Control area certificate circulation management method and system based on multi-modal identity verification

Abstract

The application discloses a control area certificate circulation control method and system based on multi-modal identity verification, relates to the technical field of intelligent certificate control, and comprises the following steps: S1, real-time acquisition of certificate circulation data, and pre-processing of the certificate circulation data; S2, screening of candidate storage locations, selection of a target storage location position according to the product of the storage location fitting evaluation value and the entry continuous evaluation value, generation of a certificate taking-out mark, a certificate homing mark and a to-be-checked circulation mark; S3, construction of a certificate storage location binding matrix, generation of a certificate original position closed loop record, a certificate misplacement return record and a certificate out-of-position circulation record; and S4, construction of a certificate circulation judgment control feature sequence and input of the sequence into a double-branch gate control cycle unit judgment control model, output of a circulation control state, and execution of a corresponding control instruction. The application solves the problems that in the existing control area certificate circulation control, identity verification and certificate storage location mapping are difficult to correspond, the taking-out and homing closed loop state is difficult to judge, and the abnormal out-of-position circulation is difficult to control.

Description

Technical Field

[0001] This invention relates to the field of intelligent document management technology, specifically to a method and system for managing the circulation of documents in a control area based on multimodal identity verification. Background Technology

[0002] With the continuous development of smart airport, smart security, and intelligent access control technologies, related technologies for personnel identification, document information management, equipment linkage control, and access behavior supervision are constantly being improved. Existing technologies have already emerged that utilize methods such as identity information collection, facial recognition, terminal recognition, and document scheduling control to process personnel identity verification, attendance records, document management, and equipment collaboration.

[0003] For example, application CN111798582B provides a method and system for real-name safety management of construction site workers. The method includes the following steps: an information collection step, which collects the worker's identity information and stores it in the project database accordingly. The identity information includes project data, identity identification data, document image data, facial recognition data, and terminal identification data; an attendance recording step, which collects the terminal holder's facial image and terminal identification after the terminal enters the attendance area; queries the project database for matching data based on the terminal identification and the facial image; if matching data is found in the project database, an attendance record for the terminal holder's identity information is added, indicating a normal status.

[0004] For example, application CN115171267B discloses a control method, device, system, and storage medium for document management. The method includes: obtaining a document retrieval request and obtaining the device for storing the document to be retrieved and the target document issuing device corresponding to the document to be retrieved; when the device for storing the document to be retrieved and the target document issuing device are the same device, generating a first prompt message to indicate that the document is on the current device; when the device for storing the document to be retrieved and the target document issuing device are different devices, and the distance difference between the locations of the device for storing the document to be retrieved and the target document issuing device is less than or equal to a preset range, generating a second prompt message to indicate whether to change the target document issuing device; when the device for storing the document to be retrieved and the target document issuing device are different devices, and the distance difference between the locations of the device for storing the document to be retrieved and the target document issuing device is greater than a preset range, generating a document scheduling instruction to schedule the document to be retrieved to the target document issuing device.

[0005] However, existing technologies primarily focus on personnel identification, terminal identity matching, document scheduling prompts, or document transfer control between devices. They have not yet established a unified transfer control mechanism covering the entire process of documents within the controlled area: verification, issuance, storage location departure, external circulation, return, closed-loop judgment, and anomaly control. Especially in controlled area document management scenarios, document transfer involves not only whether identity verification is successful, but also whether the physical storage location corresponding to the document's RFID tag value can be accurately mapped, whether the issuance and return processes can form a reliable closed loop, whether misplaced returns and abnormal external circulation can be identified in a timely manner, and whether control commands such as locking, warnings, and verification can be further output based on the identification results. Existing technologies still lack effective technical solutions for the dynamic binding relationship between the physical location of documents and their electronic circulation status, closed-loop judgment of document circulation based on multimodal identity verification, and model-based control execution for abnormal states, making it difficult to meet the needs of refined, continuous, and intelligent transfer control for documents in controlled areas.

[0006] Therefore, in response to the above problems, there is an urgent need for a control method and system for the circulation and management of control zone documents based on multimodal identity verification. Summary of the Invention

[0007] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method and system for controlling the circulation of control zone documents based on multimodal identity verification. This solves the problems in existing control zone document circulation management, such as the difficulty in mapping identity verification to document storage location, the difficulty in judging the closed-loop status of document retrieval and return, and the difficulty in controlling abnormal external circulation.

[0008] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a control method for the circulation and management of identification documents in a control area based on multimodal identity verification, comprising: S1, real-time acquisition of identification document circulation data, preprocessing the identification document circulation data, and outputting the preprocessed identification document circulation data; S2, generation of verification identification document circulation segments based on the preprocessed identification document circulation data, and selection of candidate storage locations from the segments; selection of target storage location based on the product of the storage location fit evaluation value and the continuous entry evaluation value corresponding to the candidate storage location; generation of identification document retrieval marker, identification document return marker, and circulation marker to be verified based on the circulation event timing judgment result corresponding to the target storage location; S3, reading the identification document retrieval marker, identification document return marker, and circulation marker to be verified. The document is marked with a return tag and a transfer tag to be checked. A document storage location binding matrix is ​​constructed, and transfer return state pairs are extracted. When the target storage location corresponding to the departure state and the return state is consistent, a document in-situ closed-loop record is generated. When the target storage location corresponding to the departure state and the return state is inconsistent, a document misplacement return record is generated. When no transfer return state pair is formed and the departure delay time difference exceeds the external judgment threshold, a document external transfer record is generated. S4, the document in-situ closed-loop record, document misplacement return record, document external transfer record and transfer tag to be checked are read. A document transfer judgment and control feature sequence is constructed and input into the dual-branch gating loop unit judgment and control model. The transfer control status is output and the corresponding control command is executed.

[0009] Furthermore, the specific steps for real-time collection of document circulation data and preprocessing of this data are as follows: Real-time collection of document circulation data from the control area, including verification time values, face image acquisition time values, fingerprint image acquisition time values, password input end time values, document RFID tag identification values, document RFID read / write time values, document RFID signal strength values, horizontal position values ​​corresponding to storage location identifiers, vertical position values ​​corresponding to storage location identifiers, depth position values ​​corresponding to storage location identifiers, photoelectric trigger time values ​​at storage location entrances, photoelectric trigger time values ​​inside storage locations, storage location read / write antenna signal strength values, and document... The data includes the time values ​​for leaving the storage location, entering the storage location, the lateral displacement of the robotic arm, the longitudinal displacement of the robotic arm, the depth displacement of the robotic arm, and the time value for the cabinet door opening. For the collected document circulation data, a network time protocol synchronization algorithm is used to perform multi-source timestamp alignment; an event sequence verification algorithm is used to correct the chronological relationship of circulation events; a box plot outlier detection algorithm is used to identify and remove abnormal data; a Hample filter algorithm is used to suppress discrete mutation document circulation records; and a minimum-maximum normalization algorithm is used to unify the numerical scale, outputting the preprocessed document circulation data.

[0010] Further, the specific steps for generating verification document flow segments based on the preprocessed document flow data and selecting candidate storage locations from the segments are as follows: Read the preprocessed control area document flow data; for each document RFID read / write time value, extract the verification events corresponding to the most recent verification time value before the document RFID read / write time value, including the face image acquisition time value, fingerprint image acquisition time value, and password input end time value, as well as the corresponding flow events, to generate verification document flow segments; based on each verification document flow segment, for each document RFID tag identifier value, first read the robot arm's lateral displacement value, longitudinal displacement value, and depth position corresponding to the current document RFID tag identifier value. The system then reads the lateral, longitudinal, and depth position values ​​corresponding to all storage location identifiers. It calculates the absolute values ​​of the differences between the lateral displacement value of the robotic arm and the lateral position value corresponding to each storage location identifier, the absolute values ​​of the differences between the longitudinal displacement value of the robotic arm and the longitudinal position value corresponding to each storage location identifier, and the absolute values ​​of the differences between the depth displacement value of the robotic arm and the depth position value corresponding to each storage location identifier. Storage locations with lateral, longitudinal, and depth position differences all less than the corresponding position tolerance threshold are retained to form a preliminary set of storage locations. The system then reads the storage location read / write antenna signal strength values ​​corresponding to each storage location in the preliminary set, and retains storage locations with read / write antenna signal strength values ​​greater than the antenna response threshold as candidate storage locations.

[0011] Further, the specific steps for selecting the target storage location based on the product of the storage location fit evaluation value and the continuous entry evaluation value corresponding to the candidate storage location are as follows: For each candidate storage location, add the document RF signal strength value and the storage location read / write antenna signal strength value, add one, and take the natural logarithm to obtain the signal correspondence quantity; square the differences between the lateral displacement value of the robotic arm and the lateral position value corresponding to the storage location identifier, the differences between the longitudinal displacement value of the robotic arm and the longitudinal position value corresponding to the storage location identifier, and the differences between the depth displacement value of the robotic arm and the depth position value corresponding to the storage location identifier, respectively, sum them, take the square root, and add one to obtain the spatial deviation quantity; divide the signal correspondence quantity by the spatial deviation quantity to obtain the storage location fit evaluation value of the document RF tag identifier value relative to the current candidate storage location; compare the storage location entrance photoelectric trigger time value with the storage location's... The absolute values ​​of the differences between the photoelectric trigger times within the storage location, the absolute values ​​of the differences between the photoelectric trigger times within the storage location and the document entry times, and the absolute values ​​of the differences between the document entry times and the document RFID read / write times are added sequentially. The sum is then increased by one and its natural logarithm is taken. The natural logarithm is then increased by one and its reciprocal is taken to obtain the continuous evaluation value of the document RFID tag identification value relative to the current candidate storage location. The storage location fitting evaluation value is multiplied by the continuous evaluation value, and the candidate storage location with the largest product is selected as the target storage location corresponding to the current document RFID tag identification value, generating a document storage location mapping record. When multiple candidate storage locations have the same product of storage location fitting evaluation value and continuous evaluation value, or when the candidate storage location is empty, it is determined that the target storage location cannot be uniquely determined.

[0012] Furthermore, based on the timing judgment results of the circulation events corresponding to the target storage location, a document retrieval mark, a document return mark, and a circulation mark to be verified are generated: based on the time sequence of the document leaving the storage location and the document entering the storage location in each document storage location mapping record, the circulation direction corresponding to the current document RFID tag identification value is determined; when the cabinet door opening time value appears before the document leaving the storage location time value, and the document leaving the storage location time value appears before the document entering the storage location time value, a document retrieval mark is generated; when the photoelectric trigger time value at the storage location entrance, the photoelectric trigger time value inside the storage location, and the document entering the storage location time value appear sequentially, a document return mark is generated; when a complete circulation direction is not formed within the same verification document circulation segment, or the target storage location cannot be uniquely determined, a circulation mark to be verified is generated.

[0013] Further, the specific steps for reading the document retrieval marker, document return marker, and pending verification transfer marker, constructing the document storage location binding matrix, and extracting the transfer return status pairs are as follows: Classify the document retrieval marker, document return marker, and pending verification transfer marker according to the document RFID tag identification value, and extract the target storage location and document RFID read / write time value corresponding to each marker; construct the document storage location binding matrix using the document RFID tag identification value as the row index and the target storage location as the column index; wherein, the matrix element corresponding to the document retrieval marker is written... The matrix element corresponding to the document return status value is written to the departure status value, and the matrix element corresponding to the pending verification flow mark is written to the verification status value. For each document RFID tag identifier value, the corresponding tag records are arranged in ascending order according to the document RFID read / write time value, and the status writing order of the corresponding matrix elements in the document storage binding matrix is ​​updated in sequence. When the departure status value is written first and the return status value is written later in the record corresponding to the same document RFID tag identifier value, the departure status value and the return status value are extracted to form a flow return status pair.

