An intelligent warehousing goods identification and inventory system

Abstract

This invention relates to the field of intelligent warehouse management technology, specifically an intelligent warehouse cargo identification and inventory system. The system synchronously collects multi-dimensional signals caused by cargo removal or replenishment actions through a multi-dimensional sensing unit; extracts signal feature components representing the current cargo removal or replenishment action from each dimension signal, and merges them to form a multi-dimensional joint signal feature vector; pre-establishes an indirect sensing mapping model for deeply obscured storage locations; inputs the multi-dimensional joint signal feature vector into the indirect sensing mapping model, outputs a joint posterior probability distribution of the change in cargo quantity in the target deeply obscured storage location, obtains the cargo quantity change with the highest probability in the joint posterior probability distribution and the corresponding target deeply obscured storage location, and updates the baseline inventory data of the target deeply obscured storage location in real time to form inventory change details; thus achieving real-time identification of cargo quantity changes in deeply obscured storage locations and dynamic and accurate updating of inventory data.

Description

Technical Field

[0001] This invention relates to the field of intelligent warehouse management technology, specifically an intelligent warehouse cargo identification and inventory system. Background Technology

[0002] Intelligent warehousing systems are a core infrastructure for modern logistics and supply chain management. Through real-time identification and accurate inventory counting of goods on shelves, they enable dynamic maintenance and efficient control of inventory data. As the requirements for warehousing efficiency continue to increase, shelf layouts are becoming increasingly high-density and multi-layered, with a significant increase in the number of storage locations and more complex types and turnover frequencies of goods. This places higher demands on the real-time performance and accuracy of inventory identification and counting systems.

[0003] However, in high-density shelving environments, storage locations are often stacked in multiple rows along the depth, with goods in the front row continuously obstructing the view of those behind. Existing goods identification and inventory technologies mainly rely on visual cameras, RFID tag scanning, or manual inspection. These methods can meet the needs in open storage locations or single-layer shelving scenarios, but they have fundamental limitations when dealing with deeply obstructed storage locations: visual sensors cannot capture changes in the status of goods in deep storage locations due to physical obstruction; RFID systems suffer from reduced read reliability in dense metal shelving environments due to multipath interference and signal obstruction; and manual inventory requires frequent movement of goods in the front row to verify deep storage locations, which is time-consuming, labor-intensive, and prone to human error. More importantly, every hidden retrieval or replenishment action that occurs in a deeply concealed storage location cannot be captured and recorded in a timely manner by any of the existing methods mentioned above. Changes in the quantity of goods can only be discovered during regular manual inventory checks, causing inventory data to lag behind the actual physical state for a long time. Errors continue to accumulate, ultimately leading to a series of chain problems such as inaccurate inventory, delayed replenishment decisions, and even a decline in order fulfillment capabilities. Summary of the Invention

[0004] (1) Technical problems to be solved The purpose of this invention is to provide an intelligent warehouse goods identification and inventory system to solve the problem that in high-density shelving, existing systems cannot detect changes in the quantity of goods in deeply obscured locations in real time without moving the goods in front, and every hidden picking or replenishment action cannot be effectively recorded, resulting in the inability to accurately identify and track changes in inventory quantity in real time.

[0005] (2) Technical solution To achieve the above objectives, on the one hand, the present invention provides an intelligent warehouse cargo identification and inventory system, the system comprising: an initial inventory data generation module, a multi-dimensional signal acquisition and processing module, an indirect perception mapping model establishment module, a cargo quantity change determination module, and an inventory data update and recording module, wherein each module is sequentially connected in communication. The initial inventory data generation module is used to initially identify the goods in each location on the shelf, record the type and quantity of goods in each location, and generate the baseline inventory data for each location.

[0006] A multi-dimensional signal acquisition and processing module is used to deploy multi-dimensional sensing units on the load-bearing structure of the shelf. The multi-dimensional sensing units synchronously acquire multi-dimensional signals caused by the removal or replenishment of goods in real time. The multi-dimensional signals include stress change signals, vibration signals, radio frequency scattering pattern change signals, and local weight distribution change signals. After synchronizing and aligning the multi-dimensional signals, the module extracts the signal feature components representing the removal or replenishment of goods from each dimension signal, and merges the signal feature components of each dimension signal to form a multi-dimensional joint signal feature vector.

[0007] The indirect perception mapping model building module is used to pre-build an indirect perception mapping model for deeply obscured storage locations based on the structural parameters of the shelving and the spatial positional relationship of each storage location relative to the multi-dimensional perception unit.

[0008] The cargo quantity change determination module is used to input the multi-dimensional joint signal feature vector into the indirect perception mapping model, output the joint posterior probability distribution of the cargo quantity change of the target deep occlusion cargo location, obtain the cargo quantity change with the highest probability in the joint posterior probability distribution, and determine the target deep occlusion cargo location where the cargo quantity change has occurred.

[0009] The inventory data update and recording module is used to update the baseline inventory data of the target deep-covered storage location in real time based on the change in the quantity of goods in the target deep-covered storage location, and record the timestamp of this change, storage location number and change in the quantity of goods, forming an inventory change detail.

[0010] Furthermore, the indirect perception mapping model establishment module includes: The shelving parameter acquisition and coordinate determination module is used to acquire the elastic modulus, material density, structural damping coefficient, cross-sectional dimensions and connection methods between components of the shelving load-bearing structure, establish a three-dimensional coordinate system for the shelving, and determine the spatial coordinates of each storage location and the installation position coordinates of each multi-dimensional sensing unit in the three-dimensional coordinate system of the shelving.

[0011] The forward mapping relationship matrix construction module is used to establish a signal transmission path model for multi-dimensional signals, using the spatial coordinates of each cargo location as the signal excitation point and the installation position coordinates of each multi-dimensional sensing unit as the signal receiving point. Taking the change in the quantity of goods per unit as the benchmark excitation input, it calculates the theoretical signal response vectors of each dimension corresponding to each deeply obscured cargo location according to the signal transmission path model. It integrates and arranges the theoretical signal response vectors of each dimension of the same deeply obscured cargo location to form a multi-dimensional joint theoretical signal response vector. It summarizes the multi-dimensional joint theoretical signal response vectors of all deeply obscured cargo locations and constructs a forward mapping relationship matrix between each deeply obscured cargo location and the multi-dimensional joint theoretical signal response vector.

[0012] The indirect perception mapping model forming module is used to introduce a Bayesian inference framework based on the positive mapping relationship matrix. The historical replenishment frequency of each deeply obscured cargo location is set as the prior probability. The likelihood function is constructed by the residual between the multidimensional joint theoretical signal response vector and the multidimensional joint signal feature vector actually collected by the multidimensional sensing unit. This forms an indirect perception mapping model with the multidimensional joint signal feature vector as input and the joint posterior probability distribution of the change in the quantity of goods in each deeply obscured cargo location as output.

