Ultrahigh-frequency radio frequency tableware batch identification and transaction control system based on artificial intelligence

Abstract

The invention discloses an ultrahigh-frequency radio frequency tableware batch identification and transaction control system based on artificial intelligence, and the system comprises a tableware radio frequency tag module which is used for embedding an ultrahigh-frequency RFID tag in a tableware body; the radio frequency reading and signal acquisition module is used for outputting structured data records; the artificial intelligence identification and error correction module is used for generating a credible tableware identifier set and a state feature set; the array antenna control module is used for dynamically adjusting a beam direction angle and a transmitting power parameter value of the array antenna; the edge computing transaction processing module is used for inputting the trusted tableware identifier set into an edge computing transaction processing flow and executing repeated charging risk verification; the payment control module is used for returning a payment result; and the background management and tracking module is used for generating operation statistics and abnormal record entries. The system has high-precision tableware identification, low-delay settlement and abnormity prevention and control capabilities, and is suitable for automatic identification and intelligent settlement management of batch tableware in a catering scene.

Description

Technical Field

[0001] This invention relates to the fields of Internet of Things and radio frequency identification (RFID) technology, and in particular to an artificial intelligence-based UHF radio frequency tableware batch identification and transaction control system. Background Technology

[0002] In the catering industry, UHF (Radio Frequency Identification) technology is widely used for tableware identification and circulation management due to its advantages such as long reading distance, strong batch identification capability, and resistance to environmental interference. Existing catering settlement systems typically embed UHF RFID tags within the tableware itself, using fixed or handheld readers to read tag information in batches, enabling tableware entry and exit from storage, circulation tracking, and transaction settlement. However, traditional systems generally suffer from the following problems:

[0003] In environments with high-density tableware stacking, proximity to metal, and high humidity, missed readings, misreads, and signal strength fluctuations are prone to occur, leading to incomplete or erroneous identification data and affecting the accuracy of subsequent transaction settlements. Most systems use fixed beam direction and power parameters, unable to dynamically adjust based on real-time reading performance, resulting in reading blind spots, beam jitter, or excessive transmission power. While existing anti-collision algorithms can reduce collisions to some extent, they are insufficient in handling temporal overlap, duplicate recordings, and signal anomalies in multi-tag, high-frequency response scenarios, leading to instability in the construction of trusted tag sets. When tableware frequently enters and exits the reading area, the lack of effective session management and duplicate billing verification can easily result in multiple billings for the same tag within a short period, increasing operational risks. Existing array antenna control mainly relies on manual settings or simple feedback adjustments, failing to incorporate advanced algorithms such as dual-critic delayed updates for strategy optimization, making it difficult for the system to maintain optimal reading performance continuously in variable environments.

[0004] In summary, existing technologies are insufficient in terms of accuracy, real-time performance, and transaction control security for batch identification of tableware. There is an urgent need for a comprehensive solution that combines artificial intelligence-based error correction, improved TD3 algorithm adaptive beam control, and transaction security verification mechanisms to improve the stability and security of the system. Summary of the Invention

[0005] One objective of this invention is to propose an artificial intelligence-based method for batch identification and transaction control of UHF RFID tableware. This invention integrates UHF RFID tag identification, multi-tag anti-collision and data error correction, an improved TD3 algorithm-driven array antenna adaptive beam control mechanism, and a transaction security verification process. It describes in detail the whole process implementation method in batch identification of tableware, dynamic optimization of beam and power, and transaction settlement, and has the advantages of high identification accuracy, strong reading stability, and good transaction security.

[0006] The artificial intelligence-based batch identification and transaction control system for ultra-high frequency radio frequency tableware according to an embodiment of the present invention includes:

[0007] A tableware radio frequency tag module is used to embed ultra-high frequency RFID tags into the tableware body;

[0008] The radio frequency reading and signal acquisition module is used to output structured data records when tableware enters the reading area;

[0009] The artificial intelligence recognition and error correction module is used to preprocess the collected structured data records to generate a set of trustworthy tableware identifiers and a set of state features;

[0010] The array antenna control module is used to dynamically adjust the array antenna beam direction angle and transmit power parameter values ​​based on the dual-Critic delay update adaptive beam control mechanism.

[0011] The edge computing transaction processing module is used to input the set of trusted tableware identifiers into the edge computing transaction processing flow and perform duplicate billing risk verification.

[0012] The payment control module is used to trigger payment methods based on billing information to complete transaction settlement and return payment results;

[0013] The backend management and tracking module is used to generate operational statistics and anomaly record entries.

[0014] Optionally, modules can be integrated using the following methods:

[0015] S1. Embed an ultra-high frequency RFID tag in the body of the tableware and write and bind it, and establish a mapping relationship between the EPC code and the TID number;

[0016] S2. When the tableware enters the reading area, the radio frequency reading and signal acquisition process acquires the EPC code, TID number and original tag reading feature information in batches.

[0017] S3. Aggregate structured data records and filter trusted tableware identifiers to generate a set of trusted tableware identifiers and their change ratios. Combine the distribution of received signal strength indication values ​​and the reread request queue to construct a set of state features and form a complete output data packet.

[0018] S4. The adaptive beam control mechanism for array antenna based on dual-Critic delay update adopts the improved TD3 algorithm, uses offline behavior and online strategy to realize adaptive control of array antenna beam direction and transmit power, and updates the Actor network cyclically through constraint post-action, reward signal and experience playback mechanism.

[0019] S5. Input the set of trusted tableware identifiers into the edge computing transaction processing flow and perform duplicate billing risk verification.

[0020] S6. Complete the payment processing of the statement information and trusted tableware identifier set, and generate operational statistics and abnormal record entries in the back-end management and tracking module.

[0021] Optionally, step S1 includes the following specific steps:

[0022] S11. Select UHF RFID tags based on the material type and geometric dimensions of the tableware body, and determine the tag embedding position;

[0023] S12. Process the installation slot at the tag embedding position, place the UHF RFID tag in it and fix it with epoxy potting material, and perform a sealing and washing cycle test after curing.

[0024] S13. Define the field structure and length configuration for the unique identifier code of tableware;

[0025] S14. At the coding station, use a calibrated RFID reader to write the unique identification code of the tableware into the EPC storage area of ​​the tag, read back to check the consistency of the written content, read the TID number of the tag and form the correspondence between the EPC code and the TID number.

[0026] S15. Set an access password for the tag, enable write protection for the EPC storage area, write production date, capacity and specification parameter information to the user area as needed, and maintain the read-only attribute of the TID area.

[0027] S16. Perform a tag performance sampling inspection on each piece of tableware that has completed the writing of the unique identifier, including sampling inspection of the stability of the reading distance and the received signal strength indication value, record the reading distance range, signal strength fluctuation range, antenna channel information and test time, and conduct supplementary sampling inspections for metal proximity and humid and hot conditions.

[0028] S17. Record the tableware type, capacity, batch, EPC code, TID number, packaging process batch number and test results to the ledger database of the background management and tracking module, complete the filing of tableware and label binding relationship and generate a searchable index.

