A precious metal video storage traceability system and method based on information kernel reconstruction and filter net learning

Abstract

This invention belongs to the field of anti-counterfeiting and traceability technology, and provides a precious metal video evidence preservation and traceability system and method based on information kernel reconstruction and filter learning. The method establishes a unique traceability identifier for precious metal objects and constructs an inverted residual three-segment information receiving structure consisting of an import receiving module, a dimension-upgrading receiving module, and an amplification receiving module. Important information kernels are formed through core information center placement, peripheral information placement, core priority reception, waiting replacement, and matrix transposition. These kernels are then input into a filter neural network model via an activation function, where the important information kernels undergo elliptical segmentation, center learning, flanking contraction, and merging to generate filter information kernels. Video and business data from weighing, label binding, packaging, warehousing, and handover nodes are collected and linked with the filter information kernels to form a complete video evidence preservation chain. This invention improves the efficiency of precious metal traceability identification, node video verification capability, and anti-tampering capability.

Description

Technical Field

[0001] This invention belongs to the field of anti-counterfeiting technology, specifically a precious metal video evidence preservation and traceability system and method based on information kernel reconstruction and filter learning. Background Technology

[0002] The anti-counterfeiting and traceability system for the precious metals supply chain is mainly used to record and verify the authenticity of raw materials, semi-finished products, finished products, and their packaging of precious metals such as gold, silver, platinum, and palladium during the production, weighing, packaging, warehousing, delivery, handover, and sales inquiry processes. Due to the high unit value, numerous circulation stages, heavy handover responsibilities, and the sensitivity of weight and batch information of precious metals, relying solely on ordinary labels, QR codes, or batch numbers for traceability is prone to problems such as label duplication, data addition at key points, discrepancies between weighing records and actual operations, and a lack of verifiable video evidence for handover processes. This results in subsequent queries only seeing static traceability fields, making it difficult to determine the actual operation of the precious metal object at each circulation stage.

[0003] In existing technologies, for example, prior art document CN112085511B discloses a method for generating anti-counterfeiting traceability codes, an anti-counterfeiting traceability method, and a related system. This method primarily involves running an application (APP) to acquire multimedia images related to the product taken by an authenticated user, generating a QR code, a coded message, and detailed information for the corresponding multimedia image. The authenticated multimedia image is then associated with the product's traceability stamp set, which is further linked to the product's batch code. An internal and external code are bound to the batch code, thus forming the anti-counterfeiting traceability code. After scanning the anti-counterfeiting traceability code, consumers can obtain the traceability stamp set associated with the batch code, batch information, and scan query information to understand real-world video footage of the product's production process.

[0004] However, the improvement of the above method focuses on collecting the correlation between multimedia images, QR codes, cryptographic codes and batch codes through the APP, so that the production process of goods can be queried in the form of images or videos. It does not establish a full video evidence chain for continuous nodes such as weighing, label binding, packaging, warehousing and handover in the precious metal supply chain, nor does it perform hierarchical carrying, feature extraction and anti-tampering core processing of the core information, auxiliary information, node information and video index information inside the unique traceability identifier.

[0005] Therefore, when precious metal objects flow through different nodes, there may still be problems such as weak correspondence between video clips and node business data, mixed storage of core traceability information and auxiliary information, difficulty in timely screening of abnormal nodes, and difficulty in identifying traceability identifiers from the information structure level after they are copied. These issues make it difficult to meet the requirements of the high-value precious metal supply chain for full-process video verification, node data consistency verification, and information verification and anti-tampering identification. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a precious metal video evidence preservation and traceability system and method based on information kernel reconstruction and filter learning, which solves the problems mentioned in the background art.

[0007] A method for tracing the provenance of precious metal videos based on information kernel reconstruction and filter learning includes the following steps:

[0008] S1. Establish a unique traceability identifier for precious metal objects that enter the supply chain. The precious metal objects include precious metal raw materials, precious metal semi-finished products, precious metal finished products, or packaging containing precious metal finished products. S2. Construct the unique traceability identifier into an inverted residual three-segment information receiving structure. The inverted residual three-segment information receiving structure includes an import receiving module, an up-dimensional receiving module, and an amplification receiving module connected in sequence. The information carrying area of ​​the import receiving module and the amplification receiving module is greater than the information carrying area of ​​the up-dimensional receiving module, so that the unique traceability identifier forms an information processing structure that is wide at both ends and narrow in the middle. S3. Receive the initial traceability information of the precious metal object through the import receiving module, and classify and lay out the initial traceability information according to information category, information importance level and information area, so that the core information is arranged in the central receiving area of ​​the import receiving module, and the remaining information is arranged in the corner receiving areas of the import receiving module. S4. Receive the radiation information output by the import receiving module through the up-dimensional receiving module, and perform hierarchical feature extraction on the import information based on at least one of the parameters of information level, information area, information export speed and information export time of the radiation information to form source tracing feature information. S5. When performing hierarchical feature extraction in the up-dimensional receiving module, core information is received first, and other information is in a waiting state before entering the up-dimensional receiving module. After the core information is exported by the up-dimensional receiving module, the other information enters the core receiving area of ​​the up-dimensional receiving module after a period of time corresponding to the core information export process, so as to replace the storage location of the previous core information. S6. The amplified receiving module receives the source traceability feature information output by the up-dimensional receiving module, and performs matrix transpose processing on the source traceability feature information within an equal unit time, so that the amplified receiving module receives the information exported by the up-dimensional receiving module from different information angles, forming an important information core for rapid identification. S7. The important information kernel is activated by an activation function. When the important information kernel meets the preset activation conditions, the important information kernel is input into the filter neural network model. S8. The important information kernel is elliptical segmented by the filter neural network model, so that the features of the core information are placed at the center of the elliptical segmentation region, and the remaining information with a lower information level than the core information is compressed into the narrow areas on both sides of the elliptical segmentation region. S9. The filter neural network model prioritizes learning the core information at the center of the elliptical segmentation region and merges the remaining information in the narrow areas on both sides of the elliptical segmentation region to form a filter information kernel with an area smaller than or equal to the area of ​​the original elliptical segmentation region. Then, the center learning and merging process is repeated for the next elliptical segmentation region. S10. Collect operation process videos at the nodes of weighing, sorting, label binding, packaging, warehousing, warehousing and handover of precious metal objects, and associate the filter information core, unique traceability identifier, node business data and operation process videos to form node evidence data. Then, form a full video evidence chain according to the supply chain flow sequence, and provide it to the query terminal for anti-counterfeiting traceability query.

[0009] Preferably, in the inverted residual three-segment information receiving structure, the import receiving module is used to perform wide-domain classification and carrying of the initial traceability information, the up-dimensional receiving module is used to perform narrow-domain feature extraction of the information output by the import receiving module, and the amplification receiving module is used to perform wide-domain amplification and reception of the traceability feature information output by the up-dimensional receiving module. Among them, the information carrying area of ​​the import receiving module and the amplification receiving module is the same or similar, while the information carrying area of ​​the up-dimensional receiving module is smaller than that of the import receiving module and the amplification receiving module, so that the information is transmitted within the unique traceability identifier in the manner of wide-domain import, narrow-domain extraction, and wide-domain amplification.

[0010] Preferably, the import receiving module is configured as a rectangular information receiving structure, the central area of ​​the rectangular information receiving structure is used to lay the core information of the precious metal object, and the four corner areas of the rectangular information receiving structure are used to lay the remaining information of the precious metal object. The core information includes at least one of the following: the unique code of the precious metal object, the precious metal category, the batch number, the weight information, and the current circulation node information; The remaining information includes at least one of the following: operator information, operation time information, packaging information, equipment number information, workstation information, logistics information, and video clip index information.

[0011] Preferably, the radiation information is an information output sequence formed by expanding from the central receiving area to the peripheral receiving areas in the receiving module; The radiation information includes at least the information level, information area, information export path, information export speed, and information export time generated when the core information expands to the remaining information; The upscaling receiving module performs hierarchical feature extraction on the imported information based on different information levels, information areas, information export speeds, or information export times in the radiation information, rather than performing upscaling on the imported information only through a single path.

[0012] Preferably, when core information and other information arrive at the up-dimensional receiving module at the same time, the up-dimensional receiving module only allows the core information to enter the core receiving area, while the other information enters the waiting buffer area; Once the core information has completed feature extraction and been exported by the up-dimensional receiving module, the remaining information in the waiting buffer will have an equal waiting time based on the export time of the core information. After the equal waiting time has elapsed, the remaining information enters the core receiving area of ​​the up-dimensional receiving module and occupies the original storage location of the core information, so that the remaining information can obtain the core receiving weight corresponding to the core information in the subsequent feature extraction process.