[0013] Furthermore, the positive mapping matrix construction module includes: The stress change signal theoretical response vector generation module is used to construct a finite element static mechanical model for stress change signals of multi-dimensional signals based on the elastic modulus, cross-sectional dimensions, and connection methods of each component of the rack bearing structure. The finite element static mechanical model is used as the signal transmission path model of the stress change signal. In the rack three-dimensional coordinate system, the spatial coordinates of the i-th deep-shaded storage location are used as the load application point. The standard weight corresponding to the change in the number of goods per unit is used as the concentrated force load value and applied along the direction of gravity. The installation position coordinates of the multi-dimensional sensing unit are used as the stress observation point. The stress increment value caused by the concentrated force load at each stress observation point is solved through the finite element static mechanical model. The stress increment values ​​of all stress observation points are arranged in order to form the stress change signal theoretical response vector.

[0014] Furthermore, the positive mapping matrix construction module includes: The vibration signal theoretical response vector generation module is used to construct transient dynamic wave equations for multi-dimensional vibration signals based on the material density, structural damping coefficient, and elastic modulus of each component of the rack bearing structure. These transient dynamic wave equations serve as the signal transmission path model for the vibration signal. The spatial coordinates of the i-th deeply obscured cargo location are used as the vibration excitation source point. The time function of the force generated by the cargo removal or replenishment action corresponding to the unit change in cargo quantity is used as the excitation boundary condition. The transient dynamic wave equations are solved to obtain the vibration displacement versus time curve at the installation position coordinates of each multi-dimensional sensing unit. Three characteristic quantities—peak amplitude, dominant frequency energy ratio, and vibration decay time constant—are extracted from each vibration displacement versus time curve. These three characteristic quantities of all multi-dimensional sensing units are then sequentially concatenated to form the vibration signal theoretical response vector.

[0015] Furthermore, the vibration signal theoretical response vector generation module includes the following steps: The maximum absolute value of the vibration displacement in the vibration displacement versus time curve is determined as the peak amplitude. The power spectral density distribution of the vibration signal is obtained by performing a Fourier transform on the vibration displacement versus time curve. The frequency corresponding to the maximum power spectral density value is identified from the power spectral density distribution as the dominant frequency. The dominant frequency energy value centered on the dominant frequency and with a bandwidth of a preset multiple is calculated. The ratio of the dominant frequency energy value to the total energy value of the power spectral density distribution is determined as the dominant frequency energy proportion.

[0016] Extract the time interval from the vibration initiation moment to the vibration displacement decaying to a preset proportional threshold from the vibration displacement change curve over time. Fit the envelope of the vibration displacement change curve over time within the time interval with an exponential decay function. Determine the time constant in the fitted exponential decay function as the vibration decay time constant.

[0017] Furthermore, the positive mapping matrix construction module includes: The radio frequency scattering pattern change signal theoretical response vector generation module is used to establish a ray tracing propagation model for the radio frequency scattering pattern change signal of multi-dimensional signals based on the geometric dimensions and material electromagnetic parameters of the shelf bearing structure. The ray tracing propagation model is used as the signal transmission path model of the radio frequency scattering pattern change signal. The spatial coordinates of the i-th deep-shaded cargo location are used as the scattering disturbance center. The change in dielectric constant at the scattering disturbance center caused by the change in the quantity of goods is set as the unit disturbance intensity.

[0018] Using the unit disturbance intensity as input excitation, the complex changes in the radio frequency forward scattering coefficient and back scattering coefficient received at the installation position coordinates of each multidimensional sensing unit are calculated through the ray tracing propagation model. The real and imaginary parts of the complex changes of each multidimensional sensing unit are arranged in order to form the theoretical response vector of the radio frequency scattering pattern change signal.

[0019] Furthermore, the radio frequency scattering pattern change signal theoretical response vector generation module includes the following steps: A three-dimensional spatial geometric model is established based on the geometric dimensions of the rack support structure. In the three-dimensional spatial geometric model, the position coordinates of the radio frequency transmitter, the operating frequency and transmission power of the radio frequency transmitter, the installation position coordinates of the radio frequency receiving sensor in each multi-dimensional sensing unit, and the spatial coordinates of each deeply obscured storage location are set. Based on the material electromagnetic parameters of the rack support structure, the relative permittivity and conductivity of each component in the three-dimensional spatial geometric model are set. The three-dimensional spatial geometric model is divided into multiple discrete reflecting surfaces and diffraction edges, and a ray tracing propagation model including direct path, reflection path and diffraction path is constructed.

[0020] Furthermore, the positive mapping matrix construction module includes: The local weight distribution change signal theoretical response vector generation module is used to construct a weight distribution transfer matrix for local weight distribution change signals of multi-dimensional signals based on the cross-sectional dimensions, elastic modulus, and connection methods of each component of the rack bearing structure. The weight distribution transfer matrix is ​​used as the signal transmission path model of the local weight distribution change signal. The spatial coordinates of the i-th deeply obscured cargo location are used as the weight input node, and the weight increment corresponding to the change in the number of goods per unit is used as the non-zero element in the input vector. The change in support reaction force at the installation position coordinates of each multi-dimensional sensing unit is calculated through the weight distribution transfer matrix. All the changes in support reaction force are arranged in order to form the local weight distribution change signal theoretical response vector.

[0021] Furthermore, the local weight distribution change signal theoretical response vector generation module includes the following steps: The rack support structure is discretized into multiple structural units. Degrees of freedom are defined at the nodes of each structural unit. These degrees of freedom include the linear displacement of the node along each coordinate axis and the angular displacement around each coordinate axis in the rack's three-dimensional coordinate system. The element stiffness matrix of each structural unit is established based on its cross-sectional dimensions and elastic modulus. The element stiffness matrices are assembled into the overall stiffness matrix of the rack support structure according to the connection method between components. The overall stiffness matrix is ​​inverted and multiplied by the constraint force transformation matrix corresponding to the constraint degrees of freedom to form the weight distribution transfer matrix from the load vector of each node to the support reaction force vector of each constraint node.

[0022] (3) Beneficial effects Compared with the prior art, the beneficial effects of the present invention are: 1. By deploying multi-dimensional sensing units on the shelving structure, signals from four dimensions—stress change, vibration, radio frequency scattering pattern change, and local weight distribution—are simultaneously collected and fused to form a multi-dimensional joint signal feature vector. This overcomes the fundamental limitation of traditional visual perception and single sensing methods, which cannot perceive the status of deep storage locations due to physical obstruction. It enables the effective capture of the actions of retrieving or replenishing goods in deep, obscured storage locations without moving the goods in front or interrupting normal warehousing operations.