[0029] Optionally, the generation of the searchable index refers to using the unique identification information of tableware as the primary key (including EPC code and TID number) in the ledger database of the background management and tracking module, and storing fields such as tableware type, capacity, batch, packaging process batch number and test results in association according to a preset data table structure; while writing data, the system automatically builds a B+ tree index structure for the EPC code, TID number and batch field, and builds a hash index for the tableware type and capacity field to support efficient multi-condition combination retrieval and accurate matching, thereby ensuring that the corresponding tableware and its label information can be quickly located in the subsequent transaction, traceability and statistical process.

[0030] Optionally, the radio frequency (RF) reading and signal acquisition process is as follows: When the tableware with the unique identifier written and bound enters the reading area of ​​the RF reading and signal acquisition module, the RF reading and signal acquisition module transmits an ultra-high frequency (UHF) RF signal with a working frequency band of 860MHz to 960MHz through a multi-channel array antenna to excite the UHF RFID tag embedded in the tableware body, causing the tag's EPC storage area to return the unique identifier code, TID number, and user area extended information; the system simultaneously collects the tag reading feature information of each tag and the current array antenna beam direction and transmission power parameters, and completes the acquisition of batch tag data by combining a multi-antenna polling strategy and an anti-collision reading mechanism, and organizes the acquired EPC code, TID number, and tag reading feature information into a structured data record. The tag reading feature information includes signal strength indication value, timestamp, and antenna channel information;

[0031] Optionally, step S3 includes the following specific steps:

[0032] S31. Receive and buffer the structured data records output by the RF reading and signal acquisition module, aggregate the preprocessed structured data records according to EPC encoding, and generate tag reading feature information for each EPC encoding in each antenna channel.

[0033] S32. Define the tag records in the aggregation results that conform to the unique identification coding rules, have complete fields, and are consistent with the ledger database of the back-end management and tracking module as trusted tableware identifiers. Combine all trusted tableware identifiers into a trusted tableware identifier set and count the total number of trusted tableware identifiers as the number of trusted tableware identifiers.

[0034] S33. Calculate the ratio of the change in the number of trusted tableware identifiers in the current period to the number of trusted tableware identifiers in the previous period;

[0035] The change ratio is the ratio of the difference between the two to the number of trusted tableware identifiers in the previous period. When the number of trusted tableware identifiers in the previous period is zero, the calculation is performed with one as the denominator.

[0036] The change ratio, along with the distribution of the received signal strength indication value and the reread request queue, constitute a set of state features;

[0037] S34. Pack the trusted tableware identifier set and the state feature set into a data package to form a complete output data package containing EPC code, TID number, timestamp of the unique record retained in the time window and antenna channel number and its statistical characteristics, received signal strength distribution, ratio of change of trusted tableware identifier quantity and reread request queue information.

[0038] Optionally, step S4 includes the following specific steps:

[0039] S41. Receive the complete output data packet, extract the distribution of received signal strength indication value, the change ratio of the number of trusted tableware identifiers, the historical beam direction angle and transmit power parameters, and the channel state information, and combine them into a state vector according to a fixed field order, which is used as the state input for the current reading cycle and input to the Actor network.

[0040] S42. Before the system runs online, the heuristic scanning and small-step power ladder strategy of the array antenna control module is called to collect the offline state-action trajectory set;

[0041] Under the heuristic scanning and small-step power ladder strategy, a heuristic strategy label corresponding to each state vector is generated based on the beam direction angle and transmit power parameter values ​​obtained from each set of state vectors.

[0042] Using the offline state-action trajectory set as supervised data, the Actor network is trained by behavior cloning so that the difference between the beam direction angle and transmission power output by the Actor network and the corresponding heuristic policy label is less than the preset error threshold. This completes the offline warm-up and serves as the initial Actor parameters for the improved TD3 algorithm.

[0043] S43. During the online phase, the Actor network receives the state vector of the current reading cycle and generates the raw action output of beam direction angle and transmit power.

[0044] S44. Perform range constraint mapping processing on the beam direction angle and transmit power output by the Actor network: trim the beam direction angle to the physically scannable range of the array antenna, trim the transmit power to the power range permitted by RF regulations, and generate constraint actions that meet the hardware implementation conditions and RF transmission specifications.

[0045] S45. The constrained action is differentially calculated from the beam direction angle and transmit power parameters executed by the array antenna control module in the previous reading cycle. The changes in beam direction angle and transmit power are calculated separately, and maximum rate limiting is applied to both. Any excess is clipped to obtain the safe action output. The above range constraint mapping and rate limiting processing are collectively referred to as the safe action layer, and its output is the safe action output.

[0046] S46. The safety action output is sent to the array antenna control module to perform array element phase and amplitude adjustment, and complete the radio frequency reading and signal acquisition of the current cycle;

[0047] S47. Collect the structured data records and complete output data packets generated in this cycle, construct a reward signal based on the periodic changes in the number of trusted tableware identifiers, the changes in beam direction angle and transmission power, and form an empirical sample with the state vector of the current cycle, the safety action output and the state vector of the next cycle.

[0048] S48. Write the experience samples into the experience replay pool; sample a small batch of samples from the experience replay pool, update the dual Critic network first, and then update the Actor network according to the delay update step size; add smoothing noise to the target Actor output when updating the target network.

[0049] S49. Repeat S41 to S48 in each reading cycle; when the change ratio of the number of trusted tableware identifiers calculated based on the complete output data packet is lower than the preset threshold, the security action layer triggers a rollback mechanism to roll back the current security action output to the most recent stable security action output, and continues to execute the online update process of S41 to S48 during the rollback.

[0050] Optionally, the online strategy refers to a dynamic control strategy in which the Actor network generates a set of decision rules for beam direction angle and transmit power parameters based on the state vector of the current reading cycle during real-time system operation. This set of rules is then modified and executed within each cycle using mechanisms such as range constraints, rate limiting, and safety action layers. This strategy can be continuously optimized during operation based on real-time acquired structured data records and complete output data packets. Through experience playback and network updates, it continuously improves decision accuracy and stability, thereby achieving adaptive adjustment of the array antenna beam and power.

[0051] Optionally, the behavior cloning training is a supervised learning-based policy parameter initialization method, the process of which is as follows: using the state vector in the offline state-action trajectory set as the network input, and the beam direction angle and transmit power parameters in the heuristic policy label corresponding to the state vector as the expected output, a supervised training sample set is constructed; the state vector is input into the Actor network, the beam direction angle and transmit power parameter values ​​output by the network are obtained, the difference between the output and the expected output is calculated, and the difference is used as the loss function for backpropagation and parameter update; the training process is repeated until the difference is less than a preset error threshold, thereby obtaining the initial Actor parameters that meet the accuracy requirements.

[0052] Optionally, step S5 includes the following specific steps:

[0053] S51. Receive the timestamp and antenna channel number of the trusted tableware identifier set and the complete output data packet in the edge computing transaction processing module, create a settlement session and record the session identifier;

[0054] S52. Based on each EPC code, query the ledger database and food database of the back-end management and tracking module to obtain the food identifier, pricing method, unit price parameter and confirm the price rule version;

[0055] S53. Generate a candidate billing item list. The candidate billing item list is a collection composed of billing item structures. The billing item structure includes: EPC code, TID number, food item identifier, unit price, quantity, timestamp, and antenna channel number. These are written into the billing item structure to form a set of items to be billed.

