Intelligent clothing inventory real-time monitoring and automatic checking system based on RFID

Abstract

The invention relates to the field of intelligent warehousing, and discloses an intelligent clothing inventory real-time monitoring and automatic checking system based on RFID, which aims at solving the problems of poor environmental adaptability, high dynamic sensing delay, high cross-scene deployment cost and the like in the prior art, and comprises a reconfigurable RFID tag, an integrated spiral-dipole composite antenna and a temperature compensation circuit, dual-band switching of 860 to 960 MHz and self calibration are supported; a heterogeneous read-write network is formed by a fixed reader-writer and AGV mobile nodes, and signal coverage is enhanced through beam forming; the edge calculation module adopts an LSTM trajectory prediction and hybrid filtering algorithm to realize millisecond-level processing; the cloud platform maps the inventory in real time based on the digital twin model and traces the source of the block chain, the dynamic time slot distribution protocol optimizes the cooperation of the reader-writer, and the transfer learning model realizes cross-scene rapid adaptation. According to the invention, the tag reading rate is 99.7%, the inventory efficiency is improved by 300%, the deployment period is shortened by 80%, and the problems of accurate inventory management and dynamic tracking in a complex environment are solved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent warehousing, and more particularly, to an RFID-based intelligent clothing inventory real-time monitoring and automatic inventory counting system. Background Art

[0002] As the apparel retail industry transitions to fast fashion, inventory management faces increasingly severe challenges. Traditional manual inventory counting methods suffer from low efficiency (a single full-warehouse inventory count takes an average of 48 hours) and high error rates (the industry average inventory discrepancy rate is 3.5%). Although existing technologies have improved upon this with barcode scanning or basic RFID systems, the following technical bottlenecks still exist in practical applications:

[0003] 1. Environmental adaptability deficiencies: The electromagnetic shielding effect created by the dense metal shelves in clothing warehouses causes the read rate of conventional UHF RFID tags to drop to 72%-85%. Tag impedance mismatch caused by temperature and humidity fluctuations increases the tag miss rate by 37% in warehouses during the summer.

[0004] 2. Insufficient dynamic sensing capabilities: Existing systems often use fixed readers, which cannot effectively capture inventory location changes caused by dynamic operations such as AGV handling and manual picking. MIT Auto-ID laboratory tests show that the location update delay for moving goods generally exceeds 15 minutes.

[0005] 3. Data processing efficiency limitations: When faced with the demand for processing over 10,000 tag data points per second, centralized cloud processing architectures experience exponentially increased system response times. Data from a Walmart pilot project showed that when 200 readers were activated simultaneously, the traditional system experienced a data packet loss rate of up to 22%.

[0006] 4. High cross-scenario migration costs: Existing solutions rely on specific warehouse environment parameter configurations. System reconstruction for different warehouse types requires redeploying hardware and takes 2-3 weeks for algorithm tuning.

[0007] Therefore, to address the above problems, an RFID-based smart clothing inventory real-time monitoring and automatic inventory system is proposed. Summary of the Invention

[0008] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an RFID-based smart clothing inventory real-time monitoring and automatic inventory system to solve the problems raised in the above-mentioned background technology.

[0009] To achieve the above objectives, the present invention provides the following technical solution: an RFID-based intelligent clothing inventory real-time monitoring and automatic inventory system, comprising:

[0010] The reconfigurable RFID tag set is encapsulated in the clothing lining through a flexible substrate, integrates a spiral-dipole composite antenna and temperature compensation circuit, and supports 860-960MHz dual-band adaptive switching;

[0011] A heterogeneous reader / writer network consists of a three-dimensional scanning array consisting of a primary reader, fixed auxiliary readers, and mobile readers. The primary reader is deployed on the warehouse's top truss, the fixed auxiliary readers are embedded in the shelf columns, and the mobile readers are mounted on AGV robots.

[0012] The edge computing layer, an FPGA processor deployed near the reader, runs a multi-source data fusion algorithm, including an LSTM motion prediction unit and a hybrid filtering denoising module;

[0013] The cloud management platform communicates with the edge layer via the OPC-UA protocol to build a digital twin warehouse model and implement blockchain traceability;

[0014] Collaborative working mechanism: the main reader performs global time slot scheduling, the mobile reader performs blind scanning along dynamic paths, and the edge layer pre-processes the data and uploads it to the cloud to generate a real-time inventory map.