[0023] 2. A multi-dimensional signal transmission path model is pre-established for each deeply obscured cargo location. By combining a Bayesian inference framework with prior information on historical replenishment, an indirect perception mapping model is constructed from the multi-dimensional joint signal feature vector to the joint posterior probability distribution of the change in cargo quantity in each deeply obscured cargo location. This enables the system to accurately locate the changed target cargo location and quantify the change in cargo quantity in a probabilistically optimal manner in complex scenarios where the signal responses of multiple cargo locations are highly mixed. This significantly improves the accuracy and robustness of identifying deeply obscured cargo locations.

[0024] 3. Each identification result is written to the baseline inventory data in real time and the timestamp, location number and change in quantity of goods are recorded synchronously to form a continuous and complete inventory change detail. This ensures that the inventory data is always highly consistent with the actual status of the goods on the shelves, thereby eliminating the problem of long-term lag and error accumulation in inventory data caused by the inability to record changes in deep-seated locations. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the module composition of an intelligent warehousing cargo identification and inventory system according to the present invention. Detailed Implementation

[0026] 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.

[0027] Example 1: As Figure 1 As shown, this embodiment provides an intelligent warehouse cargo identification and inventory system. The system includes: an initial inventory data generation module, a multi-dimensional signal acquisition and processing module, an indirect perception mapping model establishment module, a cargo quantity change determination module, and an inventory data update and recording module. Each module is connected to the other in a sequential manner. The initial inventory data generation module is used to initially identify the goods in each location on the shelf, record the type and quantity of goods in each location, and generate the baseline inventory data for each location.

[0028] For example, initial identification can be completed through manual inventory checks at each location, barcode scanning and data entry, or visual recognition system photography; the specific implementation method is not limited. Taking a fast-moving consumer goods warehouse as an example, the warehouse deploys a set of standard heavy-duty shelving with eight layers. Each layer has four rows of storage locations arranged longitudinally, and each layer has twenty storage locations arranged laterally, for a total of 640 storage locations. After the system is deployed, warehouse managers use handheld barcode scanners to scan the barcodes of goods at each storage location and enter the quantity. The system writes the goods category code (e.g., "SKU-20XX1001") and the corresponding quantity (e.g., 12 items of goods with category code "SKU-20XX1001" are stored in the storage location of the 5th column of the 3rd layer) into the database, forming the baseline inventory record for that storage location. After all 640 storage locations are scanned, the system database contains a complete baseline inventory dataset, with each record containing three basic fields: storage location number, goods category code, and goods quantity.

[0029] A multi-dimensional signal acquisition and processing module is used to deploy multi-dimensional sensing units on the load-bearing structure of the shelf. The multi-dimensional sensing units synchronously acquire multi-dimensional signals caused by the removal or replenishment of goods in real time. The multi-dimensional signals include stress change signals, vibration signals, radio frequency scattering pattern change signals, and local weight distribution change signals. After synchronizing and aligning the multi-dimensional signals, the module extracts the signal feature components representing the removal or replenishment of goods from each dimension signal, and merges the signal feature components of each dimension signal to form a multi-dimensional joint signal feature vector.

[0030] For example, after initial identification, multi-dimensional sensing units are deployed at designated installation locations on the load-bearing structure of the shelving (including uprights, beams, shelves, and other load-bearing components). A multi-dimensional sensing unit is a composite sensing device integrating multiple types of sensors. Each unit integrates four types of sensors: stress sensors, vibration acceleration sensors, radio frequency transceiver sensors, and pressure sensors. These four types of sensors share the same clock reference and can synchronously trigger data acquisition when goods are removed or replenished. The installation locations of the multi-dimensional sensing units are uniformly planned according to the shelving structure, typically installed at the junctions of beams and uprights and near the support points of shelves to ensure effective sensing of various signals. Taking the aforementioned fast-moving consumer goods warehouse as an example, each shelving unit has a multi-dimensional sensing unit installation node every two layers along the upright height direction. Simultaneously, a pressure sensor measuring point is installed at each of the four corner points of each shelf layer. A total of sixteen multi-dimensional sensing units are deployed across the entire shelving unit, forming a three-dimensional sensing network covering the entire shelving.

[0031] The multi-dimensional sensing unit synchronously collects four types of multi-dimensional signals caused by the removal or replenishment of goods, as follows: Stress change signal reflects the change in stress distribution within the load-bearing structure caused by changes in goods weight. A stress sensor collects static stress values ​​before and after the goods movement, and the difference between the two values ​​is the stress increment. Vibration signal reflects the transient impact and elastic wave propagation generated on the rack structure by the removal or replenishment of goods. A vibration acceleration sensor continuously collects vibration acceleration time series within 500 milliseconds before and after the goods movement at a sampling frequency of no less than 1000 times per second. The vibration displacement versus time curve is obtained by quadratic integration of the acceleration time series. Radio frequency scattering pattern change signal reflects the change in electromagnetic scattering characteristics within the rack space caused by changes in the quantity of goods. An radio frequency transceiver sensor transmits a detection pulse before and after the goods movement and receives the scattered echo. The difference in scattering coefficients between the two echoes is extracted as the radio frequency scattering pattern change. Local weight distribution change signal reflects the change in support reaction force distribution at each support point of the rack caused by changes in goods weight. A pressure sensor collects static pressure values ​​before and after the goods movement, and the difference between the two values ​​is the support reaction force change.

[0032] Although the four types of sensors share a clock reference, slight deviations may occur in the arrival times of various signals at the data processing module due to data transmission link delays. Therefore, synchronization and alignment of the four types of signals are necessary before extracting signal features. The specific method for synchronization and alignment is as follows: using the earliest detected signal change among the four types of signals as the reference time zero point, the remaining signals are shifted along the time axis according to the calibrated time delay difference between their respective sensors and the reference sensor. This ensures that the start times of the same physical event corresponding to the four types of signals remain strictly consistent on the time axis. The time delay difference is pre-determined through calibration experiments during the system installation and commissioning phase and stored in the system parameter file, which is then directly accessed during operation.