[0056] S54. Perform duplicate billing risk verification, compare the set of items to be billed with the billed details in the current settlement session and the preset time window; if the same EPC code exists in an unclosed session or a recently settled order, mark the duplicate billing risk and remove it or set it for review, and record the abnormal reason code.

[0057] S55. Calculate the amount payable according to the pricing rules, generate the item amount and the total bill amount, and summarize them into the amount field and item details;

[0058] S56. Generate billing information (order number, billing generation time, summary of trusted tableware identifier set [quantity and EPC code list], item details, amount field, and duplicate billing risk marker list) and cache it in the edge computing transaction processing module;

[0059] S57. Output the bill information and the final set of trusted tableware identifiers to the payment control module, and write the candidate bill items and duplicate billing risk records into the transaction details of the back-end management and tracking module for querying and reconciliation.

[0060] Optionally, step S6 includes the following specific steps:

[0061] S61. The payment control module receives the billing information and the set of trusted tableware identifiers, verifies the consistency of the order number, amount field, price rule version and session identifier, creates a payment session and locks the order number.

[0062] S62. Generate a payment instruction based on the payment channel parameters. The payment instruction includes the order number, total bill amount, merchant identifier, terminal identifier, and payment timeout time, and generate corresponding payment payload data.

[0063] S63. Send a payment instruction to the selected payment channel and listen for the channel's response, recording the transaction number and payment status;

[0064] S64. Upon successful payment, verify that the order number, merchant ID, and amount received match. After verification, generate a payment result record and mark the bill as paid.

[0065] S65. If payment fails or the payment timeout period is exceeded, retry within the maximum number of attempts according to the preset retry strategy; if it still fails, mark it as payment failure and record the failure reason code.

[0066] S66. Return the payment result receipt to the edge computing transaction processing module to update the bill status, and write the transaction record and trusted tableware identifier set into the transaction details and reconciliation data table of the backend management and tracking module; the transaction record includes the order number, payment channel transaction number, payment time, and amount fields;

[0067] S67. In the back-end management and tracking module, update the tableware circulation path and usage count based on the payment result, associate the corresponding EPC code transaction record with the order number, and generate operational statistics and abnormal record entries.

[0068] The beneficial effects of this invention are:

[0069] By introducing a multi-antenna polling strategy, the system can achieve orderly switching of multiple channels and dynamic optimization of coverage during RF reading, ensuring that tableware RFID tags receive relatively uniform signal illumination from different directions and positions. This significantly reduces missed readings caused by single-channel obstruction, spatial attenuation, or antenna directivity limitations, and provides a multi-view source of raw data for subsequent tag feature aggregation. Combined with an anti-collision reading mechanism, the system can dynamically allocate response time slots and adjust the Q value in high-concurrency scenarios with multiple tags responding simultaneously, effectively reducing data loss and misreads caused by signal collisions, and improving the integrity and accuracy of tag data acquisition. Relying on the artificial intelligence recognition and error correction module and the array antenna adaptive beam control mechanism, the system can optimize beam direction and transmit power parameters online while reading batch tags, forming a closed loop of reading-evaluation-adjustment. This ensures both high-throughput batch reading performance and high accuracy and stability of tag recognition, thereby achieving a higher degree of automation and lower manual intervention requirements in tableware batch recognition and transaction control scenarios. Attached Figure Description

[0070] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0071] Figure 1 This is a flowchart of the artificial intelligence-based ultra-high frequency radio frequency tableware batch identification and transaction control system proposed in this invention.

[0072] Figure 2 This is a system flowchart of the artificial intelligence-based ultra-high frequency radio frequency tableware batch identification and transaction control system proposed in this invention;

[0073] Figure 3 This is a schematic diagram of the adaptive beam control mechanism of the array antenna based on dual-Critic delay update in the artificial intelligence-based UHF radio frequency tableware batch identification and transaction control system proposed in this invention.

[0074] Figure 4 This is a schematic diagram of the improved TD3, an artificial intelligence-based ultra-high frequency radio frequency tableware batch identification and transaction control system proposed in this invention. Detailed Implementation

[0075] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0076] refer to Figure 1-4 An AI-powered UHF radio frequency tableware batch identification and transaction control system includes:

[0077] A tableware radio frequency tag module is used to embed ultra-high frequency RFID tags into the tableware body;

[0078] The radio frequency reading and signal acquisition module is used to output structured data records when tableware enters the reading area;

[0079] The artificial intelligence recognition and error correction module is used to preprocess the collected structured data records to generate a set of trustworthy tableware identifiers and a set of state features;

[0080] The array antenna control module is used to dynamically adjust the array antenna beam direction angle and transmit power parameter values ​​based on the dual-Critic delay update adaptive beam control mechanism.

[0081] The edge computing transaction processing module is used to input the set of trusted tableware identifiers into the edge computing transaction processing flow and perform duplicate billing risk verification.

[0082] The payment control module is used to trigger payment methods based on billing information to complete transaction settlement and return payment results;

[0083] The backend management and tracking module is used to generate operational statistics and anomaly record entries.

[0084] This invention achieves high-precision batch identification of tableware, dynamic beam optimization, real-time settlement, and full-process data tracking by embedding ultra-high frequency RFID tags into the tableware body and combining the coordinated operation of modules such as radio frequency reading and signal acquisition, artificial intelligence recognition and error correction, array antenna adaptive beam control, edge computing transaction processing, payment control, and back-end management and tracking. This significantly improves the identification accuracy, settlement efficiency, and traceability and intelligence level of operation management.

[0085] In this embodiment, the modules are interconnected using the following method:

[0086] S1. Embed an ultra-high frequency RFID tag in the body of the tableware and write and bind it, and establish a mapping relationship between the EPC code and the TID number;

[0087] S2. When the tableware enters the reading area, the radio frequency reading and signal acquisition process acquires the EPC code, TID number and original tag reading feature information in batches.

[0088] S3. Aggregate structured data records and filter trusted tableware identifiers to generate a set of trusted tableware identifiers and their change ratios. Combine the distribution of received signal strength indication values ​​and the reread request queue to construct a set of state features and form a complete output data packet.

[0089] S4. The adaptive beam control mechanism for array antenna based on dual-Critic delay update adopts the improved TD3 algorithm, uses offline behavior cloning initialization and online strategy to realize adaptive control of array antenna beam direction and transmit power, and updates the Actor network cyclically through safety action constraints, reward signal construction and experience playback mechanism.

[0090] S5. Input the set of trusted tableware identifiers into the edge computing transaction processing flow and perform duplicate billing risk verification.

[0091] S6. Complete the payment processing of the statement information and trusted tableware identifier set, and generate operational statistics and abnormal record entries in the back-end management and tracking module.

[0092] This invention achieves efficient batch identification of tableware, dynamic beam optimization, and real-time transaction settlement by embedding ultra-high frequency RFID tags into the tableware body and establishing a mapping relationship between EPC codes and TID numbers. It combines steps such as radio frequency reading and signal acquisition, trusted tableware identification screening, adaptive beam control of array antenna based on dual-critic delay update, edge computing transaction processing, and payment processing. This significantly improves the identification accuracy, transaction processing efficiency, and traceability of operational data, helps reduce the risk of missed identification and misbilling, and enhances the overall intelligence level of the system.