[0014] Furthermore, the specific steps for generating a document in-situ closed-loop record when the target storage locations corresponding to the out-of-position state and the return-of-position state are consistent, and generating a document misplacement return record when the target storage locations corresponding to the out-of-position state and the return-of-position state are inconsistent, and generating a document external circulation record when no circulation return state pair is formed and the out-of-position retention time difference exceeds the external judgment threshold, are as follows: Based on each circulation return state pair, read the target storage locations corresponding to the out-of-position state value and the return-of-position state value respectively. When the two target storage locations are consistent, generate a document in-situ closed-loop record; when the two target storage locations are inconsistent, read the matrix element where the corresponding return-of-position state value is located, and when the matrix element is... When no verification status value is written, a document misplacement return record is generated; when the matrix element containing the document has been written with a verification status value, the pending verification transfer mark is maintained; for out-of-position status values ​​that have not formed a transfer return status pair, the system unified timestamp after alignment and the document RFID read / write time value corresponding to the out-of-position status value are read, and the time difference between the system unified timestamp after alignment and the document RFID read / write time value corresponding to the out-of-position status value is calculated; when the time difference is greater than the external judgment threshold, a document external transfer record is generated; when the time difference is not greater than the external judgment threshold, or when the matrix element containing the corresponding out-of-position status value has been written with a verification status value, the pending verification transfer mark is maintained.

[0015] Further, the specific steps for reading the original closed-loop record of the document, the record of the document misplacement and return, the record of the document circulating outside, and the circulation mark to be verified, constructing the document circulation judgment and control feature sequence and inputting it into the dual-branch gated loop unit judgment and control model, and outputting the circulation control status are as follows: Read the original closed-loop record of the document, the record of the document misplacement and return, the record of the document circulating outside, and the circulation mark to be verified, classify them according to the document RFID tag identification value, and arrange them in ascending order according to the document RFID read / write time value; encode the original closed-loop record of the document, the record of the document misplacement and return, the record of the document circulating outside, and the circulation mark to be verified into different circulation status values, and extract the corresponding target storage location and storage location fit evaluation. The document circulation judgment and control feature sequence is constructed by estimating the value, the continuous evaluation value of the position, and the time interval between two adjacent records. The document circulation judgment and control feature sequence is input into a pre-trained dual-branch gating loop unit judgment and control model. The first branch reads the circulation status value and the time interval value to extract the circulation status temporal feature. The second branch reads the target storage location, the storage location fitting evaluation value, and the continuous evaluation value of the position to extract the storage location binding feature. The circulation status temporal feature and the storage location binding feature are input into the gating fusion layer to generate the document circulation judgment and control vector and output the circulation control status. The circulation control status includes normal release status, misalignment locking status, external warning status, and manual review status.

[0016] Furthermore, the specific steps for executing the corresponding control instructions are as follows: When the circulation control status is normal release status, maintain normal circulation permissions; when the circulation control status is misalignment lock status, generate a storage location lock instruction and execute the corresponding storage location lock; when the circulation control status is out-of-town warning status, generate a document out-of-town warning instruction and execute the corresponding warning output; when the circulation control status is manual review status, generate a manual review instruction and freeze subsequent circulation confirmation.

[0017] The second aspect of this invention provides a control system for the circulation and management of identification documents in a control area based on multimodal identity verification, comprising: a data acquisition and processing module, a verification storage location mapping module, a closed-loop regression discrimination module, and a judgment and control execution output module, wherein: the data acquisition and processing module is used to acquire identification document circulation data in real time, perform preprocessing on the identification document circulation data, and output the preprocessed identification document circulation data; the verification storage location mapping module is used to generate verification document circulation segments based on the preprocessed identification document circulation data, and filter candidate storage locations from the segments; select a target storage location based on the product of the storage location fit evaluation value and the continuous entry evaluation value corresponding to the candidate storage location; and generate an identification document retrieval mark, an identification document return mark, and a circulation mark to be verified based on the circulation event time sequence discrimination result corresponding to the target storage location. The closed-loop regression discrimination module is used to read the document issuance mark, document return mark, and pending verification circulation mark, construct a document storage location binding matrix, and extract circulation regression state pairs. When the target storage location corresponding to the departure state and the return state is consistent, a document in-situ closed-loop record is generated. When the target storage location corresponding to the departure state and the return state is inconsistent, a document misplacement return record is generated. When no circulation regression state pair is formed and the departure delay time difference exceeds the external judgment threshold, a document external circulation record is generated. The judgment and control execution output module is used to read the document in-situ closed-loop record, document misplacement return record, document external circulation record, and pending verification circulation mark, construct a document circulation judgment and control feature sequence, input it into the dual-branch gating loop unit judgment and control model, output the circulation control status, and execute the corresponding control instructions.

[0018] Beneficial effects The present invention has the following beneficial effects: (1) A control method and system for the circulation of documents in the control area based on multimodal identity verification, by associating the verification time value, the face image acquisition time value, the fingerprint image acquisition time value, the password input end time value with the document radio frequency reading and writing time value, and then combining the robot displacement information, storage trigger information, and cabinet door opening information to generate a verification document circulation segment, the identity verification process and the actual document circulation process can be unified, thereby improving the continuous tracking capability of the document circulation process.

[0019] (2) The control method and system for the circulation of documents in the control area based on multimodal identity verification introduces the document radio frequency signal strength value, the storage space read and write antenna signal strength value, the robot displacement information and the entry timing information around the candidate storage space to construct the storage space fitting evaluation value and the entry continuous evaluation value respectively, and determines the target storage space location by multiplying the two, which can improve the accuracy and stability of the target storage space location determination.

[0020] (3) The control method and system for the circulation of documents in the control area based on multimodal identity verification, by constructing a document storage location binding matrix, recording the writing order of the off-site status value, the return status value, and the verification status value, and extracting the circulation return status pair, can identify the original closed-loop record of the document, the record of the document misplaced return, and the record of the document circulating outside within the same structure, thereby improving the distinguishability and traceability of the document circulation results.

[0021] (4) The control method and system for the circulation of documents in the control area based on multimodal identity verification constructs the document circulation judgment and control feature sequence by recording the original closed loop of documents, the document misplacement return record, the document circulation record outside, and the circulation mark to be verified. Combined with the dual-branch gated loop unit judgment and control model to output the circulation control status, it can realize differentiated control of normal circulation, misplacement return, external stay, and pending verification, and improve the proactive identification and targeted handling of abnormal circulation. Attached Figure Description

[0022] Figure 1 A flowchart of a control area document circulation management method based on multimodal identity verification; Figure 2 This is a structural diagram of a control area document circulation management system based on multimodal identity verification; Figure 3 A three-dimensional mapping determination diagram for candidate storage sites; Figure 4 This is a flowchart for the closed-loop regression judgment of document circulation. Detailed Implementation

[0023] 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.

[0024] Please see Figures 1-4This invention provides a technical solution: a method for controlling the circulation of identification documents in a control area based on multimodal identity verification, comprising: S1, collecting identification document circulation data in real time, performing preprocessing on the identification document circulation data, and outputting the preprocessed identification document circulation data; S2, generating verification identification document circulation segments based on the preprocessed identification document circulation data, and selecting candidate storage locations from the segments; selecting a target storage location based on the product of the storage location fit evaluation value and the continuous entry evaluation value corresponding to the candidate storage location; generating an identification document retrieval marker, an identification document return marker, and a circulation marker to be verified based on the circulation event timing judgment result corresponding to the target storage location; S3, reading the identification document retrieval marker, the identification document return marker, and the... The following steps are performed: S1) Check the circulation markers, construct a document storage location binding matrix, and extract circulation return state pairs. When the target storage location corresponding to the departure state and the return state is consistent, generate a document in-situ closed-loop record. When the target storage location corresponding to the departure state and the return state is inconsistent, generate a document misplacement return record. When no circulation return state pair is formed and the departure delay time difference exceeds the external judgment threshold, generate a document external circulation record. S4) Read the document in-situ closed-loop record, document misplacement return record, document external circulation record, and circulation markers to be checked. Construct a document circulation judgment and control feature sequence and input it into the dual-branch gating loop unit judgment and control model. Output the circulation control status and execute the corresponding control instructions.