[0033] After synchronization and alignment are completed, signal feature components are extracted from the four types of signals. For stress change signals, the stress increment value at each stress sensor measuring point is directly used as the feature component. Assume the system has a total of [number missing] sensors. If there are 10 stress sensor measurement points, then the characteristic components of the stress change signal are: The multidimensional vector has each component in megapascals (MPA). For vibration signals, three features are extracted from the vibration displacement versus time curve at each multidimensional sensing unit: peak amplitude, dominant frequency energy ratio, and vibration decay time constant. The peak amplitude is the maximum absolute value of the vibration displacement, measured in millimeters. The dominant frequency energy ratio is the ratio of the energy in the frequency band near the dominant frequency to the total signal energy, a dimensionless value ranging from 0 to 1. The vibration decay time constant is the envelope decaying exponentially. The time constant obtained after fitting The unit is milliseconds, where The initial amplitude, The variable is time. Assume the system has a total deployment... If there are multiple sensing units, then the characteristic components of the vibration signal are: Multidimensional vector. For a radio frequency scattering pattern variation signal, the real and imaginary parts of the complex changes in the forward scattering coefficient and backscattering coefficient at the radio frequency receiving sensor in each multidimensional sensing unit are respectively taken as feature components. Each multidimensional sensing unit contributes four components (real part of forward scattering coefficient, imaginary part of forward scattering coefficient, real part of backscattering coefficient, and imaginary part of backscattering coefficient). Assume the system has a total of [number missing] [units missing]. If there are multiple sensing units, then the characteristic components of the radio frequency scattering pattern change signal are: A dimensionless scattering coefficient vector. For signals showing localized weight distribution changes, the change in support reaction force at each pressure sensor measuring point is directly used as a characteristic component. Assume the system has a total of [number missing] sensors. If there are 1 pressure sensor measuring points, then the characteristic components of the local weight distribution change signal are: A dimensional vector, with units of Newtons. The four types of signal feature components are sequentially concatenated in a fixed order: stress change signal feature components, vibration signal feature components, radio frequency scattering pattern change signal feature components, and local weight distribution change signal feature components, forming a vector with a total dimension of... The multidimensional joint signal feature vector is denoted as vector x.

[0034] Taking the aforementioned fast-moving consumer goods warehouse as an example, the warehouse shelves are equipped with sixteen multi-dimensional sensing units ( Each multidimensional sensing unit contains a stress sensor. ), vibration acceleration sensors and radio frequency transceiver sensors, and pressure sensors are also installed at the four corners of the shelf, for a total of thirty-two ( Then, the total dimension of the multidimensional joint signal feature vector x is . When a picker retrieves two items from the second column of the third shelf, sixteen multidimensional sensing units simultaneously detect the signal change. Within approximately two hundred milliseconds after the action occurs, the system completes signal acquisition, synchronization alignment, and feature extraction, outputting a 160-dimensional multidimensional joint signal feature vector for subsequent inference by the indirect sensing mapping model.

[0035] The indirect perception mapping model building module is used to pre-build an indirect perception mapping model for deeply obscured storage locations based on the structural parameters of the shelving and the spatial positional relationship of each storage location relative to the multi-dimensional perception unit. The indirect perception mapping model establishment module includes: The shelving parameter acquisition and coordinate determination module is used to acquire the elastic modulus, material density, structural damping coefficient, cross-sectional dimensions and connection methods between components of the shelving load-bearing structure, establish a three-dimensional coordinate system for the shelving, and determine the spatial coordinates of each storage location and the installation position coordinates of each multi-dimensional sensing unit in the three-dimensional coordinate system of the shelving.

[0036] For example, the physical parameters of each component of the rack's load-bearing structure are obtained, including the steel's elastic modulus (typically 210 GPa), material density (typically 7850 kg / m³), structural damping coefficient (typically a dimensionless value between 0.02 and 0.05), the cross-sectional dimensions of each component (including width, height, and wall thickness in millimeters), and the connection method between components (bolted or welded). A three-dimensional coordinate system for the rack is established, with a fixed corner point at the bottom of the rack as the origin, the rack's depth as the positive X-axis, the rack's transverse direction as the positive Y-axis, and the vertical upward direction as the positive Z-axis. Based on the rack design drawings, the spatial coordinates of all deeply obscured storage locations (represented by the coordinates of the geometric center point of the storage location) and the installation position coordinates of all multi-dimensional sensing units are determined within the three-dimensional coordinate system. Taking the aforementioned fast-moving consumer goods warehouse as an example, there are four rows of storage locations in the longitudinal direction of the shelving. The first row is the front row of visible storage locations, and the second to fourth rows are deep-covered storage locations. There are a total of three layers of depth multiplied by eight layers of height multiplied by twenty horizontal storage locations, totaling four hundred and eighty deep-covered storage locations. Their spatial coordinates are entered one by one according to the shelving drawings.

[0037] The forward mapping relationship matrix construction module is used to establish a signal transmission path model for multi-dimensional signals, using the spatial coordinates of each cargo location as the signal excitation point and the installation position coordinates of each multi-dimensional sensing unit as the signal receiving point; and to calculate the theoretical signal response vectors of each dimension corresponding to each deeply obscured cargo location based on the signal transmission path model, using the change in the quantity of goods per unit as the benchmark excitation input. The positive mapping relationship matrix construction module includes: The stress change signal theoretical response vector generation module is used to construct a finite element static mechanical model for stress change signals of multi-dimensional signals based on the elastic modulus, cross-sectional dimensions, and connection methods of each component of the rack bearing structure. The finite element static mechanical model is used as the signal transmission path model of the stress change signal. In the rack three-dimensional coordinate system, the spatial coordinates of the i-th deep-shaded storage location are used as the load application point. The standard weight corresponding to the change in the number of goods per unit is used as the concentrated force load value and applied along the direction of gravity. The installation position coordinates of the multi-dimensional sensing unit are used as the stress observation point. The stress increment value caused by the concentrated force load at each stress observation point is solved through the finite element static mechanical model. The stress increment values ​​of all stress observation points are arranged in order to form the stress change signal theoretical response vector.

[0038] For example, based on the elastic modulus, cross-sectional dimensions, and connection methods of each component of the rack's load-bearing structure, the finite element method is used to discretize the rack's load-bearing structure into a finite element mesh composed of beam elements and plate elements, establishing a finite element static mechanical model. In the rack's three-dimensional coordinate system, the first... The spatial coordinates of a deeply obscured cargo location are used as the application point of a concentrated force load, with the standard weight corresponding to a unit change in cargo quantity along the gravity load. A concentrated force load is applied in the axial direction. The standard weight corresponding to the change in the number of goods per unit is determined based on the standard weight of a single item stored in that location. For example, if the weight of a single item stored in that location is 1.5 kg, then the concentrated force load value corresponding to the change in the number of goods per unit is... Newton. Solving all problems using a finite element static mechanics model. The stress increment value at each stress sensor measuring point caused by the concentrated force load will be The stress increment values ​​are arranged in order of sensor number to form the [number]. The theoretical response vector of the stress change signal corresponding to a deeply obscured cargo location is denoted as s. , dimension .