[0093] In this embodiment, step S1 includes the following specific steps:

[0094] S11. Select UHF RFID tags based on the material type and geometric dimensions of the tableware body, and determine the tag embedding position;

[0095] S12. Process the installation slot at the tag embedding position, place the UHF RFID tag in it and fix it with epoxy potting material, and perform a sealing and washing cycle test after curing.

[0096] S13. Define the field structure and length configuration for the unique identifier code of tableware;

[0097] S14. At the coding station, use a calibrated RFID reader to write the unique identification code of the tableware into the EPC storage area of ​​the tag, read back to check the consistency of the written content, read the TID number of the tag and form a mapping relationship between the EPC code and the TID number.

[0098] S15. Set an access password for the tag, enable write protection for the EPC storage area, write production date, capacity and specification parameter information to the user area as needed, and maintain the read-only attribute of the TID area.

[0099] S16. Perform a tag performance sampling inspection on each piece of tableware that has completed the writing of the unique identifier, including sampling inspection of the stability of the reading distance and the received signal strength indication value, record the reading distance range, signal strength fluctuation range, antenna channel information and test time, and conduct supplementary sampling inspections for metal proximity and humid and hot conditions.

[0100] S17. Record the tableware type, capacity, batch, EPC code, TID number, packaging process batch number and test results to the ledger database of the background management and tracking module, complete the filing of tableware and label binding relationship and generate a searchable index.

[0101] This invention selects and embeds UHF RFID tags based on the material and geometry of tableware, and combines these steps with sealing and curing treatment, unique identification coding and protection, tag performance sampling inspection, and filing of the binding relationship between tableware and tags. This ensures the stability and readability of the tags in complex usage environments, and achieves reliable identification and efficient traceability management of tableware. As a result, it significantly improves the accuracy of subsequent batch identification and transaction processing and the long-term stability of the system operation.

[0102] In this embodiment, the radio frequency reading and signal acquisition process is as follows:

[0103] When the tableware with the unique identifier written and bound enters the reading area of ​​the radio frequency reading and signal acquisition module, the radio frequency reading and signal acquisition module transmits an ultra-high frequency radio frequency signal with a working frequency band of 860MHz to 960MHz through a multi-channel array antenna to excite the ultra-high frequency RFID tag embedded in the tableware body, so that the tag's EPC storage area returns the unique identifier code, TID number and user area extended information.

[0104] The system synchronously collects tag reading characteristic information and current array antenna beam direction and transmit power parameters from each tag. Combining a multi-antenna polling strategy and an anti-collision reading mechanism, it acquires batch tag data and organizes the acquired EPC code, TID number, and tag reading characteristic information into structured data records. Tag reading characteristic information includes signal strength indication value, timestamp, and antenna channel information.

[0105] In this embodiment, the multi-antenna polling strategy refers to the process in the RF reading and signal acquisition module where, to improve tag coverage and reading success rate, multiple independent RF channels of the array antenna are sequentially activated according to a preset polling order. Within each channel's activation period, the coverage area is progressively scanned based on beam direction angle and transmit power parameters. When the tag reading result of the current channel stabilizes or reaches a preset acquisition duration threshold, the system switches to the next channel to continue reading until all channels are polled. This strategy can balance signal coverage in different directions and areas in densely distributed multi-tag scenarios, reduce the probability of missed readings due to single-channel obstruction or signal attenuation, and provide multi-view raw data support for subsequent signal feature aggregation and trusted identifier generation.

[0106] In this embodiment, the anti-collision reading mechanism addresses signal conflicts caused by multiple tags responding simultaneously during RF reading and signal acquisition. It employs a time-slot anti-collision algorithm based on the EPCglobalClass-1Gen2 protocol and an adaptive Q-value adjustment strategy. By dynamically allocating and adjusting tag response time slots during the reading session, the probability of multiple tags transmitting data simultaneously within the same time slot is reduced. Simultaneously, combined with real-time monitoring of the received signal strength indicator, tags with high signal quality and stable responses are prioritized, reducing misreads and duplicate reads caused by signal superposition or interference. This mechanism can work in conjunction with a multi-antenna polling strategy to optimize throughput and improve data accuracy during high-concurrency tag reading.

[0107] In this embodiment, step S3 includes the following specific steps:

[0108] S31. Receive and buffer the structured data records output by the RF reading and signal acquisition module, aggregate the preprocessed structured data records according to EPC encoding, and generate tag reading feature information for each EPC encoding in each antenna channel.

[0109] S32. Define the tag records in the aggregation results that conform to the unique identification coding rules, have complete fields, and are consistent with the ledger database of the back-end management and tracking module as trusted tableware identifiers. Combine all trusted tableware identifiers into a trusted tableware identifier set and count the total number of trusted tableware identifiers as the number of trusted tableware identifiers.

[0110] S33. Calculate the ratio of the change in the number of trusted tableware identifiers in the current period to the number of trusted tableware identifiers in the previous period;

[0111] The change ratio is the ratio of the difference between the two to the number of trusted tableware identifiers in the previous period. When the number of trusted tableware identifiers in the previous period is zero, the calculation is performed with one as the denominator.

[0112] The change ratio, along with the distribution of the received signal strength indication value and the reread request queue, constitute a set of state features;

[0113] S34. Pack the trusted tableware identifier set and the state feature set into a data package to form a complete output data package containing EPC code, TID number, timestamp of the unique record retained in the time window and antenna channel number and its statistical characteristics, received signal strength distribution, ratio of change of trusted tableware identifier quantity and reread request queue information.

[0114] This invention aggregates and processes structured data records output by the RFID reading and signal acquisition module, filters trusted tableware identifiers, constructs a set of state features by combining the proportion of changes in the number of trusted tableware identifiers, the distribution of received signal strength indicators, and the reread request queue, and packages key information to generate a complete output data packet. This achieves high-precision filtering, dynamic state quantification, and structured encapsulation of tableware RFID tag data, thereby significantly improving the stability and accuracy of subsequent antenna beam control and transaction processing.

[0115] In this embodiment, the preprocessing includes:

[0116] The cached structured data records are processed for tag conflict residue. Based on the adjacency of the conflict flag bit marked by the anti-collision reading mechanism and the timestamp, abnormal responses within the same time slice are eliminated and merged, retaining records that can be parsed and have complete fields.

[0117] Perform time window deduplication. Set a deduplication time window. For multiple return records of the same EPC code within the window, sort them from high to low according to the received signal strength indication value. Combine the timestamp sequence with the consistency of the antenna channel number to select the unique record to retain and delete the other duplicate records.

[0118] Perform misread verification by comparing the EPC code and TID number in each record with the correspondence established in the ledger database of the back-end management and tracking module. Records that do not match are marked as misread and removed. Records whose field format does not conform to the unique identifier coding rules are invalidated.

[0119] Perform a missed read detection and check the consistency of the EPC code occurrence sequence in adjacent scan cycles. When an EPC code exists in both the previous and next cycles but is missing in the current cycle, add the EPC code to the reread request queue, along with the antenna channel number of the most recent occurrence and the corresponding array antenna beam direction and transmit power parameters.