[0015] Preferably, the flexible antenna of the reconfigurable RFID tag adopts a silver nanowire grid structure prepared by a photolithography process, and switches the working frequency band through a PIN diode. Its self-calibration function is specifically manifested as follows: the temperature sensor monitors environmental changes in real time, the microcontroller adjusts the antenna impedance matching network according to the temperature gradient, and automatically triggers the retuning process when the signal strength fluctuation exceeds ±3dB.

[0016] Preferably, the deployment method of the heterogeneous reader-writer network includes: spatial topology construction, calculating the optimal reader-writer layout based on the warehouse CAD model, and the spacing between the main readers is 1.2-1.5 times the height of the shelf; signal coverage optimization, using beamforming technology to directionally enhance the field strength in the deep area of ​​the shelf to compensate for the signal attenuation caused by the metal shelf; mobile node collaboration, the reader-writer carried by the AGV establishes time-division duplex communication with the fixed reader-writer during inspection to avoid channel conflicts.

[0017] Preferably, the hybrid filtering and denoising module of the edge computing layer includes: first, a primary filtering unit, which uses an improved Kalman filter to eliminate Gaussian noise in the RFID signal; second, a secondary processing unit, which applies a particle filtering algorithm to solve the signal distortion caused by the multipath effect; finally, a dynamic parameter adjustment mechanism, which automatically adjusts the filter window width according to the RSSI fluctuation rate, and enables the alternative filter coefficient library when the environmental interference level is greater than level 4.

[0018] Preferably, the digital twin warehouse model construction process of the cloud management platform includes a physical mapping layer, a real-time data layer and a business logic layer, wherein the physical mapping layer imports the warehouse three-dimensional structure data through the BIM interface; the real-time data layer fuses RFID coordinate information with UWB positioning data to generate a clothing location heat map; the business logic layer integrates the inventory forecasting algorithm and the replenishment strategy optimization engine, and triggers an intelligent replenishment work order when the inventory of a certain SKU is lower than the safety threshold.

[0019] Preferably, the navigation system of the AGV robot adopts: multi-sensor fusion positioning, integrating UWB anchor point signals, IMU inertial data and visual odometry, with a positioning accuracy of <5cm; adaptive path planning, generating a global path based on the Dijkstra algorithm, combined with the dynamic window method for real-time obstacle avoidance; task scheduling strategy, dynamically adjusting the inventory priority according to the inventory difference rate, and automatically increasing the scanning frequency in high-deviation areas.

[0020] Preferably, the global time slot scheduling method specifically includes three stages. The first stage is the conflict detection stage, in which each reader broadcasts the channel occupancy status and constructs a conflict probability matrix; the second stage is the time slot allocation stage, in which an improved Q-learning algorithm is used to calculate the optimal time slot combination, and a reward function comprehensively evaluates channel utilization and energy consumption indicators; the third stage is the dynamic adjustment stage, in which when a new reader is connected or the tag density changes by more than 15%, distributed negotiation is triggered to reallocate time slot resources.

[0021] Preferably, the workflow of the blockchain traceability module is as follows: data is uploaded to the chain, the clothing circulation information collected by RFID is generated into a Merkle tree structure and stored in Hyperledger Fabric; smart contracts, preset quality inspection rules automatically verify the integrity of the clothing information entering the warehouse, abnormal data triggers an alarm and freezes related transactions; cross-chain interaction, realizes data atomic exchange with the blockchain system of suppliers and logistics parties through the cross-chain gateway.

[0022] Preferably, the execution logic of the dynamic parameter adjustment mechanism is: first, establish an environmental interference level assessment model, and divide the interference intensity into 1-5 levels according to the background noise power spectrum density; when the interference level increases, gradually increase the process noise covariance matrix Q value of the Kalman filter; simultaneously reduce the particle filter resampling threshold, retaining more large-weight particles to maintain tracking stability.

[0023] Preferably, the intelligent replenishment work order generation method includes: demand forecasting, using an LSTM network to analyze historical sales data and seasonal trend factors; supplier evaluation, generating supplier scores based on the on-time delivery rate and quality qualification rate recorded in the blockchain; automatic order dispatching, matching replenishment demand with the inventory data of preferred suppliers, and generating an electronic purchase order with an attached smart contract.