[0039] The positive mapping relationship matrix construction module includes: The vibration signal theoretical response vector generation module is used to construct transient dynamic wave equations for multi-dimensional vibration signals based on the material density, structural damping coefficient, and elastic modulus of each component of the rack bearing structure. These transient dynamic wave equations serve as the signal transmission path model for the vibration signal. The spatial coordinates of the i-th deeply obscured cargo location are used as the vibration excitation source point. The time function of the force generated by the cargo removal or replenishment action corresponding to the unit change in cargo quantity is used as the excitation boundary condition. The transient dynamic wave equations are solved to obtain the vibration displacement versus time curve at the installation position coordinates of each multi-dimensional sensing unit. Three characteristic quantities—peak amplitude, dominant frequency energy ratio, and vibration decay time constant—are extracted from each vibration displacement versus time curve. These three characteristic quantities of all multi-dimensional sensing units are then sequentially concatenated to form the vibration signal theoretical response vector.

[0040] The vibration signal theoretical response vector generation module includes the following steps: The maximum absolute value of the vibration displacement in the vibration displacement versus time curve is determined as the peak amplitude. The power spectral density distribution of the vibration signal is obtained by performing a Fourier transform on the vibration displacement versus time curve. The frequency corresponding to the maximum power spectral density value is identified from the power spectral density distribution as the dominant frequency. The dominant frequency energy value centered on the dominant frequency and with a bandwidth of a preset multiple is calculated. The ratio of the dominant frequency energy value to the total energy value of the power spectral density distribution is determined as the dominant frequency energy proportion.

[0041] Extract the time interval from the vibration initiation moment to the vibration displacement decaying to a preset proportional threshold from the vibration displacement change curve over time. Fit the envelope of the vibration displacement change curve over time within the time interval with an exponential decay function. Determine the time constant in the fitted exponential decay function as the vibration decay time constant.

[0042] For example, for vibration signals, based on the material density, structural damping coefficient, and elastic modulus of each component of the rack's load-bearing structure, a transient dynamic wave equation is established using the mass matrix, damping matrix, and stiffness matrix: .

[0043] in, The structural mass matrix, Here is the structural damping matrix. Here is the structural stiffness matrix. Let be the structural node displacement vector. Let be the time function vector of the external force. (The rest of the text appears to be a list of vectors or parameters, possibly The spatial coordinates of a deeply obscured cargo location are used as the vibration excitation source point. The force-time function generated by the cargo removal or replenishment action corresponding to the unit change in cargo quantity is applied to this node as the excitation boundary condition. The force-time function adopts the form of a half-sine pulse. .

[0044] in, This is the peak force; The duration of the action is typically 0.05 to 0.2 seconds. Numerical solution of the transient dynamic wave equation yields all... Vibration displacement versus time curves at the installation locations of the multi-dimensional sensing units were analyzed. Peak amplitude, dominant frequency energy percentage, and vibration decay time constant were extracted from each curve. The three feature quantities of the first multidimensional sensing unit are sequentially concatenated according to the multidimensional sensing unit numbering order to form the second... The theoretical vibration signal response vector corresponding to a deeply obstructed storage location is denoted as v. , dimension .

[0045] The positive mapping relationship matrix construction module includes: The module for generating theoretical response vectors of radio frequency scattering pattern change signals is used to establish a ray tracing propagation model for radio frequency scattering pattern change signals of multi-dimensional signals based on the geometric dimensions and material electromagnetic parameters of the shelf-supporting structure.

[0046] The theoretical response vector generation module for the radio frequency scattering pattern change signal includes the following steps: A three-dimensional spatial geometric model is established based on the geometric dimensions of the rack support structure. In the three-dimensional spatial geometric model, the position coordinates of the radio frequency transmitter, the operating frequency and transmission power of the radio frequency transmitter, the installation position coordinates of the radio frequency receiving sensor in each multi-dimensional sensing unit, and the spatial coordinates of each deeply obscured storage location are set. Based on the material electromagnetic parameters of the rack support structure, the relative permittivity and conductivity of each component in the three-dimensional spatial geometric model are set. The three-dimensional spatial geometric model is divided into multiple discrete reflecting surfaces and diffraction edges, and a ray tracing propagation model including direct path, reflection path and diffraction path is constructed.

[0047] The ray tracing propagation model is used as the signal transmission path model for the radio frequency scattering pattern change signal. The spatial coordinates of the i-th deep-shaded cargo location are taken as the scattering disturbance center, and the change in dielectric constant at the scattering disturbance center caused by the change in the quantity of cargo is set as the unit disturbance intensity.

[0048] Using the unit disturbance intensity as input excitation, the complex changes in the radio frequency forward scattering coefficient and back scattering coefficient received at the installation position coordinates of each multidimensional sensing unit are calculated through the ray tracing propagation model. The real and imaginary parts of the complex changes of each multidimensional sensing unit are arranged in order to form the theoretical response vector of the radio frequency scattering pattern change signal.

[0049] For example, for radio frequency (RF) scattering pattern change signals, a three-dimensional spatial geometric model is established based on the geometric dimensions of the rack support structure. Within this model, the location coordinates of the RF transmitter (typically installed at the center of the top of the rack), its operating frequency (typically 2.4 GHz or 5.8 GHz in the industrial license-free band), and its transmission power (typically 100 milliwatts) are set. The installation location coordinates of the RF receiving sensors in each multi-dimensional sensing unit and the spatial coordinates of each deeply obscured storage location are also set. Based on the material electromagnetic parameters of the rack support structure (the typical relative permittivity of steel is approximately 1, and the typical conductivity is approximately one million Siemens per meter), the electromagnetic properties of each component in the three-dimensional spatial geometric model are set. The surface of the three-dimensional spatial geometric model is divided into multiple discrete reflecting surfaces and diffraction edges, constructing a ray tracing propagation model that includes direct paths, reflected paths, and diffracted paths. Using the first... The spatial coordinates of the deeply obscured cargo location are taken as the scattering disturbance center. The change in dielectric constant at the scattering disturbance center caused by a change in the quantity of cargo is set as the unit disturbance intensity (valued as 1, dimensionless). The unit disturbance intensity is used as the input excitation. The complex changes in the forward and backscattering coefficients of the radio frequency received by the radio frequency receiving sensor in each multidimensional sensing unit are calculated using a ray tracing propagation model. The real and imaginary parts of the complex changes of each multidimensional sensing unit are arranged sequentially according to the multidimensional sensing unit numbering order to form the... The theoretical response vector of the radio frequency scattering pattern change signal corresponding to a deeply obscured cargo location is denoted as r. , dimension .