[0120] The records after deduplication and verification are aggregated according to EPC codes to generate statistical information and timestamp sequence information of the received signal strength indication value for each EPC code in each antenna channel, forming a description of the distribution and time distribution of the received signal strength indication value, and retaining the corresponding TID number and the most recent antenna channel number.

[0121] This invention performs tag conflict residue processing, time window deduplication, misread verification, and missed read detection sequentially on the cached structured data records, and combines EPC encoding to generate a distribution of received signal strength indication values ​​and a time distribution description. This effectively removes abnormal and redundant records, corrects misreads, and marks missed tableware tags, thereby significantly improving the integrity and accuracy of tableware RFID tag data and providing a high-quality data foundation for subsequent status feature construction and transaction processing.

[0122] In this embodiment, step S4 includes the following specific steps:

[0123] S41. Receive the complete output data packet, extract the distribution of received signal strength indication value, the change ratio of the number of trusted tableware identifiers, the historical beam direction angle and transmit power parameters, and the channel state information, and combine them into a state vector according to a fixed field order, which is used as the state input for the current reading cycle and input to the Actor network.

[0124] S42. Before the system runs online, the heuristic scanning and small-step power ladder strategy of the array antenna control module is called to collect the offline state-action trajectory set;

[0125] Under the heuristic scanning and small-step power ladder strategy, a heuristic strategy label corresponding to each state vector is generated based on the beam direction angle and transmit power parameter values ​​obtained from each set of state vectors.

[0126] Using the offline state-action trajectory set as supervised data, the Actor network is trained by behavior cloning so that the difference between the beam direction angle and transmission power output by the Actor network and the corresponding heuristic policy label is less than the preset error threshold. This completes the offline warm-up and serves as the initial Actor parameters for the improved TD3 algorithm.

[0127] S43. During the online phase, the Actor network receives the state vector of the current reading cycle and generates the raw action output of beam direction angle and transmit power.

[0128] S44. Perform range constraint mapping processing on the beam direction angle and transmit power output by the Actor network: trim the beam direction angle to the physically scannable range of the array antenna, trim the transmit power to the power range permitted by RF regulations, and generate constraint actions that meet the hardware implementation conditions and RF transmission specifications.

[0129] S45. The constrained action is differentially calculated from the beam direction angle and transmit power parameters executed by the array antenna control module in the previous reading cycle. The changes in beam direction angle and transmit power are calculated separately, and maximum rate limiting is applied to both. Any excess is clipped to obtain the safe action output. The above range constraint mapping and rate limiting processing are collectively referred to as the safe action layer, and its output is the safe action output.

[0130] S46. The safety action output is sent to the array antenna control module to perform array element phase and amplitude adjustment, and complete the radio frequency reading and signal acquisition of the current cycle;

[0131] S47. Collect the structured data records and complete output data packets generated in this cycle, construct a reward signal based on the periodic changes in the number of trusted tableware identifiers, the changes in beam direction angle and transmission power, and form an empirical sample with the state vector of the current cycle, the safety action output and the state vector of the next cycle.

[0132] S48. Write the experience samples into the experience replay pool; sample a small batch of samples from the experience replay pool, update the dual Critic network first, and then update the Actor network according to the delay update step size; add smoothing noise to the target Actor output when updating the target network.

[0133] S49. Repeat S41 to S48 in each reading cycle; when the change ratio of the number of trusted tableware identifiers calculated based on the complete output data packet is lower than the preset threshold, the security action layer triggers a rollback mechanism to roll back the current security action output to the most recent stable security action output, and continues to execute the online update process of S41 to S48 during the rollback.

[0134] This invention introduces a dynamic optimization mechanism combining offline behavior cloning preheating and an improved TD3 algorithm during the adaptive beam control process of the array antenna. It constructs a state vector using complete output data packets, generates actions through an Actor network, and performs range constraints and rate limiting at the safety action layer. This ensures that the adjustment of beam direction and transmit power meets both hardware and regulatory requirements while maintaining operational safety. Combined with the continuous updates of reward signal construction and experience playback, it achieves efficient adaptive adjustment of the array antenna beam direction angle and transmit power, effectively improving the stable reading rate of trusted tableware identifiers and the reliability of system operation.

[0135] In this embodiment, the improved TD3 algorithm includes the following specific structure and steps:

[0136] The system receives the complete output data packet, extracts the distribution of received signal strength indication value, the proportion of changes in the number of trusted tableware identifiers, the historical beam direction angle, the transmit power parameter value, and the channel state information, and combines them in a fixed field order to form a state vector, which is used as the state input for the current reading cycle and input to the Actor network.

[0137] Before the system goes online, the heuristic scanning and small-step power ladder strategy of the array antenna control module is invoked to collect offline state action trajectory sets. A corresponding heuristic policy label, including beam direction angle and transmit power parameter values, is generated for each set of state vectors. Using the offline state action trajectory set as supervised data, behavioral cloning training of the Actor network is performed, ensuring that the difference between the beam direction angle and transmit power output by the Actor network on each training sample and the heuristic policy label is less than a preset error threshold. This difference is used as the initial parameters of the Actor network.

[0138] The Actor network receives the state vector of the current reading cycle and outputs the raw action results of beam direction angle and transmit power based on the current online strategy.

[0139] A range constraint mapping operation is performed on the beam direction angle and transmit power of the original action output. The beam direction angle is cropped to the physically scannable range of the array antenna, and the transmit power is cropped to the power range permitted by RF regulations, generating an action result that meets the constraints. This result is then differentiated from the actual action executed in the previous cycle to calculate the changes in beam direction angle and transmit power, respectively. A maximum rate limiting threshold is applied to crop the portion exceeding the rate of change limit, and a safe action is output.

[0140] Safety actions are used to control the array antenna to perform real-time adjustments to the phase and amplitude of the array elements, completing the radio frequency reading and signal acquisition tasks for the current cycle.

[0141] Collect the structured data records and complete output data packets generated in the current cycle, construct reward signals based on the change ratio of the number of trusted tableware identifiers, the change in beam direction angle, and the change in transmission power, and combine the current cycle's state vector, safety action output, and the next cycle's state vector to form experience samples, which are then written into the experience playback pool.

[0142] A small batch of empirical samples is sampled from the empirical replay pool. First, the evaluation target value update operation of the dual Critic network is performed, and then the parameters of the Actor network are updated using a delayed step mechanism. When updating the target network, smooth noise is added to the output action of the target Actor network.

[0143] The above steps are repeated as the reading cycle continues. When the change ratio of the number of trusted tableware identifiers calculated based on the complete output data packet is lower than the preset threshold, the security action layer triggers a rollback mechanism to replace the security action generated in the current cycle with the most recent stable security action, and continues to execute the online update and optimization process of the Actor network.

[0144] The improved TD3 algorithm proposed in this invention introduces offline heuristic policy label-guided behavior cloning training to initialize the Actor network, and combines it with an online state vector-driven adaptive action generation mechanism to achieve dynamic optimization and adjustment of the array antenna beam direction angle and transmit power. By constructing a safe action layer through action range pruning and change rate limiting, the risk of physical control exceeding limits is effectively avoided. Furthermore, a reward signal is constructed based on the change in the number of trusted tableware identifiers, and a continuous Actor-Critic joint update and policy rollback mechanism are implemented. This ensures reading stability while improving the dynamic robustness and recognition accuracy of the system in complex electromagnetic environments.