[0025] Specifically, the system collects real-time data on document circulation within the control area. This data includes verification time values, facial image acquisition time values, fingerprint image acquisition time values, password input completion time values, document RFID tag identification values, document RFID read / write time values, document RFID signal strength values, the horizontal position value corresponding to the storage location identifier, the vertical position value corresponding to the storage location identifier, the depth position value corresponding to the storage location identifier, the photoelectric trigger time value at the storage location entrance, the photoelectric trigger time value inside the storage location, the storage location read / write antenna signal strength value, the document leaving the storage location, the document entering the storage location, the lateral displacement value of the robotic arm, the vertical displacement value of the robotic arm, the depth displacement value of the robotic arm, and the cabinet door opening time value. The verification time value is taken from the moment an identity verification process is completed and the document enters the storage location. The time records for the document transfer confirmation stage include: the face image acquisition time value (taken from the time record of the face acquisition device completing the frame capture of the current operator's face image); the fingerprint image acquisition time value (taken from the time record of the fingerprint acquisition device completing the fingerprint image capture of the current operator); the password input completion time value (taken from the time record of the password input interface completing the last digit of the password and triggering the submission verification); the document RFID tag identification value (taken from the unique tag identification content returned after the internal RFID tag of the document is activated); the document RFID read / write time value (taken from the read / write time record generated when the RFID read / write antenna completes the current document tag recognition); and the document RFID signal strength value (taken from the signal strength detection result fed back by the current document RFID tag during the current read / write process). The horizontal position corresponding to the storage location identifier is also included. The vertical position value and depth position value corresponding to the storage location identifier are the three-dimensional spatial position parameters of each storage location pre-calibrated inside the document cabinet in the control area. During calibration, an internal coordinate system is established with the fixed installation reference point of the document cabinet as the origin. Then, the center position of each storage location is mapped to a unique horizontal position value, vertical position value, and depth position value, so that the current position of the robotic arm and the fixed position of the storage location can be directly compared under a unified spatial reference. The photoelectric trigger time value at the storage location entrance is taken from the first photoelectric detection record triggered when the document passes through the storage location entrance detection position. The photoelectric trigger time value inside the storage location is taken from the second photoelectric detection record triggered when the document continues to move into the storage location and passes through the internal detection position. The signal strength value of the storage location read / write antenna is taken from... The tag response intensity detected by the near-field read / write antenna bound to the current storage location at the current moment; the time value of the document leaving the storage location is taken from the time record of the edge change of the corresponding departure detection signal after the document leaves the original storage location; the time value of the document entering the storage location is taken from the entry confirmation time record generated when the document is sent into the target storage location and reaches the set entry position; the lateral displacement value, longitudinal displacement value, and depth displacement value of the robot arm are taken from the real-time position feedback results of the robot arm on the lateral guide rail, longitudinal guide rail, and depth execution unit, respectively. The position feedback results are converted by the encoder pulse accumulation value. After conversion, the same position reference is used with the storage location space coordinates to ensure that the deviation between the current position of the robot arm and the candidate storage location position can be directly quantified;The cabinet door opening time value is taken from the event time record generated when the cabinet door status sensor detects the cabinet door switching from a closed state to an open state. This data is used to constrain the sequential relationship between manual contact actions and document removal actions, preventing abnormal document transfer detection without the access control system being open. For the collected document transfer data, a network time protocol synchronization algorithm is used to perform multi-source timestamp alignment processing. During alignment, the local times generated by the face capture device, fingerprint capture device, password input terminal, RFID reader / writer, photoelectric detection device, robotic arm position feedback unit, and cabinet door status detection device are first mapped to the same standard clock reference. Then, the clock offset of each acquisition source relative to the standard clock reference is calculated, and the corresponding clock offset is used to correct the verification time value, face image acquisition time value, fingerprint image acquisition time value, password input end time value, and document tracking time value. The timing values ​​of frequency read / write, photoelectric trigger at the storage location entrance, photoelectric trigger inside the storage location, document leaving the storage location, document entering the storage location, and cabinet door opening are used to bring the time records generated by different acquisition links into a unified time reference frame. The technical principle of this processing is that the clocks inside different acquisition devices have cumulative drift. If unified time synchronization is not completed first, when extracting the document transfer segment based on time proximity, the same real transfer action will be split into multiple records, which will affect the accurate binding between the verification event and the document RF read / write event. An event sequence verification algorithm is used to correct the order of events in the workflow. During the correction process, event records corresponding to the same RFID tag value are merged, and then the order of events—the cabinet door opening time, the document leaving the storage space time, the storage space entrance photoelectric trigger time, the storage space internal photoelectric trigger time, the document entering the storage space time, and the document RFID read / write time—is compared with the preset workflow logic. If a document leaving the storage space time has occurred even though the cabinet door has not been opened, or the document entering the storage space time is earlier than the storage space internal photoelectric trigger time, or the storage space internal photoelectric trigger time is earlier than the storage space internal photoelectric trigger time, the correction is applied. When a time value is earlier than the photoelectric trigger time value at the storage location entrance, or when the RFID read / write time value of the document deviates significantly from the adjacent time interval of the entry action, the corresponding record is marked as an abnormal sequence record. The sequence is then corrected based on the relative positions of other sensor records within the same time slice. The technical principle behind this process is that each detection point has a clear physical path during document circulation. Once the physical path is determined, the order of events is irreversible. Therefore, physical path constraints can be used to correct time sequence errors caused by communication buffering, reporting delays, and write queuing. A box plot outlier detection algorithm is used to perform abnormal data identification and removal.During the identification process, the detection targets are the RF signal strength of the document, the signal strength of the storage space read / write antenna, the lateral displacement of the robotic arm, the longitudinal displacement of the robotic arm, and the depth displacement of the robotic arm. Within a fixed-length sliding sampling window, the upper quartile, lower quartile, and interquartile range are calculated. Data records exceeding the box plot discrimination interval are identified as abnormal records and removed. After removal, document circulation records that meet the time continuity requirements are retained. The technical principle of this process is that the RF signal strength of the document is prone to sudden highs and lows due to obstruction, instantaneous interference, and posture jitter. The lateral displacement, longitudinal displacement, and depth displacement of the robotic arm may also show significant jumps under encoder jitter and instantaneous pulse loss conditions. The box plot outlier detection algorithm can use the statistical distribution boundary to identify outliers that deviate from the main distribution, reducing the interference of abnormal sampling values ​​on the storage space mapping judgment. The Hample filter algorithm is used to suppress discrete mutations in document circulation records. During the suppression process, based on the time-series records corresponding to the RFID tag values ​​of each document, local neighborhood windows are constructed around the document's RFID signal strength, storage read / write antenna signal strength, lateral displacement of the robotic arm, longitudinal displacement of the robotic arm, and depth displacement of the robotic arm. Within each local neighborhood window, the median and the absolute deviation of the median are calculated. Records that deviate from the median by more than a set multiple of the absolute deviation of the median are replaced with the results corresponding to the local median, thereby suppressing single-point discrete mutations. The technical principle of this processing is that document storage and retrieval is a continuous displacement process, and the change in RFID response is also a gradual process that changes with the degree of spatial proximity. If isolated spikes that do not conform to the laws of continuous physical motion appear between adjacent sampling points, these spikes usually do not represent the true circulation state. The Hample filter algorithm can suppress single-point pulse disturbances while maintaining the true edge change position. A minimum-maximum normalization algorithm is used to perform numerical scale unification processing, outputting preprocessed document circulation data. During unification processing, minimum and maximum values ​​are extracted for the document's radio frequency signal strength, the storage location read / write antenna signal strength, the robotic arm's lateral displacement, longitudinal displacement, and depth displacement, and the lateral, longitudinal, and depth positions corresponding to the storage location markers. These values ​​are then mapped to a unified closed interval, ensuring that different physical quantities, despite their different sources, dimensions, and numerical ranges, can participate in subsequent candidate storage location selection, storage location fit evaluation value calculation, and continuous placement evaluation value calculation at the same numerical scale. The technical principle behind this processing is that if spatial location quantities and signal strength quantities from different sources directly participate in subsequent calculations, variables with large numerical ranges will have a disproportionately dominant influence on the results. The minimum-maximum normalization algorithm compresses all data into a unified scale, making the subsequent judgment results, which involve spatial proximity, radio frequency response strength, and the continuity of the placement process, more stable, ultimately outputting preprocessed document circulation data.

[0026] In this implementation plan, by uniformly constraining the collection, correction, filtering, suppression, and scaling of verification time values, facial image acquisition time values, fingerprint image acquisition time values, password input end time values, document RFID reading and writing time values, storage space entrance photoelectric trigger time values, storage space internal photoelectric trigger time values, document leaving storage space time values, document entering storage space time values, robotic arm lateral displacement value, robotic arm longitudinal displacement value, robotic arm depth displacement value, and cabinet door opening time values, a genuine document circulation process corresponding to the document RFID tag identification value can be ensured. This establishes a data foundation with clear temporal references, unified spatial references, and clear state boundaries. It avoids the problems of incorrect segmentation, incorrect splicing, and incorrect mapping of the same document circulation behavior caused by time drift, out-of-order events, outlier jumps, and amplitude imbalances in multi-source acquisition links. This improves the support stability of preprocessed document circulation data for verifying document circulation fragment construction, candidate storage location determination, target storage location determination, and circulation direction identification. It also makes the subsequently generated document retrieval markers, document return markers, and circulation markers to be verified have higher consistency, traceability, and reliability.

[0027] Specifically, the steps for generating verification document flow segments based on preprocessed document flow data and selecting candidate storage locations from these segments are as follows: Read the preprocessed control area document flow data. For each document RFID read / write time value, extract the verification events corresponding to the most recent verification time value before the document RFID read / write time value, including the face image acquisition time value, fingerprint image acquisition time value, and password input end time value, as well as the corresponding flow events, to generate a verification document flow segment. First, all records are arranged in ascending order of document RFID read / write time values. Then, for each document RFID read / write record, all verification time values ​​are retrieved backwards. The verification time value that occurred earlier than the current document RFID read / write time value and has the smallest time interval is retained as the associated verification time value corresponding to the current document RFID read / write time value. This processing method is used to correlate the current document flow action with the closest identity verification process in time, avoiding the incorrect association of earlier verification events with the current document flow action. The facial image acquisition time value, fingerprint image acquisition time value, and password input end time value corresponding to the associated verification time value are then extracted to form a verification event. At the same time, the document RFID tag identification value, the document RFID signal strength value, the robot arm lateral displacement value, the robot arm longitudinal displacement value, the robot arm depth displacement value, the photoelectric trigger time value at the storage space entrance, the photoelectric trigger time value inside the storage space, the document leaving the storage space, the document entering the storage space, and the cabinet door opening time value are extracted to form a flow event. The verification event and the flow event are then combined according to the same document RFID read / write time value to generate a verified document flow segment. This processing method is used to simultaneously retain identity verification information, document read / write information, robot arm position information, storage space trigger information, and flow sequence information within the same segment, providing a unified data foundation for subsequent storage space determination. Based on the flow segments of each verification document, for each document's RFID tag identifier value, the robot's lateral displacement value, longitudinal displacement value, and depth displacement value corresponding to the current document's RFID tag identifier value are first read. Then, the lateral position value, longitudinal position value, and depth position value corresponding to all storage location identifiers are read. Among them, the robot's lateral displacement value, longitudinal displacement value, and depth displacement value are taken from the position feedback results at the sampling time corresponding to the current document's RFID read / write time value. The lateral position value, longitudinal position value, and depth position value corresponding to all storage location identifiers are taken from the fixed position parameters of each storage location in the pre-established storage location table. This processing method is adopted to make the robot's current position and all storage location fixed positions compare one by one under a unified spatial reference.The absolute values ​​of the differences between the lateral displacement value of the robotic arm and the lateral position value corresponding to each storage location marker, the absolute values ​​of the differences between the longitudinal displacement value of the robotic arm and the longitudinal position value corresponding to each storage location marker, and the absolute values ​​of the differences between the depth displacement value of the robotic arm and the depth position value corresponding to each storage location marker are calculated separately. These calculations yield the lateral position difference, longitudinal position difference, and depth position difference for each storage location. This processing method quantifies the degree of spatial deviation of the robotic arm's current position relative to each storage location in three directions. Storage locations where the lateral position difference, longitudinal position difference, and depth position difference are all less than the corresponding position tolerance threshold are retained, forming a preliminary set of storage locations. The lateral position difference corresponds to the lateral position tolerance threshold, the longitudinal position difference corresponds to the longitudinal position tolerance threshold, and the depth position difference corresponds to the depth position tolerance threshold. The current storage location is retained only when all three position differences meet the corresponding position tolerance threshold. This processing method is used to first eliminate obviously impossible storage locations based on spatial proximity, thus narrowing the subsequent judgment range. Next, the signal strength values ​​of the storage location read / write antennas for each storage location in the initial screening set are read. Storage locations with signal strength values ​​greater than the antenna response threshold are retained as candidate storage locations. During the reading process, the signal strength values ​​of the storage location read / write antennas for each storage location in the initial screening set at the same time are extracted based on the current document RFID read / write time value. These values ​​are then compared with the antenna response threshold one by one, and storage locations with signal strength values ​​greater than the antenna response threshold are retained to obtain candidate storage locations. This processing method is adopted because spatial proximity only indicates that the current document may be close to several storage locations. The signal strength value of the storage location read / write antenna can further characterize whether a valid read / write response is formed between the current document RFID tag value and the specific storage location, thereby improving the accuracy of the candidate storage location screening results.

[0028] In this implementation plan, a stable correspondence is established between the verification time value and the document RFID reading and writing time value. The time values ​​of facial image acquisition, fingerprint image acquisition, password input completion, document RFID signal strength, robotic arm lateral displacement, robotic arm longitudinal displacement, robotic arm depth displacement, photoelectric triggering at the storage location entrance, photoelectric triggering inside the storage location, document leaving the storage location, document entering the storage location, and cabinet door opening are all constrained within the same document flow segment. Combined with the lateral, longitudinal, and depth position values ​​corresponding to the storage location identifier, candidate storage location screening results are formed. This ensures that the identity verification information, spatial location information, and flow sequence information corresponding to the document RFID tag identifier value remain consistent on the same judgment basis. It avoids problems such as mismatch of verification objects, inaccurate storage location range, and overly broad judgment basis for target storage location due to relying solely on a single reading and writing record. This improves the convergence between candidate storage locations and actual flow locations, allowing subsequent storage location alignment evaluation value calculations and continuous entry evaluation value calculations to be based on a more concentrated and reliable data foundation.