[0050] The positive mapping relationship matrix construction module includes: The local weight distribution change signal theoretical response vector generation module is used to construct a weight distribution transfer matrix for local weight distribution change signals of multi-dimensional signals, based on the cross-sectional dimensions, elastic modulus, and connection methods between components of the rack load-bearing structure.

[0051] The module for generating the theoretical response vector of the local weight distribution change signal includes the following steps: The rack support structure is discretized into multiple structural units. Degrees of freedom are defined at the nodes of each structural unit. These degrees of freedom include the linear displacement of the node along each coordinate axis and the angular displacement around each coordinate axis in the rack's three-dimensional coordinate system. The element stiffness matrix of each structural unit is established based on its cross-sectional dimensions and elastic modulus. The element stiffness matrices are assembled into the overall stiffness matrix of the rack support structure according to the connection method between components. The overall stiffness matrix is ​​inverted and multiplied by the constraint force transformation matrix corresponding to the constraint degrees of freedom to form the weight distribution transfer matrix from the load vector of each node to the support reaction force vector of each constraint node.

[0052] The weight distribution transfer matrix is ​​used as the signal transmission path model for the local weight distribution change signal. The spatial coordinates of the i-th deeply obscured cargo location are used as the weight input node. The weight increment corresponding to the change in the number of goods per unit is used as the non-zero element in the input vector. The change in support reaction force at the installation position coordinates of each multi-dimensional sensing unit is calculated through the weight distribution transfer matrix. All the changes in support reaction force are arranged in order to form the theoretical response vector of the local weight distribution change signal.

[0053] For example, for a local weight distribution change signal, the rack support structure is discretized into multiple structural elements using the finite element method, and six degrees of freedom are defined at the nodes of each structural element (along...). axis, axis, Linear displacement of the axis and its surroundings axis, axis, (Angular displacement of the axis) Based on the cross-sectional dimensions and elastic modulus of each structural unit, establish the element stiffness matrix of each structural unit. According to the connection method between components, assemble the element stiffness matrices according to the nodal degrees of freedom number to form the overall stiffness matrix K of the rack load-bearing structure. After inverting the overall stiffness matrix K, multiply it with the constraint force transformation matrix B corresponding to the constraint degrees of freedom to form the weight distribution transfer matrix from the load vector of each nodal to the support reaction force vector of each constraint node: ; with the first The structural nodes corresponding to the spatial coordinates of each deeply obscured cargo location are used as weight input nodes, with the weight increment corresponding to the unit change in cargo quantity as the value of that node. The axial load component is calculated using the weight distribution transfer matrix T. The change in support reaction force at each pressure sensor measuring point will The changes in the support reaction force are arranged in order of the pressure sensor numbers, forming the [number]. The theoretical response vector of the local weight distribution change signal corresponding to a deeply obscured cargo location is denoted as w. , dimension .

[0054] The theoretical signal response vectors of each dimension of the same deep-shaded cargo location are integrated and arranged to form a multi-dimensional joint theoretical signal response vector. The multi-dimensional joint theoretical signal response vectors of all deep-shaded cargo locations are summarized, and a positive mapping relationship matrix between each deep-shaded cargo location and the multi-dimensional joint theoretical signal response vector is constructed. For example, for the first For each deeply obscured storage location, the four types of signal theoretical response vectors are integrated and arranged in the same splicing order as the multidimensional joint signal feature vector to form a multidimensional joint theoretical signal response vector: The multidimensional joint theoretical signal response vector is obtained through offline pre-calculation before system operation; its dimension is... The dimension is completely consistent with that of the multidimensional joint signal feature vector x. For all Repeat the above calculation for each deeply obscured storage location. A multidimensional joint theoretical signal response vector Arrange the columns to construct a positive mapping matrix H with dimension . Taking the aforementioned fast-moving consumer goods warehouse as an example, the number of deeply concealed storage locations... If the dimension of the multidimensional joint signal eigenvector is 160, then the dimension of the positive mapping matrix H is... Each column in the matrix corresponds to the theoretical multidimensional perception response of a deeply occluded cargo location under the excitation of a unit change in the quantity of goods.

[0055] The indirect perception mapping model forming module is used to introduce a Bayesian inference framework based on the positive mapping relationship matrix. The historical replenishment frequency of each deeply obscured cargo location is set as the prior probability. The likelihood function is constructed by the residual between the multidimensional joint theoretical signal response vector and the multidimensional joint signal feature vector actually collected by the multidimensional sensing unit. This forms an indirect perception mapping model with the multidimensional joint signal feature vector as input and the joint posterior probability distribution of the change in the quantity of goods in each deeply obscured cargo location as output.

[0056] For example, a Bayesian inference framework is introduced based on the positive mapping relation matrix H. For the... The change in the quantity of goods at each candidate deeply obscured cargo location is: The prior probability of a change in cargo quantity at a given location is calculated by dividing the frequency of such changes by the total frequency of changes at all locations during the same statistical period. If the system is initially put into operation and has no historical data, the prior probabilities of all storage locations can be initialized to a uniform distribution, i.e. The system will be updated to a prior probability based on actual frequency after accumulating enough operational data.

[0057] Likelihood function describes the first... The change in the quantity of goods at each candidate deeply obscured cargo location is: Under the given conditions, the probability of observing the multidimensional joint signal eigenvector x is modeled in a multidimensional Gaussian form: .

[0058] in, The first in the positive mapping matrix H The corresponding multidimensional joint theoretical signal response vector of the column, The overall noise variance of the sensor is determined through repeated experiments during the system calibration phase. For example, the calibration result is... A value of 0.01 indicates that the sensor observation residuals follow a Gaussian distribution with a standard deviation of 0.1. Let be the normalization constant. According to Bayes' theorem, the th... The change in the quantity of goods in each deeply obscured storage location is: The posterior probability is: .

[0059] in, The prior distribution of changes in the quantity of goods can be set according to the actual replenishment habits of the warehouse. For example, if a single replenishment is usually 1 to 5 items, it can be set as follows: This range is set to uniform distribution. For all... The posterior probability is calculated by combining the changes in the quantity of each candidate cargo location with the total quantity of all candidate cargo locations, and then normalized to form a joint posterior probability distribution, which is the output of the indirect perception mapping model.