[0024] The technical effects and advantages of the present invention are as follows:

[0025] 1. Ultra-high reliability perception in complex environments

[0026] Through the collaborative design of dual-band reconfigurable RFID tags and three-dimensional heterogeneous read-write networks, the problems of electromagnetic shielding and temperature and humidity interference on metal shelves are overcome: the spiral-dipole composite antenna increases the tag's reading distance in a metal environment by 2.3 times, the temperature compensation circuit reduces the missed read rate caused by impedance mismatch from the industry average of 9.7% to 0.8%, and beamforming technology compensates for signal attenuation deep in the shelves, optimizing blind spot coverage from 58% to 92%.

[0027] 2. Breakthrough in real-time dynamic inventory tracking capabilities

[0028] The innovative integration of mobile read-write nodes and intelligent path planning algorithms enables real-time monitoring of the entire process: the rotating read-write array equipped with the AGV can complete multi-angle scanning while moving, reducing the position update delay from 15 minutes to 8 seconds. The integrated positioning system based on UWB / IMU / vision has a dynamic cargo tracking accuracy of ±3cm. The digital twin platform processes 12,000 tag data updates per second, and the inventory status refresh frequency is increased to sub-second level.

[0029] 3. Efficient processing architecture for massive data

[0030] The edge-cloud collaborative computing framework is used to reconstruct the data processing process and break through the bottleneck of traditional systems: FPGA edge nodes implement data preprocessing, compressing the uploaded data volume by 83%. The hybrid filtering algorithm increases the efficiency of effective signal extraction by 4.8 times, and the data packet loss rate drops sharply from 22% to 0.3%. The distributed computing architecture supports concurrent operation of 200+ readers and writers, and the system response time is stable within 85ms.

[0031] 4. Intelligent adaptation for rapid deployment across scenarios

[0032] Flexible configuration is achieved based on the transfer learning model and parameter self-optimization system: the domain adversarial network (DANN) reduces the algorithm adaptation time for new warehouse scenarios from 3 weeks to 18 hours. The dynamic parameter adjustment mechanism automatically matches the electromagnetic environment characteristics of different warehouse types, reducing deployment costs by 86% and increasing the system function reuse rate to 95%, supporting mixed warehousing management of multiple categories such as clothing, shoes and hats. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a system framework diagram of the present invention. DETAILED DESCRIPTION

[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0035] As attached Figure 1 As shown,

[0036] (1) An RFID-based intelligent clothing inventory real-time monitoring and automatic inventory system, characterized by comprising:

[0037] The reconfigurable RFID tag set is encapsulated in the clothing lining through a flexible substrate, integrates a spiral-dipole composite antenna and temperature compensation circuit, and supports 860-960MHz dual-band adaptive switching;

[0038] A heterogeneous reader / writer network consists of a three-dimensional scanning array consisting of a primary reader, fixed auxiliary readers, and mobile readers. The primary reader is deployed on the warehouse's top truss, the fixed auxiliary readers are embedded in the shelf columns, and the mobile readers are mounted on AGV robots.

[0039] The edge computing layer, an FPGA processor deployed near the reader, runs a multi-source data fusion algorithm, including an LSTM motion prediction unit and a hybrid filtering denoising module;

[0040] The cloud management platform communicates with the edge layer via the OPC-UA protocol to build a digital twin warehouse model and implement blockchain traceability;

[0041] The system uses a collaborative working mechanism: the master reader performs global time slot scheduling, mobile readers perform blind spot scanning along dynamic paths, and the edge layer preprocesses data before uploading it to the cloud to generate a real-time inventory map. During system deployment, a reconfigurable RFID tag (25×15×0.3mm) is sewn onto the inside of the clothing care label. The tag contains a built-in silver nanowire flexible antenna (50μm line width) and an SX1276 chip. An Impinj Speedway R420 reader (transmitting power 30dBm) is installed every 8 meters on the warehouse's top steel beams, and an Alien ALR-9680 reader (tilt angle 20°) is embedded 1.5 meters from the ground on each shelf column. The AGV (model MiR500) is equipped with a rotatable RFID array (rotating at 30rpm) on top. The edge computing node uses the NVIDIA Jetson AGX Xavier and runs a TensorRT-accelerated LSTM model (with a time window length of 30 seconds). The cloud platform builds a digital twin model based on Unity3D and integrates with the SAP ERP system. After the system is started, the main reader allocates time slots through the TDMA protocol, the mobile reader performs blind scanning according to the path planned by the A* algorithm, and the edge node uploads the processed data to the cloud via the MQTT protocol.