[0145] In this embodiment, the specific process of the heuristic scanning is as follows:

[0146] During the offline warm-up phase, the system invokes the array antenna control module to execute a heuristic scanning process. Within the physically scannable range of the array antenna, the beam direction angles are partitioned and traversed based on a preset search order and historical experience priorities. For multiple direction angles within each partition, the system rapidly selects a set of direction angles with high coverage efficiency and signal quality by progressively scanning and evaluating the coverage and signal strength distribution in real time. The application of heuristic scanning allows state-action samples to be concentrated in the optimal direction area, reducing the number of traversals of inefficient directions and improving the usability and representativeness of the offline trajectory set.

[0147] In this embodiment, the specific process of the small-step power ladder strategy is as follows:

[0148] After heuristically selecting the beam direction angle, the system executes a small-step power ladder strategy at each preferred direction angle, gradually increasing the transmit power parameter value with a small power increment step. After each power adjustment, the distribution of received signal strength indication values ​​and the reading status of trusted tableware identifiers are collected in real time until the power limit permitted by radio frequency regulations is reached or the reading performance stabilizes. This strategy can obtain action effect samples at different power levels while ensuring signal stability and security, forming power parameter labels that correspond one-to-one with the state vector, enriching the power dimension features of the offline state-action trajectory set.

[0149] In this embodiment, the safety action layer is an execution unit that performs safety constraints and amplitude limiting on the beam direction angle and transmit power output by the Actor network during the adaptive beamforming process of the array antenna. Its main function is to map the original actions to the physical scannable range of the array antenna and the power range permitted by radio frequency regulations, and to apply a maximum rate limit to the amount of action change in adjacent readout cycles, pruning any portion exceeding the limit. This ensures that the antenna beamforming process complies with hardware safety boundaries and regulatory requirements, preventing system instability or illegal operation caused by excessive adjustments or overpowered transmission.

[0150] In this embodiment, step S5 includes the following specific steps:

[0151] S51. Receive the timestamp and antenna channel number of the trusted tableware identifier set and the complete output data packet in the edge computing transaction processing module, create a settlement session and record the session identifier;

[0152] S52. Based on each EPC code, query the ledger database and food database of the back-end management and tracking module to obtain the food identifier, pricing method, unit price parameter and confirm the price rule version;

[0153] S53. Generate a candidate billing item list. The candidate billing item list is a collection composed of billing item structures. The billing item structure includes: EPC code, TID number, food item identifier, unit price, quantity, timestamp, and antenna channel number. These are written into the billing item structure to form a set of items to be billed.

[0154] S54. Perform duplicate billing risk verification, compare the set of items to be billed with the billed details in the current settlement session and the preset time window; if the same EPC code exists in an unclosed session or a recently settled order, mark the duplicate billing risk and remove it or set it for review, and record the abnormal reason code.

[0155] S55. Calculate the amount payable according to the pricing rules, generate the item amount and the total bill amount, and summarize them into the amount field and item details;

[0156] S56. Generate billing information (order number, billing generation time, summary of trusted tableware identifier set [quantity and EPC code list], item details, amount field, and duplicate billing risk marker list) and cache it in the edge computing transaction processing module;

[0157] S57. Output the bill information and the final set of trusted tableware identifiers to the payment control module, and write the candidate bill items and duplicate billing risk records into the transaction details of the back-end management and tracking module for querying and reconciliation.

[0158] This invention automates the entire process of tableware transaction settlement, from data collection and risk control to bill generation and reconciliation record writing, by sequentially completing the reception of trusted tableware identifier sets and reading parameters and the creation of settlement sessions in the edge computing transaction processing module; real-time acquisition of food information and pricing rules; structured generation of bill items and double billing risk verification; calculation of payable amount and cached output of bill information; and synchronizing bill data and risk records to the backend management and tracking module. This effectively improves the real-time performance, accuracy and security of transaction processing.

[0159] In this embodiment, the duplicate billing risk verification process refers to comparing the currently generated set of items to be billed with the billed details in the current settlement session and the preset time window in the edge computing transaction processing process. When the same EPC code is detected to exist in an unclosed settlement session or a recently settled order, the system determines that there is a duplicate billing risk, sets the item to the review status, and records the corresponding abnormal reason code for subsequent transaction detail tracking and manual review.

[0160] In this embodiment, step S6 includes the following specific steps:

[0161] S61. The payment control module receives the billing information and the set of trusted tableware identifiers, verifies the consistency of the order number, amount field, price rule version and session identifier, creates a payment session and locks the order number.

[0162] S62. Generate a payment instruction based on the payment channel parameters. The payment instruction includes the order number, total bill amount, merchant identifier, terminal identifier, and payment timeout time, and generate corresponding payment payload data.

[0163] S63. Send a payment instruction to the selected payment channel and listen for the channel's response, recording the transaction number and payment status;

[0164] S64. Upon successful payment, verify that the order number, merchant ID, and amount received match. After verification, generate a payment result record and mark the bill as paid.

[0165] S65. If payment fails or the payment timeout period is exceeded, retry within the maximum number of attempts according to the preset retry strategy; if it still fails, mark it as payment failure and record the failure reason code.

[0166] S66. Return the payment result receipt to the edge computing transaction processing module to update the bill status, and write the transaction record and trusted tableware identifier set into the transaction details and reconciliation data table of the backend management and tracking module; the transaction record includes the order number, payment channel transaction number, payment time, and amount fields;

[0167] S67. In the back-end management and tracking module, update the tableware circulation path and usage count based on the payment result, associate the corresponding EPC code transaction record with the order number, and generate operational statistics and abnormal record entries.

[0168] This invention utilizes a payment control module to verify the consistency between bill information and a set of trusted tableware identifiers, generate and send payment instructions, monitor and verify payment receipts, and execute a preset retry strategy when payment fails. Simultaneously, it returns a payment result receipt to update the bill status and binds the transaction record with the tableware identifier to the backend management and tracking module, further updating the tableware circulation path and usage frequency. This constructs a secure, traceable, and anomaly-recording end-to-end payment processing mechanism for tableware transactions, effectively improving transaction accuracy, stability, and operational efficiency.

[0169] In this embodiment, the retry strategy is specifically set as follows: when payment fails or the payment timeout period is exceeded, the payment control module will retry a maximum of 3 times within the same payment session for a single order; the fixed time interval between two adjacent retries is 5 seconds, and idempotency verification and order number locking operations are performed before each retry to prevent duplicate payments; retry conditions include no receipt from the payment channel, unknown payment status, network interruption, or failure reason codes that are retryable; once any retry is successful and the received amount and merchant identifier are verified, subsequent retries are immediately stopped; if three consecutive retries still fail, the order is marked as a payment failure and the failure reason code is recorded.