[0029] Specifically, the steps for selecting the target storage location based on the product of the storage location fit evaluation value and the continuous entry evaluation value corresponding to the candidate storage location are as follows: For each candidate storage location, the document radio frequency signal strength value and the storage location read / write antenna signal strength value are added together, one is added, and the natural logarithm is taken to obtain the corresponding signal quantity; wherein, the document radio frequency signal strength value is taken as the response result of the current document radio frequency tag identification value in the read / write record corresponding to the current document radio frequency read / write time value, and the storage location read / write antenna signal strength value is taken as the near-field read / write response result of the current candidate storage location under the same document radio frequency read / write time value; in this embodiment, both the document radio frequency signal strength value and the storage location read / write antenna signal strength value are taken from the preprocessed document circulation data, and in The preprocessing stage uses a minimum-maximum normalization algorithm to process the values ​​into non-negative normalized values, all within the closed interval [0,1], thus ensuring that the intensity quantities involved in the calculation remain non-negative. First, an addition process is performed, followed by an increment process, and then a natural logarithmic transformation process. The added term is a smoothing term to avoid zero-value input, preventing the natural logarithmic input from being zero when both the document's RF signal strength value and the storage location's read / write antenna signal strength value are zero, ensuring that the natural logarithmic operation input is always greater than zero. This processing method can compress the amplifying effect of high-amplitude response data on the results, keeping the response differences between different candidate storage locations within a comparable range, while ensuring that the calculation process of the corresponding signal quantities meets the definition requirements of logarithmic operations.The spatial deviation is obtained by summing the squares of the differences between the lateral displacement value of the robotic arm and the lateral position value corresponding to the storage location marker, the differences between the longitudinal displacement value of the robotic arm and the longitudinal position value corresponding to the storage location marker, and the differences between the depth displacement value of the robotic arm and the depth position value corresponding to the storage location marker, and then adding one. The lateral displacement value, longitudinal displacement value, and depth displacement value of the robotic arm are taken from the position feedback results at the sampling time corresponding to the current document RFID reading / writing time value. The lateral position value, longitudinal position value, and depth position value corresponding to the storage location marker are taken from the fixed coordinate parameters of the current candidate storage location in the storage location table. In this embodiment, the lateral displacement value, longitudinal displacement value, depth displacement value, lateral position value corresponding to the storage location marker, longitudinal position value corresponding to the storage location marker, and depth position value corresponding to the storage location marker are all taken from... The preprocessed document flow data is processed into non-negative normalized values ​​using the min-max normalization algorithm in the previous preprocessing stage, so that the position quantities in the three directions participate in the calculation under a unified numerical scale. After calculating the position differences in the three directions, the position differences in the three directions are then squared, summed, and square-rooted, and then incremented by one. The increment is to avoid the spatial deviation being zero when the current position of the robot arm completely coincides with the current candidate storage position, ensuring that the denominator is always greater than zero when the spatial deviation is used as the denominator in subsequent division operations. Using this processing method, the spatial offsets in the three directions can be uniformly converted into a single distance representation value, so that the overall closeness of the current position of the robot arm to the current candidate storage position is quantified, and the denominator has a stable numerical basis in the calculation of the storage position fit evaluation value. Dividing the corresponding signal value by the spatial deviation yields the storage location fit evaluation value of the RFID tag relative to the current candidate storage location. This processing method uses the signal proximity as the numerator and the spatial deviation as the denominator, subjecting the current candidate storage location to both RFID response intensity constraints and three-dimensional spatial position constraints. Since the input of the corresponding signal value consists of a non-negative normalized intensity value, and the natural logarithm of the input is always greater than zero after smoothing, and the spatial deviation is also always greater than zero due to the smoothing term, the calculation process of the storage location fit evaluation value has a clear domain basis. The closer the current candidate storage location is to the actual target storage location, the larger the storage location fit evaluation value.The absolute values ​​of the differences between the photoelectric trigger time value at the storage space entrance and the photoelectric trigger time value inside the storage space, the absolute values ​​of the differences between the photoelectric trigger time value inside the storage space and the time value when the document enters the storage space, and the absolute values ​​of the differences between the time value when the document enters the storage space and the time value when the document's RFID read / write is applied, are added sequentially, and then the sum is increased by one and the natural logarithm is taken. The reciprocal of this sum is then taken, and the document's RFID tag identification value relative to the current candidate storage space is obtained. Here, the photoelectric trigger time value at the storage space entrance represents the moment when the document enters the storage space entrance detection position; the photoelectric trigger time value inside the storage space represents the moment when the document continues to enter the storage space detection position; the document's entry time value represents the moment when the document reaches the entry confirmation position; and the document's RFID read / write time value represents the moment when the current document's RFID tag identification value completes reading and writing recognition. In this embodiment, the photoelectric trigger time value at the storage space entrance, the photoelectric trigger time value inside the storage space, and the document's entry time value are used to represent these three values. The time values ​​for document RFID reading and writing are taken from the preprocessed document circulation data, and multi-source timestamp alignment is performed by the network time protocol time synchronization algorithm in the preprocessing stage, so all time quantities are under a unified time base. The absolute value of the time difference between three adjacent actions is calculated first, and then the accumulation process is performed sequentially. Since the absolute value of each time difference is non-negative, the accumulation result is also non-negative. Then, the increment process, natural logarithm process, increment process, and reciprocal process are performed. Among them, the first increment process is a smoothing term to avoid the natural logarithm input being zero when the accumulation result of the three time difference absolute values ​​is zero, and is used to ensure that the natural logarithm operation input is always greater than zero. The second increment process is used to avoid the result amplification and instability caused by directly taking the reciprocal when the natural logarithm result is zero. Using this processing method, the temporal tightness between each detection action in the bit insertion process can be converted into a continuous quantitative result. The tighter the action chain corresponding to the current candidate storage position, the larger the continuous evaluation value of bit insertion. The storage location fit assessment value and the position continuity assessment value are multiplied together, and the candidate storage location with the largest product is selected as the target storage location corresponding to the current document RFID tag identification value, generating a document storage location mapping record. This processing method incorporates spatial fit and position continuity into the same judgment result, avoiding the determination of the target storage location based solely on a single location proximity relationship or a single temporal continuity relationship, thus ensuring that the target storage location simultaneously meets the spatial matching and action continuity requirements. When generating the document storage location mapping record, the current document RFID tag identification value, target storage location, storage location fit assessment value, position continuity assessment value, and document RFID read / write time value are written into the same mapping record, serving as the basis for subsequent determination of document retrieval mark, document return mark, and pending verification circulation mark.When multiple candidate storage locations have the same product of storage location fitting evaluation value and continuous entry evaluation value, and the candidate storage location is empty, it is determined that the target storage location cannot be uniquely determined. Among them, the fact that multiple candidate storage locations have the same product indicates that the current document RFID tag identification value cannot form a unique mapping result under the current document RFID read / write time value. The fact that the candidate storage location is empty indicates that the current position constraint and read / write response constraint do not support any storage location to enter the final determination range. Therefore, no target storage location is generated, and only the determination result of failing to uniquely determine the target storage location is retained.

[0030] The specific formula for calculating the storage location fit assessment value is as follows: ; In the formula, This represents the storage space matching assessment value between the i-th document transfer record and the j-th candidate storage space. The radio frequency signal strength value of the document in the i-th document transfer record. This represents the storage location read / write antenna signal strength value of the i-th document transfer record relative to the j-th candidate storage location. This represents the lateral displacement value of the robotic arm in the i-th document transfer record. This represents the longitudinal displacement value of the robotic arm in the i-th document transfer record. This represents the depth displacement value of the robotic arm in the i-th document transfer record. This represents the horizontal position value corresponding to the storage location identifier of the j-th candidate storage location. This represents the vertical position value corresponding to the storage location identifier of the j-th candidate storage location. This represents the depth position value corresponding to the storage location identifier of the j-th candidate storage location.

[0031] The specific formula for calculating the continuous evaluation value is as follows: ; In the formula, This represents the continuous evaluation value of the i-th document transfer record. This represents the photoelectric trigger time value at the storage location entrance of the i-th document transfer record. This represents the photoelectric trigger time value inside the storage location of the i-th document transfer record. This represents the time value at which the i-th document transfer record enters the storage location. This represents the time value of the RFID read / write operation for the i-th document transfer record.

[0032] In this implementation plan, by incorporating the strength values ​​of the document's radio frequency signal, the signal strength values ​​of the storage space's read / write antenna, the lateral displacement value of the robotic arm, the longitudinal displacement value of the robotic arm, the depth displacement value of the robotic arm, the photoelectric triggering time value at the storage space entrance, the photoelectric triggering time value inside the storage space, the document's entry time value, and the document's radio frequency read / write time value into the target storage space location determination process, the comparison results of candidate storage spaces corresponding to the current document's radio frequency tag identification value can simultaneously possess spatial fit discrimination capability and entry process convergence discrimination capability. This avoids the target storage space location being prematurely determined based solely on local proximity phenomena and avoids the problem of mapping basis imbalance in candidate storage spaces under conditions of similar response strength. This improves the uniqueness, stability, and reliability of the document storage space mapping record, enabling subsequent document retrieval mark generation, document return mark generation, and pending verification circulation mark generation based on the target storage space location to be established on a more rigorous mapping basis.

[0033] In this embodiment, Table 1 is a candidate storage location mapping determination data table, listing the storage location fit evaluation value, continuous entry evaluation value, and the product of the two for five candidate storage locations corresponding to the same document RFID tag identification value. Specifically: in candidate storage location 1, the storage location fit evaluation value is 0.686, the continuous entry evaluation value is 0.303, and the product value is 0.208; in candidate storage location 2, the storage location fit evaluation value is 2.936, the continuous entry evaluation value is 0.303, and the product value is 0.890; in candidate storage location 3, the storage location fit evaluation value is 0.703, the continuous entry evaluation value is 0.303, and the product value is 0.213; in candidate storage location 4, the storage location fit evaluation value is 0.332, the continuous entry evaluation value is 0.303, and the product value is 0.102; in candidate storage location 5, the storage location fit evaluation value is 0.299, the continuous entry evaluation value is 0.303, and the product value is 0.091. As shown in Table 1, under the same document circulation record, the continuous evaluation values ​​for each candidate storage location remain consistent, while the storage location fit evaluation values ​​differ significantly, resulting in significant differences in the product values ​​corresponding to each candidate storage location. Among them, the product value corresponding to candidate storage location 2 is the largest, at 0.890. Therefore, candidate storage location 2 is determined as the target storage location corresponding to the current document RFID tag identification value. Table 1 intuitively illustrates the process of determining the target storage location through the joint determination of storage location fit evaluation value and continuous evaluation value in this invention. It can reflect the matching differences between different candidate storage locations and provide a basis for generating document retrieval markers, document return markers, and circulation markers to be verified based on the circulation event time sequence discrimination results corresponding to the target storage location.