[0042] (2) The flexible antenna of the reconfigurable RFID tag adopts a silver nanowire grid structure prepared by a photolithography process, and switches the working frequency band through a PIN diode. Its self-calibration function is specifically manifested as follows: the temperature sensor monitors environmental changes in real time, and the microcontroller adjusts the antenna impedance matching network according to the temperature gradient. When the signal strength fluctuation exceeds ±3dB, the retuning process is automatically triggered. Among them, the flexible antenna is prepared by a silver nanowire grid (line spacing 200μm) on a PI substrate through a photolithography process, and the antenna feed point is connected to a BAP-64003 chip; the frequency band switching is performed by SKYWORKS SKY13405-485LF PIN diode control automatically switches to 960MHz when it detects signal attenuation in the 860MHz band greater than 10dB. During the self-calibration process, a DS18B20 temperature sensor collects environmental data every 5 seconds. The STM32F407 microcontroller adjusts the impedance matching network parameters via the ADG902 RF switch when the temperature change rate ΔT / Δt exceeds 0.5°C / min. If the RSSI remains less than -70dBm after three adjustments, the antenna resonant frequency is retuned.

[0043] (3) The deployment method of the heterogeneous reader-writer network includes: spatial topology construction, calculating the optimal reader-writer layout based on the warehouse CAD model, and the main reader-writer spacing is 1.2-1.5 times the shelf height; signal coverage optimization, using beamforming technology to directionally enhance the field strength in the deep area of ​​the shelf to compensate for the signal attenuation caused by the metal shelf; mobile node collaboration, the AGV-mounted reader-writer establishes time division duplex communication with the fixed reader-writer during inspection to avoid channel conflict, wherein the main reader-writer installation height H = shelf height × 1.3 (H = 6.5 meters when the shelf height is 5 meters), and the auxiliary reader-writer horizontal spacing is 4 meters; beamforming uses PhasedArray Toolbox to generate the directional pattern, and calculates the optimal weight vector in MATLAB to make the field strength of the third layer of the shelf (height 2.1 meters) ≥ -65dBm; AGV mobile nodes and fixed readers collaborate through time division duplex (TDD), the fixed reader-writer works in time slots 1-10, and the AGV scans the blind area in time slots 11-20, and the time slot allocation period is 200ms.

[0044] (4) The hybrid filtering denoising module of the edge computing layer includes: first, a primary filtering unit, which uses an improved Kalman filter to eliminate Gaussian noise in the RFID signal; second, a secondary processing unit, which applies a particle filter algorithm to solve the signal distortion caused by the multipath effect; finally, a dynamic parameter adjustment mechanism, which automatically adjusts the filter window width according to the RSSI fluctuation rate, and enables the alternative filter coefficient library when the environmental interference level is greater than level 4, wherein the edge node runs the improved Kalman filter: the process noise covariance matrix Q = diag([0.01,0.01]), the observation noise R = 0.1; the particle filter sets 2000 particles, and the resampling threshold is 0.6; the dynamic parameter adjustment module monitors the environmental noise power spectrum density, and when the interference power within the 50MHz bandwidth is greater than -80dBm, it is determined to be level 4 interference, and switches to the alternative filter coefficient library (Q = diag([0.05,0.05]), R = 0.3. The input dimension of the LSTM model is [30,6] (including time, coordinates, speed and other parameters), and outputs the position prediction for the next 10 seconds.

[0045] (5) The digital twin warehouse model construction process of the cloud management platform includes a physical mapping layer, a real-time data layer and a business logic layer, wherein the physical mapping layer imports the warehouse three-dimensional structure data through the BIM interface; the real-time data layer fuses the RFID coordinate information with the UWB positioning data to generate a clothing location heat map; the business logic layer integrates the inventory prediction algorithm and the replenishment strategy optimization engine, and triggers an intelligent replenishment work order when the inventory of a certain SKU is lower than the safety threshold. The digital twin model imports the BIM model through the Revit API, receives the RFID coordinates (x, y, z) and UWB positioning data every 500ms, and uses the Marching Cubes algorithm to generate an inventory heat map; the blockchain node is deployed on Hyperledger Fabric2.4, each block contains 100 transactions, and the smart contract verifies the consistency of the EPC code of the incoming clothing with the purchase order; the inventory prediction module uses the Prophet algorithm, and the input parameters include the sales data of the past 30 days, the weather index and the promotion plan. When the prediction error rate is greater than 15%, the model retraining is triggered.