[0060] The cargo quantity change determination module is used to input the multi-dimensional joint signal feature vector into the indirect perception mapping model, output the joint posterior probability distribution of the cargo quantity change of the target deep occlusion cargo location, obtain the cargo quantity change with the highest probability in the joint posterior probability distribution, and determine the target deep occlusion cargo location where the cargo quantity change has occurred. For example, when the system detects a cargo removal or replenishment action during operation (triggered by any sensor signal sudden change exceeding a preset threshold), the multi-dimensional joint signal feature vector x, collected and constructed in real time, is input into a pre-established indirect perception mapping model to calculate the joint posterior probability distribution of all candidate cargo locations and changes in the quantity of candidate cargo. The combination with the highest posterior probability value is then selected from the joint posterior probability distribution. This refers to the most likely target deeply obscured cargo location number involved in this operation. Changes in the corresponding quantity of goods A positive change in the quantity of goods indicates that the action is to add goods, while a negative change indicates that the action is to remove goods.

[0061] Continuing with the example of the aforementioned fast-moving consumer goods warehouse, at 10:32 AM one morning, a picker retrieved three bottled beverages from the eighth horizontal storage location (numbered "R3-C3-H08", a deeply obscured storage location) on the third shelf of the third column. Sixteen multi-dimensional sensing units simultaneously detected the signal abrupt change. The system completed signal acquisition and feature extraction within approximately 200 milliseconds, outputting a 160-dimensional multi-dimensional joint signal feature vector x. This x was input into the indirect sensing mapping model. The model calculated the joint posterior probability among all 480 deeply obscured storage locations and candidate change combinations. The combination with the highest probability corresponded to storage location number "R3-C3-H08" with a change in the quantity of goods of -3, perfectly matching the actual situation. The inference took approximately 50 milliseconds, meeting the requirements for real-time processing.

[0062] The inventory data update and recording module is used to update the baseline inventory data of the target deep-covered storage location in real time based on the change in the quantity of goods in the target deep-covered storage location, and record the timestamp of this change, storage location number and change in the quantity of goods, forming an inventory change detail.

[0063] For example, based on the determined target deep-coverage storage location number and the change in the quantity of goods, the baseline inventory data of the storage location stored in the system database is updated in real time. Specifically, the current inventory quantity of the storage location is added to the change in the quantity of goods (if the change is negative when goods are taken out, the inventory quantity decreases; if the change is positive when goods are replenished, the inventory quantity increases). The updated inventory quantity is the latest inventory status of the storage location.

[0064] Simultaneously, the system automatically generates an inventory change record, which includes three essential fields: the timestamp of the action (accurate to milliseconds, e.g., "20XX-10-01 10:32:15.382"), the location number of the target deeply obscured storage location (e.g., "R3-C3-H08"), and the change in the quantity of goods (e.g., -3, indicating that three items were retrieved). This record is appended to the inventory change detail log table in the system database. The inventory change detail log table continuously accumulates all change records in timestamp order, forming a complete inventory inflow and outflow ledger. The inventory change detail log can be queried and statistically analyzed by storage location number, time range, or product category at any time by warehouse management personnel. It can also be periodically compared and verified with physical inventory count results, providing continuously updated and reliable data support for warehouse replenishment planning, safety stock early warning, and inventory turnover analysis.

[0065] Finally, it should be noted that although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent warehouse cargo identification and inventory system, characterized in that, The system includes: an initial inventory data generation module, a multi-dimensional signal acquisition and processing module, an indirect perception mapping model establishment module, a goods quantity change determination module, and an inventory data update and recording module, with each module communicating with each other in sequence. The initial inventory data generation module is used to initially identify the goods in each location on the shelf, record the type and quantity of goods in each location, and generate the baseline inventory data for each location. A multi-dimensional signal acquisition and processing module is used to deploy multi-dimensional sensing units on the load-bearing structure of the shelf. The multi-dimensional sensing units synchronously acquire multi-dimensional signals caused by the removal or replenishment of goods in real time. The multi-dimensional signals include stress change signals, vibration signals, radio frequency scattering pattern change signals, and local weight distribution change signals. After synchronizing and aligning the multi-dimensional signals, the module extracts signal feature components representing the current removal or replenishment of goods from each dimension signal, and merges the signal feature components of each dimension signal to form a multi-dimensional joint signal feature vector. The indirect perception mapping model building module is used to pre-build an indirect perception mapping model for deeply obscured storage locations based on the structural parameters of the shelving and the spatial positional relationship of each storage location relative to the multi-dimensional perception unit. The cargo quantity change determination module is used to input the multi-dimensional joint signal feature vector into the indirect perception mapping model, output the joint posterior probability distribution of the cargo quantity change of the target deep occlusion cargo location, obtain the cargo quantity change with the highest probability in the joint posterior probability distribution, and determine the target deep occlusion cargo location where the cargo quantity change has occurred. The inventory data update and recording module is used to update the baseline inventory data of the target deep-covered storage location in real time based on the change in the quantity of goods in the target deep-covered storage location, and record the timestamp of this change, storage location number and change in the quantity of goods, forming an inventory change detail.

2. The intelligent warehousing cargo identification and inventory system according to claim 1, characterized in that, The indirect perception mapping model establishment module includes: The shelving parameter acquisition and coordinate determination module is used to acquire the elastic modulus, material density, structural damping coefficient, cross-sectional dimensions and connection methods between components of the shelving load-bearing structure, establish a three-dimensional coordinate system for the shelving, and determine the spatial coordinates of each storage location and the installation position coordinates of each multi-dimensional sensing unit in the three-dimensional coordinate system of the shelving. The forward mapping relationship matrix construction module is used to establish a signal transmission path model for multi-dimensional signals, using the spatial coordinates of each cargo location as the signal excitation point and the installation position coordinates of each multi-dimensional sensing unit as the signal receiving point. Taking the change in the quantity of goods per unit as the benchmark excitation input, it calculates the theoretical signal response vectors of each dimension corresponding to each deeply obscured cargo location according to the signal transmission path model. It integrates and arranges the theoretical signal response vectors of each dimension of the same deeply obscured cargo location to form a multi-dimensional joint theoretical signal response vector. It summarizes the multi-dimensional joint theoretical signal response vectors of all deeply obscured cargo locations and constructs a forward mapping relationship matrix between each deeply obscured cargo location and the multi-dimensional joint theoretical signal response vector. The indirect perception mapping model forming module is used to introduce a Bayesian inference framework based on the positive mapping relationship matrix. The historical replenishment frequency of each deeply obscured cargo location is set as the prior probability. The likelihood function is constructed by the residual between the multidimensional joint theoretical signal response vector and the multidimensional joint signal feature vector actually collected by the multidimensional sensing unit. This forms an indirect perception mapping model with the multidimensional joint signal feature vector as input and the joint posterior probability distribution of the change in the quantity of goods in each deeply obscured cargo location as output.