[0034] Table 1 Candidate Storage Location Mapping Determination Data Table

[0035] like Figure 3As shown in the figure, the spatial mapping relationship of five candidate storage locations corresponding to the same document RFID tag identification value during the candidate screening stage is displayed by combining the three-dimensional storage location distribution with the target storage location determination result. The horizontal coordinate represents the horizontal position value corresponding to the storage location identification, the vertical coordinate represents the vertical position value corresponding to the storage location identification, and the depth coordinate represents the depth position value corresponding to the storage location identification. Each cube represents the position distribution of candidate storage locations 1 to 5 in the three-dimensional storage space, used to characterize the actual storage location corresponding to each candidate storage location. The vertical dashed line and bottom projection point below each candidate storage location represent the mapping relationship of the center position of the corresponding candidate storage location on the base plate plane, used to characterize the correspondence between the center position of the candidate storage location and the base plate reference plane. The cube with darker color and thicker border in the figure represents the target storage location, used to characterize the candidate storage location with the largest product of the storage location fit evaluation value and the continuous placement evaluation value under the current document circulation record. The red highlighted vertical dashed line below the target storage location represents the determination landing point of the center position of the target storage location on the base plate plane, used to characterize the target storage location corresponding to the current document RFID tag identification value. As shown in the figure, the five candidate storage locations are clearly distinguishable in terms of horizontal, vertical, and depth positions. Candidate storage location 2 is located at the center of the position corresponding to the current determination result and is identified as the target storage location due to its darker and bolder color. This reflects that the present invention does not directly determine the storage location based on a single storage location value or a single signal value, but rather first filters multiple candidate storage locations and then combines the comprehensive results of the storage location fit evaluation value and the continuous entry evaluation value to complete the determination of the target storage location. This figure visually illustrates the technical process of candidate storage location screening, target storage location highlighting determination, and spatial mapping landing point formation in the present invention. It can reflect the positional differences between different candidate storage locations and the basis for determining the final target storage location, and provides support for the subsequent generation of document retrieval markers, document return markers, and pending verification circulation markers based on the time sequence discrimination results of the circulation events corresponding to the target storage location.

[0036] Specifically, the steps for generating document retrieval markers, document return markers, and pending verification transfer markers are as follows: Based on the chronological order of document departure and entry times in each document storage location mapping record, the transfer direction corresponding to the current document RFID tag identifier value is determined. First, the document storage location mapping records are arranged in ascending order of document RFID read / write times. Then, for the same document RFID tag identifier value, the corresponding cabinet door opening time, storage location entrance photoelectric trigger time, storage location internal photoelectric trigger time, document departure time, document entry time, and target storage location location are read record by record. Subsequently, the documents are compared... The timing of the document leaving the storage location and the document entering the storage location are determined within the current document verification flow segment. When the document leaving the storage location occurs before the document entering the storage location, the current flow direction is determined to be the direction after leaving the storage location. When the document entering the storage location occurs before the document leaving the storage location, and no complete entry trigger chain corresponding to the current target storage location appears within the current document verification flow segment, the current flow direction is determined to be in a direction-pending state. This processing method is used to distinguish the actual document leaving action from the actual document entering action on the timeline, so that subsequent marker generation is based on a clear action direction. A document retrieval marker is generated when the cabinet door opening time precedes the document leaving the storage location time, and the document leaving the storage location time precedes the document entering the storage location time. The generation of the retrieval marker involves first verifying whether a cabinet door opening time exists within the current document circulation segment, then verifying whether the cabinet door opening time is earlier than the document leaving the storage location time, and finally verifying whether the document leaving the storage location time is earlier than the document entering the storage location time. Only when the above time relationships are simultaneously met is the current document RFID tag value, target storage location, and document RFID read / write time value written into the document retrieval marker. This processing method uses the cabinet door opening action as a prerequisite for manual contact, the document leaving the storage location time as the basis for the document leaving its original storage location, and the document entering the storage location time not yet closing the current circulation process as a retrieval status constraint. This avoids misjudging cabinet disturbance records and short-term false trigger records as genuine retrieval behavior.When the photoelectric trigger time value at the storage location entrance, the photoelectric trigger time value inside the storage location, and the document entry time value appear sequentially, a document return mark is generated. Specifically, when generating the document return mark, the photoelectric trigger time value at the storage location entrance corresponding to the current target storage location is read first, then the photoelectric trigger time value inside the storage location corresponding to the same target storage location is read, and finally the document entry time value corresponding to the current document RFID tag identification value is read. Only when the photoelectric trigger time value at the storage location entrance is earlier than the photoelectric trigger time value inside the storage location, and the photoelectric trigger time value inside the storage location is earlier than the document entry time value, is the current document RFID tag identification value, the target storage location location, and the document RFID read / write time value written into the document return mark. This processing method utilizes the continuous trigger chain of the document moving along the storage location entrance towards the storage location until it reaches the entry position to confirm that the current document has completed the entry action towards the current target storage location, thereby avoiding mistaking a temporary state that is only near the storage location entrance as a true return state. When a complete flow direction is not formed within the same verification document flow segment, or the target storage location cannot be uniquely determined, a flow mark to be verified is generated. A complete flow direction is not formed, including missing document departure time values, missing document entry time values, cabinet door opening time values, and document departure time values ​​within the current verification document flow segment, as well as non-sequential constraints between these values, and non-sequential constraints between the storage location entrance photoelectric trigger time value, the storage location's internal photoelectric trigger time value, and the document's entry time value. The inability to uniquely determine the target storage location includes an empty candidate storage location corresponding to the current document RFID tag value, and the product of the storage location fit evaluation value and the continuous entry evaluation value being the same for multiple candidate storage locations. In any of these cases, the current document RFID tag value, document RFID read / write time value, and the current verification document flow segment identifier are written into the flow mark to be verified. This processing method is used to separate records with incomplete directions, non-unique mappings, and unclosed trigger chains from the confirmed flow results, preserving the basis for subsequent verification and regression analysis.

[0037] In this implementation plan, by incorporating the cabinet door opening time, document leaving the storage location time, storage location entrance photoelectric trigger time, storage location interior photoelectric trigger time, document entering the storage location time, and target storage location location into the same flow direction determination basis, and combining the document storage location mapping record to distinguish the status of the current document RFID tag identification value, the document retrieval mark, document return mark, and pending verification flow mark can correspond to clear flow confirmation boundaries, entry confirmation boundaries, and abnormal pending judgment boundaries, respectively. This avoids flow records that have not completed closure from directly entering the confirmed state, and avoids records with unclear direction and unstable mapping from being mixed into normal flow results. This improves the clarity of the determination basis for subsequent document storage location binding matrix construction, flow return status pair extraction, document in-situ closed-loop record identification, document misplacement return record identification, and document external flow record identification.

[0038] Specifically, the steps for reading document retrieval markers, document return markers, and pending verification transfer markers, constructing a document storage location binding matrix, and extracting transfer return status pairs are as follows: Classify the document retrieval markers, document return markers, and pending verification transfer markers according to the document RFID tag identification value, and extract the target storage location and document RFID read / write time value corresponding to each marker; First, perform a unified read of all document retrieval markers, document return markers, and pending verification transfer markers; then, use the document RFID tag identification value as the classification key to aggregate all marker records, ensuring that all transfer records corresponding to the same document RFID tag identification value enter the same processing sequence; after classification, extract each record individually. The target storage location, document RFID read / write time value, and tag type are retrieved for each tagged record. The target storage location is used to characterize the storage location corresponding to the current circulation record, the document RFID read / write time value is used to characterize the writing order of the current circulation record on the time axis, and the tag type is used to characterize the departure confirmation result, return confirmation result, and verification retention result corresponding to the current circulation record. This processing method is adopted to unify and merge different circulation states corresponding to the same document RFID tag identification value into the same processing basis, so that the subsequent state writing order comparison is established within the same document object, avoiding cross-interference between circulation records with different document RFID tag identification values. A document storage location binding matrix is ​​constructed using the document RFID tag identifier as the row index and the target storage location as the column index. First, a matrix row set is established based on all document RFID tag identifiers, then a matrix column set is established based on all target storage location locations. Finally, the document storage location binding matrix is ​​generated according to the row-column correspondence. Each row in the matrix corresponds to a document RFID tag identifier, each column corresponds to a target storage location, and each matrix element represents the state write result of the current document RFID tag identifier relative to the current target storage location. This processing method is used to transform the flow relationship between document objects and storage locations into an indexable, updatable, and traceable two-dimensional binding structure, enabling the state changes of the same document at different target storage locations to be recorded in a unified matrix format.The matrix element corresponding to the document retrieval mark is written with the departure status value, the matrix element corresponding to the document return mark is written with the return status value, and the matrix element corresponding to the pending verification transfer mark is written with the verification status value. During writing, if the current mark record is a document retrieval mark, the departure status value is written to the matrix element in the row corresponding to the current document RFID tag value and the column corresponding to the current target storage location. If the current mark record is a document return mark, the return status value is written to the matrix element in the row corresponding to the current document RFID tag value and the column corresponding to the current target storage location. If a document is marked as a pending verification transfer mark, a verification status value is written into the matrix element corresponding to the row of the current document RFID tag identifier value and the column of the current target storage location. When the same matrix element is written multiple times, the document RFID read / write time value and status writing order corresponding to each write are retained within the matrix element. This processing method is adopted so that the off-position status value, return status value, and verification status value can not only reflect the status attribution between the current document RFID tag identifier value and the current target storage location, but also retain the sequential information required for subsequent time series regression discrimination. For each RFID tag identifier value, the corresponding tag records are arranged in ascending order of the RFID read / write time value, and the state writing order of the corresponding matrix elements in the document storage location binding matrix is ​​updated sequentially. Specifically, the RFID read / write time values ​​corresponding to all tag records within the same RFID tag identifier value are extracted first, and then an ordered writing sequence is formed according to the RFID read / write time values ​​in ascending order. Subsequently, the departure state value, return state value, and verification state value are written to the corresponding matrix elements in sequence according to this ordered writing sequence, while recording the corresponding sequence number of each write in the matrix elements. This processing method is adopted to ensure that the state change process in the document storage location binding matrix is ​​consistent with the actual flow timeline, thereby distinguishing between the departure state write that occurs first and the return state write that occurs later, and between the verification state write that occurs first and the confirmation state write that occurs later.When the record corresponding to the same RFID tag value of a document is written with the departure status value written first and then the return status value written, the departure status value and the return status value are extracted to form a flow-back status pair. During extraction, for each RFID tag value of a document, all matrix elements in the document storage location binding matrix with existing status write records are searched one by one. The departure status value with the earlier writing order is located first, and then the return status value corresponding to the same RFID tag value is located later. When both departure and return status values ​​exist, and the writing order of the return status value is later than that of the departure status value, the current departure status value and the current return status value are combined into a flow-back status pair, while retaining the departure status value. The state value corresponds to the target storage location, the return state value corresponds to the target storage location, the departure state value corresponds to the document RFID read / write time value, and the return state value corresponds to the document RFID read / write time value. This processing method is adopted to extract the complete departure-return closure relationship from the continuous state evolution of the same document RFID tag identification value, so that the consistency between the departure state value and the return state value can be used to determine the document's original situ closed-loop result and the document's misplaced return result. When only the departure state value exists and no subsequent return state value is found, the current record does not form a flow return state pair, but is retained for subsequent external flow determination and pending verification continuation determination.