[0046] (6) The navigation system of the AGV robot adopts: multi-sensor fusion positioning, integrating UWB anchor point signals, IMU inertial data and visual odometry, with a positioning accuracy of less than 5cm; adaptive path planning, generating a global path based on the Dijkstra algorithm, and combining the dynamic window method to avoid obstacles in real time; task scheduling strategy, dynamically adjusting the inventory priority according to the inventory difference rate, and automatically increasing the scanning frequency in high-deviation areas. Among them, the AGV navigation system integrates UWB positioning (4 Anchor nodes deployed at the four corners of the warehouse), BMI160IMU (sampling rate 100Hz) and Realsense D435 visual odometry, and realizes positioning through extended Kalman filtering (EKF), with a covariance matrix update period of 50ms; path planning adopts an improved Dijkstra algorithm, with a cost function f = 0.6×path length + 0.3×time + 0.1×energy consumption; when the system detects that the inventory difference rate in a certain area is greater than 5%, it automatically increases the scanning priority of the area to Level 1, and the scanning interval is shortened from 10 minutes to 2 minutes.

[0047] (7) The global time slot scheduling method specifically includes three stages. The first stage is the conflict detection stage. Each reader broadcasts the channel occupancy status and constructs a conflict probability matrix. The second stage is the time slot allocation stage. The improved Q-learning algorithm is used to calculate the optimal time slot combination. The reward function comprehensively evaluates the channel utilization and energy consumption indicators. The third stage is the dynamic adjustment stage. When a new reader is connected or the tag density changes by more than 15%, the distributed negotiation is triggered to reallocate the time slot resources. The dynamic time slot allocation execution process is as follows: ① The reader uses CSMA / C Protocol A broadcasts channel state information (including operating frequency and transmission power); ② Constructs a collision probability matrix P_ij = 1-exp(-λτ), where λ is the tag arrival rate and τ is the time slot length; ③ The Q-learning state space dimension is [channel utilization, tag density], the action space is the time slot allocation scheme, and the reward function R = 0.5×throughput + 0.3×energy saving factor - 0.2×delay; ④ When a new reader is added or the tag density changes by more than 15%, each node negotiates time slot reorganization through the Paxos protocol, with a convergence time of less than 500ms.

[0048] (8) The workflow of the blockchain traceability module is as follows: data is uploaded to the chain, and the clothing circulation information collected by RFID is generated into a Merkle tree structure and stored in Hyperledger Fabric; smart contracts, preset quality inspection rules automatically verify the integrity of clothing information in the warehouse, abnormal data triggers alarms and freezes related transactions; cross-chain interaction, and atomic data exchange with suppliers and logistics parties’ blockchain systems is achieved through cross-chain gateways. Among them, the pre-training stage of the transfer learning model: a virtual warehouse is built in Omnet++, and parameters such as the proportion of metal shelves is set to 30%-70%, and the temperature and humidity fluctuation range is ±15% are set to generate 100,000 sets of training data; the domain adversarial network includes a feature extractor (ResNet-18), a classifier (3-layer fully connected) and a domain discriminator (gradient reversal layer), and the loss function L = 0.7L_class + 0.3L_domain; in actual deployment, 100 sets of RSSI samples are collected from the new warehouse, and the model adaptation is completed within 50 iterations through MAML meta-learning.

[0049] (9) The execution logic of the dynamic parameter adjustment mechanism is as follows: first, an environmental interference level assessment model is established, and the interference intensity is divided into levels 1-5 according to the background noise power spectrum density; when the interference level increases, the process noise covariance matrix Q value of the Kalman filter is gradually increased; the particle filter resampling threshold is simultaneously reduced to retain more large-weight particles to maintain tracking stability. Among them, the millimeter wave radar (TI AWR1843) is deployed at the warehouse entrance and outputs the personnel coordinate trajectory every 100ms; the multispectral camera (Ximea MQ022CG) collects clothing images at a rate of 5fps and detects stains through ResNet-50 (accuracy of 98.7%); the temperature and humidity sensor (Sensirion SCD40) uploads data every 2 minutes. When the humidity is greater than 70% for 1 hour, the tag operating frequency offset Δf = 0.02×(RH-70) MHz is automatically adjusted.