[0170] This invention introduces a retry strategy with a maximum number of attempts, fixed intervals, and idempotency verification in the event of payment failure or timeout. This ensures that retry operations within the same payment session for a single order can effectively address recoverable failures such as network anomalies and lack of channel receipts, while also preventing the risk of duplicate payments. Furthermore, it immediately terminates subsequent attempts after a successful retry and accurately records the reason code after multiple failures, achieving high reliability and traceability in the payment process and contributing to improved stability and security of transaction settlement.

[0171] Example 1:

[0172] To verify the feasibility and effectiveness of this invention, it was applied to the smart tableware management and transaction settlement scenario of a large chain restaurant. During peak hours, this restaurant handles over 30,000 pieces of tableware daily, including bowls, plates, cups, trays, and other categories. Each type of tableware requires multiple passes through an RFID channel for status confirmation and transaction association during cleaning, delivery, serving, and recycling. However, traditional RFID reading methods based on fixed power and fixed beams are prone to missed reads, misreads, and duplicate billing in complex environments such as densely stacked tableware, inconsistent tag orientation, and interference from metal kitchenware. This directly affects the accuracy of tableware turnover statistics and transaction settlement. Especially in buffet or weigh-based billing systems, repeated misreading of tableware tags can easily lead to duplicate charges, causing customer complaints and difficulties in financial reconciliation.

[0173] In this embodiment, the company embeds an UHF RFID tag into each piece of tableware and writes a unique EPC code and TID number into each tag, establishing a mapping relationship with information such as tableware type, capacity, batch, and production process, which is stored in the ledger database of the back-end management and tracking module. In tableware usage scenarios, when the tableware enters the RFID reading area of ​​the cashier or self-checkout area, the array antenna control module dynamically adjusts the beam direction and transmission power according to the improved TD3 algorithm, and limits the power and directional rate through a safety action layer to adapt to changes in tableware stacking and tag orientation. The RFID reading and signal acquisition module acquires the EPC code, TID number, and received signal strength indication value in real time, and combines multi-antenna polling and anti-collision mechanisms to batch collect tag data, generating structured data records.

[0174] The AI-powered identification and error correction module performs conflict and residual processing, time-window deduplication, misread verification, and missed read detection on structured data records, generating a set of trusted tableware identifiers and a set of status features. The edge computing transaction processing module utilizes the trusted tableware identifier set to perform duplicate billing risk verification, generating bill entries for eligible tableware records and calculating the amount due. On peak test days, the system completed batch identification of over 2500 pieces of tableware within 30 seconds on a single channel, reducing the duplicate billing risk rate to 0.05%, a significant decrease compared to the traditional fixed-power reading mode (duplicate billing risk rate of 0.9%). Upon receiving the billing information, the payment control module executes payment instruction generation and sending operations according to a preset retry strategy (maximum 3 retries, 5-second intervals) and idempotency verification. After successful payment, it updates the transaction record and tableware flow path to the backend management and tracking module. Statistical results show that during a week of continuous operation, the transaction success rate reached 99.94%, and the payment failure rate due to network fluctuations was controlled within 0.03%.

[0175] In this scenario, the present invention solves the problems of missed readings, misreadings, and duplicate billing that easily occur in batch identification of tableware, and significantly improves the reading accuracy through dynamic beam control and intelligent error correction mechanisms; at the same time, in the payment process, the uniqueness and reliability of settlement data are ensured through secure retries and idempotency control. Specific data are shown in Table 1:

[0176] Table 1. Comparison of performance between bulk tableware identification and transaction settlement.

[0177]

[0178] Compared with the system before the upgrade, the accuracy of tableware circulation statistics has increased from 97.8% to 99.96%, and the number of transaction reconciliation discrepancies has been reduced by more than 92%. This has effectively reduced the cost of manual review and customer complaints, and brought enterprises a stable transaction settlement experience and efficient tableware lifecycle management capabilities.

[0179] In a seven-day comparative test, the performance of the proposed solution and the traditional fixed-beam RFID reading solution was statistically analyzed in key indicators such as tableware identification accuracy, duplicate billing risk rate, payment success rate, and number of reconciliation discrepancies. The test covered three time periods: morning peak, lunch peak, and dinner peak, and simulated metal interference and humid environments to verify the system's stability and robustness under complex conditions. The test results are shown in the following paragraphs:

[0180] Table 1 shows that, under the same total turnover of tableware, the present invention maintains a tableware recognition accuracy rate of over 99.9%, with a missed reading rate of no more than 0.04% during peak hours, while the traditional solution has a missed reading rate of over 1.2% during peak hours; the risk rate of duplicate billing is reduced from 0.9% to 0.05%; the payment success rate is increased to 99.94%, and the number of reconciliation discrepancies decreases from an average of 42 per day to less than 4; in stress tests simulating humid and metal-adjacent environments, the recognition accuracy of the present invention decreases by less than 0.15%, while the traditional solution decreases by more than 1.6%. These data fully demonstrate the high precision, high reliability, and adaptability advantages of the present invention in batch tableware recognition and transaction settlement.

[0181] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An AI-based UHF radio frequency tableware batch identification and transaction control system, characterized in that, include: A tableware radio frequency tag module is used to embed ultra-high frequency RFID tags into the tableware body; The radio frequency reading and signal acquisition module is used to output structured data records when tableware enters the reading area; The artificial intelligence recognition and error correction module is used to preprocess the collected structured data records to generate a set of trustworthy tableware identifiers and a set of state features; The array antenna control module is used to dynamically adjust the array antenna beam direction angle and transmit power parameter values ​​based on the dual-Critic delay update adaptive beam control mechanism. The edge computing transaction processing module is used to input the set of trusted tableware identifiers into the edge computing transaction processing flow and perform duplicate billing risk verification. The payment control module is used to trigger payment instructions based on billing information, complete transaction settlement, and return payment results. The backend management and tracking module is used to generate operational statistics and anomaly record entries.

2. The AI-based UHF radio frequency tableware batch identification and transaction control system according to claim 1, characterized in that, The modules are connected in the following way: S1. Embed an ultra-high frequency RFID tag in the body of the tableware and write and bind it, and establish a mapping relationship between the EPC code and the TID number; S2. When the tableware enters the reading area, the radio frequency reading and signal acquisition process acquires the EPC code, TID number and original tag reading feature information in batches. S3. Aggregate structured data records and filter trusted tableware identifiers to generate a set of trusted tableware identifiers and their change ratios. Combine the distribution of received signal strength indication values ​​and the reread request queue to construct a set of state features and form a complete output data packet. S4. The adaptive beam control mechanism for array antenna based on dual-Critic delay update adopts the improved TD3 algorithm, uses offline behavior cloning initialization and online strategy to realize adaptive control of array antenna beam direction and transmit power, and updates the Actor network cyclically through safety action constraints, reward signal construction and experience playback mechanism. S5. Input the set of trusted tableware identifiers into the edge computing transaction processing flow and perform duplicate billing risk verification. S6. Complete the payment processing of the statement information and trusted tableware identifier set, and generate operational statistics and abnormal record entries in the back-end management and tracking module.