[0039] In this implementation plan, by unifying the document issuance mark, document return mark, and pending verification transfer mark into a document storage location binding matrix composed of the document RFID tag identification value and the target storage location, and by constraining the writing order of each state with the document RFID read / write time value, the changes in the out-of-position state, the return state, and the verification retention state corresponding to the same document RFID tag identification value can be continuously presented within the same structure. This avoids the problems of disconnected states, confused storage location relationships, and interrupted regression chains caused by the scattered storage of transfer records. This improves the completeness, temporal consistency, and location traceability of the transfer regression state for the extraction results, and provides a more stable basis for subsequent identification of document in-situ closed-loop records, identification of document misplaced return records, and identification of document transfer records based on the out-of-position state value and the return state value.

[0040] Specifically, the following steps are taken to generate a document in-situ closed-loop record when the target storage locations corresponding to the out-of-position state and the return-of-position state are consistent, and to generate a document misplacement return record when the target storage locations corresponding to the out-of-position state and the return-of-position state are inconsistent, and to generate a document external circulation record when no circulation return state pair is formed and the out-of-position retention time difference exceeds the external judgment threshold: Based on each circulation return state pair, the target storage locations corresponding to the out-of-position state value and the return-of-position state value are read respectively. When the two target storage locations are consistent, a document in-situ closed-loop record is generated; wherein, for each group of circulation return state pairs, the target storage location corresponding to the out-of-position state value, the target storage location corresponding to the return-of-position state value, and the document shooting location corresponding to the out-of-position state value are extracted. The RFID read / write time value and the return status value correspond to the RFID read / write time value of the document. Then, the target storage location corresponding to the departure status value and the target storage location corresponding to the return status value are compared to see if they are the same target storage location. When they correspond to the same target storage location, the current document RFID tag identification value, the target storage location corresponding to the departure status value, the document RFID read / write time value corresponding to the departure status value, and the document RFID read / write time value corresponding to the return status value are written into the document's original location closed-loop record. This processing method is used to solidify the flow result of the same document RFID tag identification value returning to the original target storage location after departure into a separate closed-loop regression result, so that subsequent flow judgment and control are based on the regression that keeps the position consistent before and after. When the target storage locations of the two are inconsistent, the matrix element containing the corresponding return status value is read. If the matrix element does not contain a verification status value, a document misplacement return record is generated; if the matrix element contains a verification status value, the pending verification transfer mark is maintained. Specifically, the matrix element containing the current return status value is first located in the document storage location binding matrix, and then a verification status value write record is checked within the current matrix element. When the target storage location corresponding to the departure status value is different from the target storage location corresponding to the return status value, and the matrix element containing the current return status value does not retain a verification status value write record, the current document RFID tag identifier value and the departure status value corresponding to the departure status value are... The target storage location, the return status value corresponding to the target storage location, the departure status value corresponding to the document RFID read / write time value, and the return status value corresponding to the document RFID read / write time value are written into the document misplacement return record. When the current return status value has already retained the verification status value written into the matrix element, the current transfer result is not directly confirmed as a document misplacement return record, but the transfer mark to be verified continues to be maintained. This processing method is adopted in order to first identify the misplacement return risk based on the difference in the target storage location, and then determine whether the current misplacement risk has a basis for direct confirmation based on the verification status value, thereby avoiding the premature solidification of records that still have verification retention attributes as document misplacement return records.For out-of-position state values ​​that have not formed a transition / return state pair, the system's unified timestamp after alignment and the document RFID read / write time value corresponding to the out-of-position state value are read, and the time difference between the system's unified timestamp after alignment and the document RFID read / write time value corresponding to the out-of-position state value is calculated. Among these, out-of-position state values ​​that have not formed a transition / return state pair are state records where the out-of-position state value has been written into the document storage location binding matrix but no subsequent return state value has been matched within the current search range. During reading, the document RFID tag identifier value, target storage location, and document RFID read / write time value corresponding to the out-of-position state value are extracted first, and then the system's unified timestamp after alignment is read. The system unified timestamp; the currently aligned system unified timestamp is taken from the unified time reference after the multi-source timestamps are aligned by the network time protocol synchronization algorithm, which is used to ensure that the currently aligned system unified timestamp and the document RFID read / write time value corresponding to the off-position state value are in the same time reference frame. Then, the time difference is obtained by subtracting the document RFID read / write time value corresponding to the off-position state value from the currently aligned system unified timestamp. This processing method is adopted in order to continue to expand the off-position state along the unified time axis and determine whether the unclosed loop state corresponding to the current document RFID tag identification value is a short-term waiting-to-return state or a continuous off-position state. When the time difference exceeds the threshold for determining whether a document is in circulation, an external circulation record is generated. The external circulation threshold is preset based on the maximum allowed circulation time of documents in the control area. When the time difference exceeds the external circulation threshold, it indicates that the current document RFID tag value has not formed a valid closed loop result within the specified time range since the departure status value was written. At this time, the current document RFID tag value, the target storage location corresponding to the departure status value, the document RFID read / write time value corresponding to the departure status value, the currently aligned system unified timestamp, and the time difference are written into the document external circulation record. This processing method is used to separate the departure status that has been continuously unclosed and has exceeded the allowed time range from the general unclosed loop record, so that the subsequent external warning judgment is based on a clear timeout basis. When the time difference is not greater than the external judgment threshold, or when the matrix element corresponding to the out-of-position state value has been written into the verification state value, the pending verification transfer mark is maintained. Specifically, a time difference not greater than the external judgment threshold indicates that the out-of-position state corresponding to the current document RFID tag value is still within the allowed waiting time range for return, and does not yet meet the conditions for direct transfer into the document's external transfer record. The matrix element corresponding to the out-of-position state value has been written into the verification state value, indicating that the transfer chain of the current out-of-position state still retains information awaiting further verification, and the current record must continue to maintain the pending verification transfer mark. This processing method is adopted to avoid prematurely identifying unclosed records still within a reasonable waiting period and unclosed records still with verification retention attributes as document external transfer records, thereby maintaining clear judgment boundaries between document in-situ closed-loop records, document misplacement return records, document external transfer records, and pending verification transfer marks.

[0041] In this implementation plan, by constructing a closed-loop regression judgment relationship around the off-position status value, return status value, target storage location, document RFID reading and writing time value, and current time, the circulation results corresponding to the same document RFID tag identification value can be clearly distinguished in three aspects: return consistency, return offset, and continuous non-return. This avoids non-closed-loop records and abnormal return records from being mixed into the same judgment result, and prevents short-term non-return status from being prematurely identified as document circulation status. This improves the boundary clarity between document in-situ closed-loop records, document misplacement return records, and document circulation records, so that subsequent circulation control status judgment is based on more stable and traceable regression results.

[0042] Specifically, the steps for reading the original closed-loop record of the document, the record of the document being returned in a different location, the record of the document being circulated outside the office, and the circulation mark to be verified, constructing the document circulation judgment and control feature sequence and inputting it into the dual-branch gated loop unit judgment and control model, and outputting the circulation control status are as follows: Read the original closed-loop record of the document, the record of the document being returned in a different location, the record of the document being circulated outside the office, and the circulation mark to be verified, classify them according to the document RFID tag identification value, and arrange them in ascending order according to the document RFID read / write time value; among them, first uniformly read all the original closed-loop record of the document, the record of the document being returned in a different location, and the record of the document being circulated outside the office. Records and pending verification of transfer markers are then grouped using the RFID tag identifier as the classification key, ensuring that all transfer records corresponding to the same RFID tag identifier enter the same judgment sequence. Subsequently, within each set of records corresponding to an RFID tag identifier, records are sorted in ascending order based on the RFID read / write time, forming a sequence of transfer records unfolding in chronological order. This processing method is used to project multiple transfer results corresponding to the same RFID tag identifier onto the same timeline, allowing subsequent state evolution relationship extraction to be based on continuous records. The original closed-loop record of the document, the record of the document being returned in a different location, the record of the document being circulated outside the document, and the record of the document being checked for circulation are encoded into different circulation status values. The corresponding target storage location, storage location fit evaluation value, continuous placement evaluation value, and time interval between two adjacent records are extracted to construct a document circulation judgment and control feature sequence. Specifically, the original closed-loop record of the document is encoded as the first circulation status value, the record of the document being returned in a different location is encoded as the second circulation status value, the record of the document being circulated outside the document is encoded as the third circulation status value, and the record of the document being checked for circulation is encoded as the fourth circulation status value. After encoding, each record forms a circulation status value uniquely corresponding to the current record type. Then, the target storage location, storage location fit evaluation value, and continuous placement evaluation value corresponding to the current record are extracted one by one. The target storage location is taken from the storage location identifier result corresponding to the current record, and the storage location fit evaluation value is taken from the current record's record location. The corresponding storage location fitting evaluation result is retained in the document storage location mapping record. The continuous evaluation value of the current record is taken from the corresponding continuous evaluation result of the current record in the document storage location mapping record. Then, the document RFID read / write time value of the previous record is subtracted from the document RFID read / write time value of the current record to obtain the time interval value corresponding to the current record. For the first record after sorting, the time interval value is set to the preset initial interval value. Subsequently, the circulation status value, target storage location, storage location fitting evaluation value, continuous evaluation value of the current record, and time interval value are written in chronological order to construct the document circulation judgment and control feature sequence. This processing method is adopted to encapsulate the status change information, location attribution information, mapping credibility information, and time evolution information into the same sequence structure, so that subsequent judgment and control no longer rely on a single record, but are based on a comprehensive identification based on the continuous circulation process.Offline training was performed on the dual-branch gated loop unit judgment and control model based on the training sample set, and the trained dual-branch gated loop unit judgment and control model was called based on the document circulation judgment and control feature sequence. The training sample set consisted of historical document in-situ closed-loop records, historical document misplacement return records, historical document circulation records, and historical circulation markers awaiting verification. For each document's RFID tag identifier value, the corresponding historical records were arranged in ascending order of the document's RFID read / write time value. Circulation status values, target storage location, storage location fit evaluation value, continuous placement evaluation value, and time interval values ​​were extracted to form the historical document circulation judgment and control feature sequence. Among these, samples from historical document in-situ closed-loop records that were subsequently verified to be without abnormalities were marked as normal release status, and samples from historical document misplacement return records that were manually confirmed as misplaced were marked as misplaced. In the locked state, samples of historical document circulation records that have not been returned for confirmation within the timeout period are marked as being in an "out-of-bounds warning" state. Samples of historical circulation records that require manual verification and confirmation and need manual intervention are marked as being in a "manual verification" state. During training, the cross-entropy loss function between the real label of circulation control status and the probability of the model output status is used as the optimization objective to iteratively update the parameters of the dual-branch gated loop unit judgment and control model until the loss function converges. After training, the trained dual-branch gated loop unit judgment and control model is deployed into the document circulation judgment and control process to perform online inference on the real-time generated document circulation judgment and control feature sequences. This approach is adopted to ensure that the parameter learning of the dual-branch gated loop unit judgment and control model is based on historical real circulation results, so that the model output results have a basis for supervised correction. The trained dual-branch gated loop unit judgment and control model includes a first branch, a second branch, a gated fusion layer, and a state discrimination layer. First, the document circulation judgment and control feature sequence is split into a time-series state input sequence and a storage location binding input sequence according to feature type. The time-series state input sequence consists of circulation state values ​​and time interval values ​​in chronological order, while the storage location binding input sequence consists of the target storage location, storage location fitting evaluation value, and continuous entry evaluation value in chronological order. The first branch sequentially reads the circulation state values ​​and time interval values ​​to extract the circulation state time-series features; the second branch sequentially reads the target storage location and storage location fitting evaluation value. Combined evaluation values ​​and continuous evaluation values ​​for position placement are used to extract reservoir binding features. The first branch uses a gating update method to retain state change information that contributes to subsequent judgments and controls, suppressing early state disturbances that contribute little to subsequent judgments and controls, and outputting the time-series features of the transitional state. The second branch uses a gating update method to retain reservoir attribution change information, mapping stability change information, and position placement convergence change information that contribute to subsequent judgments and controls, and outputs the reservoir binding features. This processing method is used to extract the evolution law of the transitional state and the evolution law of the reservoir mapping separately, avoiding mutual masking of different attribute features in the same recursive process.The time-series features of document circulation status and storage location binding features are input into the gating fusion layer to generate a document circulation judgment and control vector, and output the circulation control status, which includes normal release status, misalignment locking status, out-of-area warning status, and manual review status. Specifically, the gating fusion layer performs joint weighted updates on the time-series features of circulation status and storage location binding features to generate a document circulation judgment and control vector that simultaneously represents the historical circulation stability of the current document RFID tag value, the degree of storage location offset, the continuous trend of being out of the area, and the degree of aggregation to be verified. This document circulation judgment and control vector is then input into the status discrimination layer, which outputs the corresponding normal release status, misalignment locking status, out-of-area warning status, and manual review status. The system considers four states of the nuclear status. When the highest probability among the four states is not less than the status confirmation threshold, the state corresponding to the highest probability is determined as the flow control state. When the highest probability among the four states is less than the status confirmation threshold, the flow control state corresponding to the current document RFID tag value is determined as the manual review state. The status confirmation threshold is a pre-set minimum confirmation confidence threshold, used to prevent low-confidence output results from directly entering the confirmation state. This processing method is used to compress continuous flow records corresponding to the same document RFID tag value into a single judgment result, so that the flow control state output is based on the joint judgment of multiple records. When the maximum state probability corresponding to the document circulation control vector falls into the normal release state and is not less than the state confirmation threshold, the normal release state is output; when the maximum state probability corresponding to the document circulation control vector falls into the misaligned lock state and is not less than the state confirmation threshold, the misaligned lock state is output; when the maximum state probability corresponding to the document circulation control vector falls into the out-of-area warning state and is not less than the state confirmation threshold, the out-of-area warning state is output; when the maximum state probability corresponding to the document circulation control vector falls into the manual review state, or the maximum state probability is less than the state confirmation threshold, the manual review state is output. This processing method is adopted to ensure that the circulation control state is constrained by the model output probability and the state confirmation threshold, avoiding low-confidence samples from being incorrectly fixed as the normal release state, misaligned lock state, or out-of-area warning state.