[0050] (10) The intelligent replenishment work order generation method includes: demand forecasting, using LSTM network to analyze historical sales data and seasonal trend factors; supplier evaluation, generating supplier scores based on the supply punctuality rate and quality qualification rate recorded in the blockchain; automatic order dispatching, matching replenishment demand with preferred supplier inventory data, and generating electronic purchase orders with smart contracts, wherein the multi-level alarm triggering logic is as follows: ① Level 1 alarm: when the actual inventory of a certain SKU deviates from the system record by more than 2%, the AGV performs a local re-inventory within 5 minutes; ② Level 2 alarm: when the tag fails to respond to the read and write instructions for 3 consecutive times (with an interval of 30 seconds), the UWB positioning is activated and a 50cm radius warning zone is marked; ③ Level 3 alarm: when the difference between the RFID position and the visual detection coordinate is more than 1 meter and lasts for 5 minutes, the system freezes the inventory record and issues an audible and visual alarm (decibel value 85dB@1m), and resumes after manual confirmation.

[0051] Example 1: Specific implementation process of the present invention

[0052] Step 1: System deployment and initialization

[0053] (1) Tag installation and activation

[0054] (a) Encapsulate the reconfigurable RFID tag into the lining of the garment, and embed a unique ID tag on the collar of each garment.

[0055] (b) Use a handheld programmer (model: IMPINJ R420) to write clothing attribute data (SKU, size, color, etc.),

[0056] (c) Activate the tag self-calibration function and set the temperature compensation trigger threshold to ±5°C ambient temperature change.

[0057] (2) Read-write network deployment

[0058] (a) Install the main reader (working power 30dBm, coverage radius 15m) on the top truss of the warehouse, and set the spacing at a ratio of 1.3 times the shelf height.

[0059] (b) Each column of the shelf is embedded with a fixed auxiliary reader (working power 27dBm, tilt angle 15° downward),

[0060] (c) A rotatable reader array (including 4 sets of cross-polarized antennas, with an adjustable rotation speed of 0-60 rpm) is installed on the top of the AGV robot.

[0061] (3) Infrastructure networking

[0062] (a) Connect all fixed readers via an industrial-grade POE switch (Model: Moxa PT-7728).

[0063] (b) Deploy 5G CPE equipment (Huawei MH5000-82) to achieve wireless backhaul of AGV mobile nodes,

[0064] (c) Install UWB positioning base stations (Decawave TREK1000, accuracy ±10cm) at the four corners of the warehouse. Step 2: System calibration and modeling

[0065] (1) Environmental parameter collection

[0066] (a) Start a distributed temperature and humidity sensor network (one node per 50 m2) to continuously collect 24-hour environmental data.

[0067] (b) Use a spectrum analyzer (Keysight N9020B) to scan the warehouse electromagnetic environment and identify 2.4GHz / 5.8GHz interference sources.

[0068] (2) Three-dimensional signal field strength modeling

[0069] (a) The AGV carries the test tag and patrols along the preset path (speed 0.5m / s), recording the RSSI value at each point.

[0070] (b) Use Kriging interpolation to generate a three-dimensional field strength distribution map and identify signal blind spots (RSSI < -75dBm is defined as a blind spot).

[0071] (c) Adjust the beamforming parameters to ensure that the RSSI in the deep shelf area is ≥ -65dBm.

[0072] (3) Algorithm model pre-training

[0073] (a) Build a virtual warehouse model on the simulation platform (MATLAB / Simulink) and inject 100,000 sets of noise data.

[0074] (b) Pre-trained LSTM trajectory prediction model (hidden layer 128 nodes, time step 30s),

[0075] (c) Establish a basic transfer learning model (ResNet-50 backbone network, weighted cross entropy loss function), step three: dynamic operation and optimization

[0076] (1) Real-time data collection process

[0077] (a) The master reader broadcasts a synchronization signal every 200ms and starts the dynamic time slot allocation protocol.

[0078] (b) The fixed reader performs directional scanning (beam width 60°, cycle 5s / time),

[0079] (c) The AGV mobile node patrols along the optimal path (speed 1.2m / s) and activates the multi-angle scanning mode when it stops in the blind spot.