3. The intelligent warehousing goods identification and inventory system according to claim 2, characterized in that, The positive mapping relationship matrix construction module includes: The stress change signal theoretical response vector generation module is used to construct a finite element static mechanical model for stress change signals of multi-dimensional signals based on the elastic modulus, cross-sectional dimensions, and connection methods of each component of the rack bearing structure. The finite element static mechanical model is used as the signal transmission path model of the stress change signal. In the rack three-dimensional coordinate system, the spatial coordinates of the i-th deep-shaded storage location are used as the load application point. The standard weight corresponding to the change in the number of goods per unit is used as the concentrated force load value and applied along the direction of gravity. The installation position coordinates of the multi-dimensional sensing unit are used as the stress observation point. The stress increment value caused by the concentrated force load at each stress observation point is solved through the finite element static mechanical model. The stress increment values ​​of all stress observation points are arranged in order to form the stress change signal theoretical response vector.

4. The intelligent warehousing cargo identification and inventory system according to claim 2, characterized in that, The positive mapping relationship matrix construction module includes: The vibration signal theoretical response vector generation module is used to construct transient dynamic wave equations for multi-dimensional vibration signals based on the material density, structural damping coefficient, and elastic modulus of each component of the rack bearing structure. These transient dynamic wave equations serve as the signal transmission path model for the vibration signal. The spatial coordinates of the i-th deeply obscured cargo location are used as the vibration excitation source point. The time function of the force generated by the cargo removal or replenishment action corresponding to the unit change in cargo quantity is used as the excitation boundary condition. The transient dynamic wave equations are solved to obtain the vibration displacement versus time curve at the installation position coordinates of each multi-dimensional sensing unit. Three characteristic quantities—peak amplitude, dominant frequency energy ratio, and vibration decay time constant—are extracted from each vibration displacement versus time curve. These three characteristic quantities of all multi-dimensional sensing units are then sequentially concatenated to form the vibration signal theoretical response vector.

5. The intelligent warehouse cargo identification and inventory system according to claim 4, characterized in that, The vibration signal theoretical response vector generation module includes the following steps: The maximum absolute value of the vibration displacement in the vibration displacement versus time curve is determined as the peak amplitude. The power spectral density distribution of the vibration signal is obtained by performing a Fourier transform on the vibration displacement versus time curve. The frequency corresponding to the maximum power spectral density value is identified from the power spectral density distribution as the dominant frequency. The dominant frequency energy value with the dominant frequency as the center and the bandwidth as a preset multiple is calculated. The ratio of the dominant frequency energy value to the total energy value of the power spectral density distribution is determined as the dominant frequency energy proportion. Extract the time interval from the vibration initiation moment to the vibration displacement decaying to a preset proportional threshold from the vibration displacement change curve over time. Fit the envelope of the vibration displacement change curve over time within the time interval with an exponential decay function. Determine the time constant in the fitted exponential decay function as the vibration decay time constant.

6. The intelligent warehousing cargo identification and inventory system according to claim 2, characterized in that, The positive mapping relationship matrix construction module includes: The radio frequency scattering pattern change signal theoretical response vector generation module is used to establish a ray tracing propagation model for the radio frequency scattering pattern change signal of multi-dimensional signals based on the geometric dimensions and material electromagnetic parameters of the shelf bearing structure. The ray tracing propagation model is used as the signal transmission path model of the radio frequency scattering pattern change signal. The spatial coordinates of the i-th deep-shaded cargo location are used as the scattering disturbance center. The change in dielectric constant at the scattering disturbance center caused by the change in the quantity of goods is set as the unit disturbance intensity. Using the unit disturbance intensity as input excitation, the complex changes in the radio frequency forward scattering coefficient and back scattering coefficient received at the installation position coordinates of each multidimensional sensing unit are calculated through the ray tracing propagation model. The real and imaginary parts of the complex changes of each multidimensional sensing unit are arranged in order to form the theoretical response vector of the radio frequency scattering pattern change signal.

7. The intelligent warehousing cargo identification and inventory system according to claim 6, characterized in that, The theoretical response vector generation module for the radio frequency scattering pattern change signal includes the following steps: A three-dimensional spatial geometric model is established based on the geometric dimensions of the rack support structure. In the three-dimensional spatial geometric model, the position coordinates of the radio frequency transmitter, the operating frequency and transmission power of the radio frequency transmitter, the installation position coordinates of the radio frequency receiving sensor in each multi-dimensional sensing unit, and the spatial coordinates of each deeply obscured storage location are set. Based on the material electromagnetic parameters of the rack support structure, the relative permittivity and conductivity of each component in the three-dimensional spatial geometric model are set. The three-dimensional spatial geometric model is divided into multiple discrete reflecting surfaces and diffraction edges, and a ray tracing propagation model including direct path, reflection path and diffraction path is constructed.

8. The intelligent warehousing cargo identification and inventory system according to claim 2, characterized in that, The positive mapping relationship matrix construction module includes: The local weight distribution change signal theoretical response vector generation module is used to construct a weight distribution transfer matrix for local weight distribution change signals of multi-dimensional signals based on the cross-sectional dimensions, elastic modulus, and connection methods of each component of the rack bearing structure. The weight distribution transfer matrix is ​​used as the signal transmission path model of the local weight distribution change signal. The spatial coordinates of the i-th deeply obscured cargo location are used as the weight input node, and the weight increment corresponding to the change in the number of goods per unit is used as the non-zero element in the input vector. The change in support reaction force at the installation position coordinates of each multi-dimensional sensing unit is calculated through the weight distribution transfer matrix. All the changes in support reaction force are arranged in order to form the local weight distribution change signal theoretical response vector.

9. The intelligent warehousing goods identification and inventory system according to claim 8, characterized in that, The module for generating the theoretical response vector of the local weight distribution change signal includes the following steps: The rack support structure is discretized into multiple structural units. Degrees of freedom are defined at the nodes of each structural unit. These degrees of freedom include the linear displacement of the node along each coordinate axis and the angular displacement around each coordinate axis in the rack's three-dimensional coordinate system. The element stiffness matrix of each structural unit is established based on its cross-sectional dimensions and elastic modulus. The element stiffness matrices are assembled into the overall stiffness matrix of the rack support structure according to the connection method between components. The overall stiffness matrix is ​​inverted and multiplied by the constraint force transformation matrix corresponding to the constraint degrees of freedom to form the weight distribution transfer matrix from the load vector of each node to the support reaction force vector of each constraint node.

Images