[0043] In this implementation plan, by unifying the original closed-loop record of the document, the record of the document misplacement and return, the record of the document in external circulation, and the circulation mark to be verified into a document circulation judgment and control feature sequence based on the document RFID tag identification value, and by incorporating the circulation status value, target storage location, storage location fit evaluation value, continuous placement evaluation value, and time interval value into the same judgment and control process, the historical circulation trajectory, storage location binding change trajectory, and abnormal accumulation trend corresponding to the same document RFID tag identification value can continuously converge on the same judgment basis. This avoids fluctuations in circulation control status caused by local judgments triggered by a single record, and avoids the separation of location mapping anomalies, continuous non-return phenomena, and pending verification cluster phenomena. This improves the distinguishability and stability between normal release status, misplacement lock status, external warning status, and manual review status, making the final output result more consistent with the actual circulation of documents.

[0044] Specifically, the steps for executing the corresponding control instructions are as follows: When the circulation control status is in normal release status, maintain normal circulation permissions; first, read the circulation control status corresponding to the current document RFID tag identifier value, then verify whether the latest document in-situ closed-loop record corresponding to the current document RFID tag identifier value has formed a closed result, and verify whether the current target storage location is consistent with the target storage location corresponding to the most recent return status value. When the verification results meet the normal release conditions, maintain the normal circulation permissions corresponding to the current document RFID tag identifier value, do not write the restricted status, do not trigger locking actions, do not trigger warning actions, and do not trigger manual interception actions; This processing method is adopted to ensure that documents that have completed normal closed-loop circulation continue to maintain the circulation capability of being available, returnable, and re-verified, and to avoid mistakenly transferring documents that have been confirmed to be without abnormalities into the restricted status. When the circulation control status is in the misalignment lock status, a storage location lock instruction is generated and the corresponding storage location lock is executed. This involves first reading the document misalignment return record corresponding to the current document RFID tag identifier value, then extracting the target storage location corresponding to the return status value from the document misalignment return record, and determining this return status value as the current locked object location. Subsequently, a storage location lock instruction is generated, containing the document RFID tag identifier value, the locked object location, the lock trigger time, and the misalignment lock status code. When executing the corresponding storage location lock, the storage location lock instruction is written to the execution control queue corresponding to the current locked object location, causing the current locked object location to enter a restricted state that prohibits reopening, re-entry, and re-retrieval, and simultaneously writing the lock processing record corresponding to the current document RFID tag identifier value. This processing method is used to immediately block subsequent circulation actions at the incorrect target storage location after detecting that a document has been placed in the wrong target storage location, preventing the misalignment return status from spreading to a state of multiple documents being mixed together. When the document circulation control status is in the "outside warning" state, an "outside retention warning" instruction is generated, and the corresponding warning output is executed. This involves first reading the document circulation record corresponding to the current document RFID tag value, then extracting the target storage location corresponding to the departure status value, the document RFID read / write time value corresponding to the departure status value, the current time, and the time difference from the document circulation record. This information is written into the "outside retention warning" instruction, which also includes the warning trigger time, warning level identifier, and warning status code. When executing the corresponding warning output, retention prompt information corresponding to the current document RFID tag value is generated according to the warning level identifier in the "outside retention warning" instruction. This retention prompt information is then written into the warning output record, enabling the management side to perform tracking and handling based on the continuous non-return status corresponding to the current document RFID tag value. This processing method is used to independently output continuous non-return statuses exceeding the "outside" judgment threshold from general non-closed-loop statuses, allowing the risk of document retention to be identified in a timely manner.When the circulation control status is in manual review mode, a manual review instruction is generated, and subsequent circulation confirmation is frozen. This involves first reading the circulation marker to be verified corresponding to the current document RFID tag value, then extracting the document RFID read / write time value, target storage location, storage location fit evaluation value, and continuous placement evaluation value corresponding to the circulation marker to be verified. This information is written into the manual review instruction, which also includes the review trigger time, review reason identifier, and review status code. When freezing subsequent circulation confirmation, the subsequent document retrieval marker generation action, document return marker generation action, and circulation return status extraction action corresponding to the current document RFID tag value are suspended until the review result corresponding to the manual review instruction is written. This processing method is used to prevent the circulation record corresponding to the current document RFID tag value from continuing to enter the confirmation process under conditions of insufficient judgment basis, unstable mapping results, or an unclosed directional chain, thereby avoiding the repeated amplification of uncertain records into deterministic circulation results.

[0045] In this implementation plan, by establishing a one-to-one correspondence between the circulation control status and the corresponding handling action, the normal closed-loop result, misplaced return result, external retention result, and pending review result corresponding to the current document RFID tag identification value can enter differentiated control paths respectively. This avoids the problems of circulation authority mismatch, abnormal status continuation, and continued confirmation of undetermined records caused by using the same handling method for records of different risk levels. This improves the responsiveness between the target storage location control result, the document external circulation risk handling result, and the pending verification circulation tag interception result, so that the entire document circulation process forms a closed-loop control relationship from status determination to execution.

[0046] like Figure 2As shown, the second aspect of the present invention provides a control system for the circulation of identification documents in a control area based on multimodal identity verification, comprising: a data acquisition and processing module, a verification storage location mapping module, a closed-loop regression discrimination module, and a judgment and control execution output module, wherein: the data acquisition and processing module is used to acquire identification document circulation data in real time, perform preprocessing on the identification document circulation data, and output the preprocessed identification document circulation data; the verification storage location mapping module is used to generate verification document circulation segments based on the preprocessed identification document circulation data, and filter candidate storage locations from the segments; select the target storage location based on the product of the storage location fitting evaluation value and the continuous entry evaluation value corresponding to the candidate storage location; and generate a document retrieval mark, a document return mark, and a circulation mark to be verified based on the circulation event time sequence discrimination result corresponding to the target storage location; the closed-loop regression discrimination module generates a document retrieval mark, a document return mark, and a circulation mark to be verified based on the closed-loop regression discrimination result; the closed-loop regression discrimination module generates a verification storage location mapping module, a closed-loop regression discrimination module, and a judgment and control execution output module; the closed-loop regression discrimination module generates a verification storage location mapping module, a closed-loop regression discrimination module, and a closed-loop regression discrimination ... The loop regression discrimination module is used to read the document issuance mark, document return mark, and pending verification circulation mark, construct the document storage location binding matrix, and extract circulation regression state pairs. When the target storage location corresponding to the departure state and the return state is consistent, a document in-situ closed-loop record is generated. When the target storage location corresponding to the departure state and the return state is inconsistent, a document misplacement return record is generated. When no circulation regression state pair is formed and the departure delay time difference exceeds the external judgment threshold, a document external circulation record is generated. The judgment and control execution output module is used to read the document in-situ closed-loop record, document misplacement return record, document external circulation record, and pending verification circulation mark, construct the document circulation judgment and control feature sequence and input it into the dual-branch gating loop unit judgment and control model, output the circulation control status, and execute the corresponding control instructions.

[0047] In this implementation plan, by incorporating document circulation data processing, target storage location determination, circulation regression relationship identification, and circulation control status output into the same technical chain, the identity verification results, storage location mapping results, closed-loop regression results, and anomaly control results corresponding to the document RFID tag identification value can be continuously connected in the pre- and post-processing stages. This avoids the separation between document issuance markers, document return markers, pending verification circulation markers, and document in-situ closed-loop records, document misplacement return records, and document external circulation records. It also prevents the circulation control status from being formed independently of the target storage location change process, thereby improving the overall traceability, judgment continuity, and handling closure of the document circulation process in the control area, and enabling the true circulation status of documents to be stably identified and responded to in a timely manner.