[0080] (2) Edge computing data processing

[0081] The FPGA processor (Xilinx Zynq UltraScale+) performs three stages of filtering:

[0082] (a) First-level filtering: Improved Kalman filter (process noise Q = 0.01, observation noise R = 0.1),

[0083] (b) Secondary processing: particle filtering (resampling threshold 0.6, number of particles 2000),

[0084] (c) Dynamic parameter adjustment: When the environmental interference level is greater than level 3, the alternative filter coefficient library is switched and the LSTM model is run to predict the movement trajectory of the clothing. When the error is greater than 15cm, position correction is triggered.

[0085] (3) Cloud platform linkage

[0086] (a) The digital twin engine updates the inventory 3D heat map every 500ms (grid precision 0.5m×0.5m).

[0087] (b) Blockchain nodes (Hyperledger Fabric 2.4) record operation logs in real time and generate new blocks every 10 minutes.

[0088] (c) The inventory forecasting model (Prophet+ARIMA hybrid model) generates a replenishment recommendation list at 6:00 every day.

[0089] Step 4: System self-optimization

[0090] (1) Online learning mechanism

[0091] (a) Incremental model training starts every Friday at 23:00: Collect environmental data (temperature, humidity, electromagnetic interference) for the week to build a new feature set. Use the domain adversarial network (DANN) to align the feature distributions of the new and old scenes. Use small sample data (<100 groups) for meta-learning and rapid tuning. Evaluate system performance monthly. When the tag miss rate is >0.5% for three consecutive days, trigger a full system calibration.

[0092] (2) Exception handling process

[0093] (a) Level 1 alarm (inventory deviation > 2%): Automatically generate an inventory task and dispatch it to the AGV.

[0094] (b) Level 2 alarm (tag lost contact for > 30 minutes): Activate UWB-assisted positioning and mark the abnormal area.

[0095] (c) Level 3 alarm (multi-sensor data conflict): Start the manual review mode and record the abnormal characteristics.

[0096] Finally, a few points should be explained: First, in the description of this application, it should be noted that, unless otherwise specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense, and may refer to mechanical or electrical connections, internal communication between two components, or direct connection. "Up," "down," "left," and "right" are only used to indicate relative positional relationships. When the absolute positions of the objects being described change, the relative positional relationships may also change.

[0097] Secondly: The drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention. Other structures may refer to conventional designs. The same embodiment and different embodiments of the present invention may be combined with each other without conflict.

[0098] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An RFID-based intelligent clothing inventory real-time monitoring and automatic inventory system, characterized by: include: The reconfigurable RFID tag set is encapsulated in the clothing lining through a flexible substrate, integrates a spiral-dipole composite antenna and temperature compensation circuit, and supports 860-960MHz dual-band adaptive switching; A heterogeneous reader / writer network consists of a three-dimensional scanning array consisting of a primary reader, fixed auxiliary readers, and mobile readers. The primary reader is deployed on the warehouse's top truss, the fixed auxiliary readers are embedded in the shelf columns, and the mobile readers are mounted on AGV robots. The edge computing layer, an FPGA processor deployed near the reader, runs a multi-source data fusion algorithm, including an LSTM motion prediction unit and a hybrid filtering denoising module; The cloud management platform communicates with the edge layer via the OPC-UA protocol to build a digital twin warehouse model and implement blockchain traceability; Collaborative working mechanism: the main reader performs global time slot scheduling, the mobile reader performs blind scanning along dynamic paths, and the edge layer pre-processes the data and uploads it to the cloud to generate a real-time inventory map.

2. The RFID-based smart clothing inventory real-time monitoring and automatic inventory system according to claim 1 is characterized in that: The reconfigurable RFID tag's flexible antenna uses a silver nanowire mesh structure fabricated using a photolithographic process. It switches its operating frequency band via a PIN diode. Its self-calibration function involves a temperature sensor monitoring environmental changes in real time, a microcontroller adjusting the antenna impedance matching network based on the temperature gradient, and automatically triggering a retuning process when signal strength fluctuations exceed ±3dB.