3. The AI-based UHF radio frequency tableware batch identification and transaction control system according to claim 2, characterized in that, S1 includes the following specific steps: S11. Select UHF RFID tags based on the material type and geometric dimensions of the tableware body, and determine the tag embedding position; S12. Process the installation slot at the tag embedding position, place the UHF RFID tag in it and fix it with epoxy potting material, and perform a sealing and washing cycle test after curing. S13. Define the field structure and length configuration for the unique identifier code of tableware; S14. At the coding station, use a calibrated RFID reader to write the unique identification code of the tableware into the EPC storage area of ​​the tag, read back to check the consistency of the written content, read the TID number of the tag and form the correspondence between the EPC code and the TID number. S15. Set an access password for the tag, enable write protection for the EPC storage area, write production date, capacity and specification parameter information to the user area as needed, and maintain the read-only attribute of the TID area. S16. Perform a random inspection of the label performance of each piece of tableware for which the unique identifier has been written. S17. Record the tableware type, capacity, batch, EPC code, TID number, packaging process batch number and test results to the ledger database of the background management and tracking module, complete the filing of tableware and label binding relationship and generate a searchable index.

4. The AI-based UHF radio frequency tableware batch identification and transaction control system according to claim 2, characterized in that, The radio frequency reading and signal acquisition process is as follows: When the tableware that has completed the writing and binding of the tableware's unique identifier enters the reading area of ​​the radio frequency reading and signal acquisition module, the system synchronously acquires the original tag reading feature information of each tag and the current array antenna beam direction and transmission power parameters. Combining the multi-antenna polling strategy and the anti-collision reading mechanism, the system completes the acquisition of batch tag data and organizes the acquired EPC code, TID number and original tag reading feature information into a structured data record.

5. The AI-based UHF radio frequency tableware batch identification and transaction control system according to claim 2, characterized in that, S3 includes the following specific steps: S31. Receive and buffer the structured data records output by the RF reading and signal acquisition module, aggregate the preprocessed structured data records according to EPC encoding, and generate aggregated tag reading feature information for each EPC encoding in each antenna channel. S32. Define the tag records in the aggregated tag reading feature information that conform to the unique identification coding rules, have complete fields, and are consistent with the ledger database of the back-end management and tracking module as trusted tableware identifiers. Combine all trusted tableware identifiers into a trusted tableware identifier set and count the total number of trusted tableware identifiers as the number of trusted tableware identifiers. S33. Calculate the change ratio between the number of trusted tableware identifiers in the current period and the number of trusted tableware identifiers in the previous period, and combine the change ratio with the distribution of received signal strength indication values ​​to form a state feature set. S34. Pack the trusted tableware identifier set and the state feature set into a complete output data packet.

6. The AI-based UHF radio frequency tableware batch identification and transaction control system according to claim 2, characterized in that, S4 includes the following specific steps: S41. Receive the complete output data packet, combine it into a state vector according to a fixed field order, use it as the state input for the current reading cycle, and input it into the Actor network as the input basis for online policy execution. S42. Before the system runs online, the heuristic scanning and small-step power ladder strategy of the array antenna control module is called to collect the offline state action trajectory set; Under the heuristic scanning and small-step power ladder strategy, a heuristic strategy label corresponding to each state vector is generated based on the beam direction angle and transmit power parameter values ​​obtained from each set of state vectors. Using the offline action trajectory set as supervised data, the Actor network is trained by behavior cloning so that the difference between the beam direction angle and transmission power output by the Actor network and the corresponding heuristic policy label is less than the preset error threshold, which is used as the initial Actor parameters of the improved TD3 algorithm. S43. During the online phase, the Actor network receives the state vector of the current reading cycle and generates the original action output of beam direction angle and transmission power based on the current online strategy. S44. Perform range constraint mapping processing on the beam direction angle and transmit power output by the Actor network: trim the beam direction angle to the physically scannable range of the array antenna, trim the transmit power to the power range permitted by RF regulations, and generate the constraint action. S45. The constrained action is compared with the beam direction angle and transmit power parameters executed by the array antenna control module in the previous reading cycle. The changes in beam direction angle and transmit power are calculated respectively. Maximum rate limiting is applied to both. The excess is clipped to obtain the safe action output, forming the actual execution result of the online strategy in this cycle. S46. The safety action output is sent to the array antenna control module to perform array element phase and amplitude adjustment, and complete the radio frequency reading and signal acquisition of the current cycle; S47. Collect the structured data records and complete output data packets generated in this cycle, construct a reward signal based on the periodic changes in the number of trusted tableware identifiers, the changes in beam direction angle and transmission power, and form an empirical sample with the state vector of the current cycle, the safety action output and the state vector of the next cycle. S48. Write the experience samples into the experience playback pool; Sample a small batch of samples from the experience replay pool, update the dual Critic network first, then update the Actor network by the delay update step size, and add smooth noise to the target Actor output when updating the target network; S49. Repeat S41 to S48 in each reading cycle; when the change ratio of the number of trusted tableware identifiers calculated based on the complete output data packet is lower than the preset threshold, the security action layer triggers a rollback mechanism to roll back the current security action output to the most recent stable security action output, and continues to execute the online update process of S41 to S48 during the rollback.

7. The AI-based UHF radio frequency tableware batch identification and transaction control system according to claim 2, characterized in that, S5 includes the following specific steps: S51. In the edge computing transaction processing flow, receive the timestamp and antenna channel number of the trusted tableware identifier set and the complete output data packet, create a settlement session and record the session identifier; S52. Based on each EPC code, query the ledger database of the back-end management and tracking module to obtain the food item identifier, pricing method, unit price parameters, and confirm the price rule version; S53. Generate a candidate billing item list. The candidate billing item list is a collection composed of billing item structures, forming a set of items to be billed. S54. Perform duplicate billing risk verification: Compare the set of items to be billed with the billed details in the current settlement session and the preset time window; When the same EPC code exists in both an open session and a recently settled order, mark the risk of duplicate billing and set it for review, and record the exception reason code; S55. Calculate the amount payable according to the pricing rules, generate the item amount and the total bill amount, and summarize them into the amount field and item details; S56. Generate billing information and cache it in the edge computing transaction processing module; S57. Output the bill information and the final set of trusted tableware identifiers to the payment control module, and write the candidate bill items and the exception reason code into the transaction details of the back-end management and tracking module for querying and reconciliation.

8. The AI-based UHF radio frequency tableware batch identification and transaction control system according to claim 2, characterized in that, S6 includes the following specific steps: S61. Execute the payment processing flow, receive the bill information and the set of trusted tableware identifiers, verify the consistency of the bill information, create a payment session and lock the order number; S62. Generate payment instructions based on payment channel parameters and generate corresponding payment payload data; S63. Send a payment instruction to the selected payment channel and listen for the channel's response, recording the transaction number and payment status; S64. Upon successful payment, verify that the order number, merchant ID, and amount received match. After verification, generate a payment result record and mark the bill as paid. S65. If payment fails or the payment timeout period is exceeded, retry within the maximum number of attempts according to the preset retry strategy; if it still fails, mark it as payment failure and record the failure reason code. S66. Return the payment result receipt to the edge computing transaction processing module to update the bill status, and write the transaction record and trusted tableware identifier set into the transaction details and reconciliation data table of the back-end management and tracking module. S67. In the back-end management and tracking module, update the tableware circulation path and usage count based on the payment result, associate the corresponding EPC code transaction record with the order number, and generate operational statistics and abnormal record entries.

Images