[0048] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0049] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for controlling the flow of identification documents in a control area based on multimodal identity verification, characterized in that, Includes the following steps: S1 collects document circulation data in real time, performs preprocessing on the document circulation data, and outputs the preprocessed document circulation data; The specific steps for real-time collection of document circulation data and preprocessing of the document circulation data are as follows: The system collects real-time data on document circulation in the control area. This data includes verification time, face image acquisition time, fingerprint image acquisition time, password input completion time, document RFID tag identification value, document RFID read / write time, document RFID signal strength value, horizontal position value corresponding to the storage location identifier, vertical position value corresponding to the storage location identifier, depth position value corresponding to the storage location identifier, photoelectric trigger time at the storage location entrance, photoelectric trigger time inside the storage location, storage location read / write antenna signal strength value, document leaving the storage location, document entering the storage location, lateral displacement value of the robotic arm, vertical displacement value of the robotic arm, depth displacement value of the robotic arm, and cabinet door opening time. For the collected document circulation data, the following steps are performed: Network Time Protocol (NTP) time synchronization algorithm is used to perform multi-source timestamp alignment; event sequence verification algorithm is used to correct the chronological order of circulation events; box plot outlier detection algorithm is used to identify and remove outlier data; Hanpus filtering algorithm is used to suppress discrete mutation document circulation records; and min-max normalization algorithm is used to unify numerical scales, outputting the preprocessed document circulation data. S2, generate a document circulation segment for verification based on the preprocessed document circulation data, and select candidate storage locations from the segment; select the target storage location based on the product of the storage location fit evaluation value and the continuous entry evaluation value corresponding to the candidate storage location; generate a document retrieval mark, a document return mark, and a circulation mark to be verified based on the circulation event timing judgment result corresponding to the target storage location; The specific steps for selecting the target storage location based on the product of the storage location fit evaluation value and the continuous evaluation value of the candidate storage location are as follows: For each candidate storage location, the signal strength value of the document's radio frequency signal is added to the signal strength value of the storage location's read / write antenna, then one is added and the natural logarithm is taken to obtain the signal correspondence quantity. The differences between the lateral displacement value of the robotic arm and the lateral position value corresponding to the storage location identifier, the differences between the longitudinal displacement value of the robotic arm and the longitudinal position value corresponding to the storage location identifier, and the differences between the depth displacement value of the robotic arm and the depth position value corresponding to the storage location identifier are squared, summed, and the square root is taken, then one is added to obtain the spatial deviation quantity. The signal correspondence quantity is divided by the spatial deviation quantity to obtain the storage location fit evaluation value of the document's radio frequency tag identifier value relative to the current candidate storage location. The absolute values ​​of the differences between the photoelectric trigger time value at the entrance of the storage space and the photoelectric trigger time value inside the storage space, the absolute values ​​of the differences between the photoelectric trigger time value inside the storage space and the time value of the document entering the storage space, and the absolute values ​​of the differences between the time value of the document entering the storage space and the time value of the document's radio frequency reading and writing are added one by one, and the natural logarithm is taken. Then, the natural logarithm is added one by one and the reciprocal is taken to obtain the continuous evaluation value of the document's radio frequency tag identification value relative to the current candidate storage space. The storage location fitting evaluation value is multiplied by the continuous evaluation value of the entry location, and the candidate storage location with the largest product is selected as the target storage location corresponding to the current document RFID tag identification value, and a document storage location mapping record is generated; when there are multiple candidate storage locations with the same product of storage location fitting evaluation value and continuous evaluation value of entry location, or when the candidate storage location is empty, it is determined that the target storage location cannot be uniquely determined. S3: Read the document issuance mark, document return mark, and pending verification transfer mark, construct the document storage location binding matrix, and extract the transfer return state pair. When the target storage location corresponding to the departure state and the return state is consistent, generate the document original closed loop record. When the target storage location corresponding to the departure state and the return state is inconsistent, generate the document misplacement return record. When no transfer return state pair is formed and the departure delay time difference exceeds the external judgment threshold, generate the document external transfer record. S4 reads the original closed-loop record of the certificate, the record of the certificate being returned in a different location, the record of the certificate being transferred outside the office, and the transfer mark to be checked. It constructs the certificate transfer judgment and control feature sequence and inputs it into the judgment and control model of the dual-branch gated loop unit. It outputs the transfer control status and executes the corresponding control instructions.

2. The control area document circulation management method based on multimodal identity verification according to claim 1, characterized in that: The specific steps for generating verification document flow segments based on preprocessed document flow data and selecting candidate storage locations from these segments are as follows: Read the preprocessed control area document flow data, and for each document RFID read / write time value, extract the verification events corresponding to the face image acquisition time value, fingerprint image acquisition time value, and password input end time value corresponding to the most recent verification time value before the document RFID read / write time value, as well as the corresponding flow events, and generate a verified document flow segment; Based on the flow segments of each verification document, for each document's RFID tag identifier value, first read the lateral displacement value, longitudinal displacement value, and depth displacement value of the robotic arm corresponding to the current document's RFID tag identifier value, then read the lateral position value, longitudinal position value, and depth position value corresponding to all storage location identifiers; calculate the absolute value of the difference between the lateral displacement value of the robotic arm and the lateral position value corresponding to each storage location identifier, the absolute value of the difference between the longitudinal displacement value of the robotic arm and the longitudinal position value corresponding to each storage location identifier, and the absolute value of the difference between the depth displacement value of the robotic arm and the depth position value corresponding to each storage location identifier; retain storage locations where the lateral position difference, longitudinal position difference, and depth position difference are all less than the corresponding position tolerance threshold to form a preliminary set of storage locations; then read the storage location read / write antenna signal strength value corresponding to each storage location in the preliminary set of storage locations, and retain the storage locations where the storage location read / write antenna signal strength value is greater than the antenna response threshold as candidate storage locations.

3. The control area document circulation management method based on multimodal identity verification according to claim 2, characterized in that: The specific steps for generating a document retrieval marker, a document return marker, and a pending verification marker based on the timing judgment result of the circulation event corresponding to the target storage location are as follows: Based on the time sequence of document departure and entry times in the document storage location mapping records, the circulation direction corresponding to the current document RFID tag value is determined. When the cabinet door opening time occurs before the document departure time, and the document departure time occurs before the document entry time, a document retrieval mark is generated. When the photoelectric trigger time at the storage location entrance, the photoelectric trigger time inside the storage location, and the document entry time occur sequentially, a document return mark is generated. When a complete circulation direction is not formed within the same document circulation segment, or the target storage location cannot be uniquely determined, a circulation mark to be verified is generated.

4. The control area document circulation management method based on multimodal identity verification according to claim 3, characterized in that: The specific steps for reading the document issuance mark, document return mark, and pending verification transfer mark, constructing the document storage location binding matrix, and extracting the transfer return status pairs are as follows: The document retrieval marker, document return marker, and pending verification transfer marker are classified according to the document RFID tag identification value, and the target storage location and document RFID read / write time value corresponding to each marker are extracted. A document storage location binding matrix is ​​constructed with the document RFID tag identification value as the row index and the target storage location as the column index. Among them, the matrix element corresponding to the document retrieval marker is written with the departure status value, the matrix element corresponding to the document return marker is written with the return status value, and the matrix element corresponding to the pending verification transfer marker is written with the verification status value. For each RFID tag identifier value, the corresponding tag records are arranged in ascending order according to the RFID read / write time value of the document, and the state writing order of the corresponding matrix elements in the document storage location binding matrix is ​​updated in sequence. When the record corresponding to the same RFID tag identifier value is written with the departure state value first and then the return state value, the departure state value and the return state value are extracted to form a flow return state pair.

5. The control area document circulation management method based on multimodal identity verification according to claim 4, characterized in that: The specific steps for generating a closed-loop record of the document in its original location when the target storage locations corresponding to the out-of-position state and the return state are consistent, generating a record of the document's misplaced return when the target storage locations corresponding to the out-of-position state and the return state are inconsistent, and generating a record of the document's external circulation when no circulation and return state pair has been formed and the time difference of out-of-position retention exceeds the external determination threshold are as follows: Based on each pair of circulation and return states, the target storage locations corresponding to the out-of-position state value and the return state value are read respectively. When the two target storage locations are consistent, a closed-loop record of the document in its original position is generated. When the two target storage locations are inconsistent, the matrix element containing the corresponding return state value is read. When the matrix element has not been written with the verification state value, a record of document misplacement is generated. When the matrix element has been written with the verification state value, the circulation mark to be verified is maintained. For out-of-position status values ​​that have not formed a circulation return status pair, read the currently aligned system unified timestamp and the document RFID read / write time value corresponding to the out-of-position status value, and calculate the time difference between the currently aligned system unified timestamp and the document RFID read / write time value corresponding to the out-of-position status value; when the time difference is greater than the external judgment threshold, generate a document external circulation record; when the time difference is not greater than the external judgment threshold, or the matrix element where the corresponding out-of-position status value is located has been written into the verification status value, maintain the circulation mark to be verified.

6. The control area document circulation management method based on multimodal identity verification according to claim 5, characterized in that: The specific steps for reading the original closed-loop record of the document, the record of the document misplacement and return, the record of the document's external circulation, and the circulation mark to be verified, constructing the document circulation judgment and control feature sequence, inputting it into the dual-branch gated loop unit judgment and control model, and outputting the circulation control status are as follows: The system reads the original closed-loop record of the document, the record of the document misplaced return, the record of the document circulating outside, and the circulation mark to be verified. It classifies the documents according to the RFID tag identification value and arranges them in ascending order according to the RFID read / write time value. The original closed-loop record of the document, the record of the document misplaced return, the record of the document circulating outside, and the circulation mark to be verified are encoded into different circulation status values. The corresponding target storage location, storage location fit evaluation value, entry continuity evaluation value, and time interval value between two adjacent records are extracted to construct the document circulation judgment and control feature sequence. The document circulation judgment and control feature sequence is input into the pre-trained dual-branch gated loop unit judgment and control model. The first branch reads the circulation status value and time interval value to extract the circulation status temporal features; the second branch reads the target storage location, storage location fitting evaluation value and entry continuous evaluation value to extract the storage location binding features. Input the time-series features of the circulation status and the storage location binding features into the gating fusion layer to generate the document circulation judgment and control vector, and output the circulation control status, which includes normal release status, misalignment locking status, out-of-area warning status, and manual review status.

7. The control area document circulation management method based on multimodal identity verification according to claim 6, characterized in that: The specific steps for executing the corresponding control instructions are as follows: When the circulation control status is normal release status, normal circulation permissions are maintained; When the circulation control status is misaligned and locked, a storage location locking command is generated and the corresponding storage location is locked. When the circulation control status is in the "outside warning" status, generate a warning instruction for the document being held outside and execute the corresponding warning output; When the workflow control status is in the manual review status, a manual review instruction is generated, and subsequent workflow confirmation is frozen.

8. A control area document circulation management system based on multimodal identity verification, employing the control area document circulation management method based on multimodal identity verification as described in any one of claims 1-7, characterized in that, include: The module comprises a data acquisition and processing module, a storage location verification and mapping module, a closed-loop regression and discrimination module, and a judgment and control execution output module, among which: The data acquisition and processing module is used to acquire document circulation data in real time, perform preprocessing on the document circulation data, and output the preprocessed document circulation data. The verification storage location mapping module is used to generate verification document circulation segments based on preprocessed document circulation data, and to filter candidate storage locations from the segments; to select the target storage location based on the product of the storage location fit evaluation value and the continuous entry evaluation value corresponding to the candidate storage location; and to generate a document retrieval mark, a document return mark, and a circulation mark to be verified based on the circulation event timing judgment result corresponding to the target storage location. The closed-loop regression discrimination module is used to read the document issuance mark, document return mark, and pending verification circulation mark, construct the document storage location binding matrix, and extract circulation regression state pairs. When the target storage location corresponding to the departure state and the return state is consistent, a document in-situ closed-loop record is generated. When the target storage location corresponding to the departure state and the return state is inconsistent, a document misplacement return record is generated. When no circulation regression state pair is formed and the departure delay time difference exceeds the external judgment threshold, a document external circulation record is generated. The judgment and control execution output module is used to read the original closed-loop record of the certificate, the record of the certificate misplacement and return, the record of the certificate in external circulation and the circulation mark to be checked, construct the judgment and control feature sequence of the certificate circulation and input it into the judgment and control model of the dual-branch gated loop unit, output the circulation control status and execute the corresponding control instructions.

Images