3. The RFID-based smart clothing inventory real-time monitoring and automatic inventory system according to claim 1 is characterized in that: The deployment method of the heterogeneous reader-writer network includes: spatial topology construction, calculating the optimal reader-writer layout based on the warehouse CAD model, and the main reader-writer spacing is 1.2-1.5 times the shelf height; signal coverage optimization, using beamforming technology to directionally enhance the field strength in the deep area of ​​the shelf to compensate for the signal attenuation caused by metal shelves; mobile node collaboration, the AGV-mounted reader-writer establishes time-division duplex communication with the fixed reader-writer during inspection to avoid channel conflicts.

4. The RFID-based smart clothing inventory real-time monitoring and automatic inventory system according to claim 1 is characterized in that: The hybrid filtering and denoising module of the edge computing layer includes: first, a primary filtering unit that uses an improved Kalman filter to eliminate Gaussian noise in RFID signals; second, a secondary processing unit that applies a particle filter algorithm to address signal distortion caused by multipath effects; and finally, a dynamic parameter adjustment mechanism that automatically adjusts the filter window width based on the RSSI fluctuation rate and activates an alternative filter coefficient library when the environmental interference level is greater than level 4.

5. The RFID-based smart clothing inventory real-time monitoring and automatic inventory system according to claim 1 is characterized in that: The digital twin warehouse model construction process of the cloud management platform includes a physical mapping layer, a real-time data layer, and a business logic layer. The physical mapping layer imports warehouse three-dimensional structure data through the BIM interface; the real-time data layer fuses RFID coordinate information with UWB positioning data to generate a clothing location heat map; and the business logic layer integrates the inventory forecasting algorithm and the replenishment strategy optimization engine to trigger an intelligent replenishment work order when the inventory of a certain SKU falls below a safety threshold.

6. The RFID-based smart clothing inventory real-time monitoring and automatic inventory system according to claim 3 is characterized in that: The AGV robot's navigation system uses: multi-sensor fusion positioning, integrating UWB anchor point signals, IMU inertial data and visual odometry, with a positioning accuracy of less than 5cm; adaptive path planning, generating a global path based on the Dijkstra algorithm, combined with the dynamic window method for real-time obstacle avoidance; and a task scheduling strategy that dynamically adjusts inventory priority based on inventory discrepancy rates and automatically increases scanning frequency in high-deviation areas.

7. The RFID-based smart clothing inventory real-time monitoring and automatic inventory system according to claim 1 is characterized in that: The global time slot scheduling method specifically includes three stages. The first stage is the conflict detection stage, in which each reader broadcasts the channel occupancy status and constructs a conflict probability matrix; the second stage is the time slot allocation stage, in which an improved Q-learning algorithm is used to calculate the optimal time slot combination, and a reward function comprehensively evaluates channel utilization and energy consumption indicators; the third stage is the dynamic adjustment stage, in which distributed negotiation is triggered to reallocate time slot resources when a new reader is connected or the tag density changes by more than 15%.

8. The RFID-based smart clothing inventory real-time monitoring and automatic inventory system according to claim 1 is characterized in that: The blockchain traceability module's workflow is as follows: data is uploaded to the blockchain, generating a Merkle tree structure from clothing flow information collected by RFID and storing it in Hyperledger Fabric; smart contracts use pre-set quality inspection rules to automatically verify the integrity of incoming clothing information, triggering alarms and freezing related transactions if abnormal data occurs. Cross-chain interaction: atomic data exchange with suppliers and logistics parties’ blockchain systems is achieved through cross-chain gateways.

9. The RFID-based smart clothing inventory real-time monitoring and automatic inventory system according to claim 4, characterized in that: The execution logic of the dynamic parameter adjustment mechanism is as follows: first, an environmental interference level assessment model is established, and the interference intensity is divided into levels 1-5 according to the background noise power spectrum density; when the interference level increases, the process noise covariance matrix Q value of the Kalman filter is gradually increased; and the particle filter resampling threshold is simultaneously reduced to retain more large-weight particles to maintain tracking stability.

10. The RFID-based smart clothing inventory real-time monitoring and automatic inventory system according to claim 5, characterized in that: The intelligent replenishment work order generation method includes: demand forecasting, using an LSTM network to analyze historical sales data and seasonal trend factors; supplier evaluation, generating supplier scores based on the on-time delivery rate and quality compliance rate recorded in the blockchain; and automatic order dispatching, matching replenishment demand with the inventory data of preferred suppliers to generate an electronic purchase order with an attached smart contract.

Images