial distribution of the center of gravity. The visual acquisition module, including a high-definition camera and a supplementary lighting unit, is used to acquire panoramic images of the material's appearance and perform multi-angle imaging. The edge trusted computing module has a built-in material fingerprint database and behavior pattern recognition model, which is used to perform multimodal identity anchoring, binding deviation index calculation, binding integrity index generation and trust level decision-making. The blockchain storage module is used to encrypt and record differentiated control instructions and verification process data in the local operation log of the blockchain structure, ensuring the immutability and traceability of the data; The multi-terminal collaborative network consists of multiple AI intelligent fusion terminals deployed at different work nodes interconnected through an industrial local area network. It is used to execute the distributed consistency protocol in the multi-terminal collaborative verification mode. Each terminal exchanges material historical operation records and radio frequency environmental noise characteristics through a broadcast mechanism and jointly updates the binding integrity index. The warehouse master station system, which is connected to the multi-terminal collaborative network, includes: The data audit module is used to periodically collect blockchain operation logs from various terminals to conduct cross-terminal data auditing and trace abnormal behavior. The model optimization module is used to globally optimize the weight parameters of the radio frequency environment interference model and the distributed consensus protocol based on the radio frequency environment noise characteristics and historical operation records uploaded by each terminal. The personnel access control module is used to dynamically adjust the access control permissions of corresponding personnel based on the cumulative number of violations in the violation database. The material transfer control mechanism, connected to the AI ​​intelligent fusion terminal, includes intelligent access control, cargo location status indicator lights, and an AGV scheduling module, used to receive and execute the differentiated control commands.

Images