[0013] Preferably, when the amplified receiving module receives the source traceability feature information output by the up-dimensional receiving module, it performs matrix transposition on the information matrix corresponding to the source traceability feature information with an equal unit time as the transposition period, so that the row information is converted into column information, or the column information is converted into row information. By receiving data from multiple angles after matrix transposition, the amplified receiving module forms an important information core with an expanded information area; The important information core retains the original classification information of the import receiving module and the feature extraction information of the upgrade receiving module, enabling the important information core to be used for both rapid identification and anti-tampering verification. Preferably, the activation function is used to connect the amplification receiving module and the filter neural network model; When the core information completeness, information level, information area, node business data matching degree, or video segment index completeness in the important information core meet the preset activation conditions, the activation function outputs an entry signal, allowing the important information core to enter the filter neural network model. When the critical information core does not meet the preset activation conditions, the activation function outputs a blocking signal, causing the critical information core to enter an abnormal pending verification state. Preferably, the filter neural network model includes an elliptical segmentation layer, a central learning layer, a flanking contraction layer, and an information kernel merging layer; The elliptical segmentation layer is used to divide the important information kernel into elliptical segmentation regions; The central learning layer is used to prioritize learning the core information features at the center of the elliptical segmentation region. The flank contraction layer is used to spatially contract the remaining information on both sides of the elliptical segmentation region whose information level is lower than that of the core information. The information core merging layer is used to merge the remaining information after the two sides have shrunk to form a filter information core, and to make the area of ​​the filter information core less than or equal to the area of ​​the original elliptical segmentation region. A precious metal video evidence preservation and traceability system based on information kernel reconstruction and filter learning includes: a traceability identifier construction module, used to establish a unique traceability identifier for precious metal objects. The traceability identifier construction module includes an import receiving module, an up-dimensional receiving module, and an amplification receiving module, which together constitute an inverted residual three-segment information receiving structure. The activation connection module is used to determine, through the activation function, whether the important information kernel output by the amplified receiving module meets the conditions for entering the filter neural network model; The filter neural network model is used to perform elliptical segmentation, center information learning, bilateral information shrinkage, and information kernel merging on important information kernels to form filter information kernels. The video capture module is used to capture video of the operation process at the nodes of weighing, sorting, label binding, packaging, warehousing, warehousing and handover of precious metal objects; The node data acquisition module is used to collect node business data generated by barcode scanning devices, electronic weighing devices, label output devices, or back-end business systems. The video evidence association module is used to associate the unique traceability identifier, filter information core, node business data and operation process video to form node evidence data; The evidence processing module is used to link the node evidence data of different circulation nodes in a chain according to the circulation sequence of the precious metal object in the supply chain, forming a full-process video evidence chain. The anti-counterfeiting and traceability query module is used to output the circulation node information, video evidence data, node business data and filter information verification results of precious metal objects based on the unique traceability identifier. Preferably, the filter neural network model includes an ellipse segmentation unit, a center learning unit, a flanking contraction unit, a merging kernel unit, and a recurrent processing unit; The elliptical segmentation unit is used to divide the important information core into at least one elliptical segmentation region. The central learning unit is used to prioritize the extraction of core information features at the center of the elliptical segmentation region. The flank contraction unit is used to compress and place the remaining information in the narrow areas on both sides of the elliptical segmentation region. The merging and core-forming unit is used to merge the remaining information in the narrow areas on both sides to form a filter information core with an area smaller than or equal to the area of ​​the original elliptical segmented region; The loop processing unit is used to input the next elliptical segmentation region into the elliptical segmentation unit after the current elliptical segmentation region is processed, and to repeatedly perform center learning, flanking contraction and merging into a nucleus process.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention constructs an inverted residual three-segment information receiving structure by using an import receiving module, an upscaling receiving module, and an amplification receiving module. The core information of the unique traceability identifier is placed in the central area, while the remaining information is placed in the corner areas. Important information cores are formed through core priority, waiting replacement, and matrix transposition. This ensures that the unique code, batch, weight, node, and video index of the precious metal object are no longer stored as ordinary fields, but form structured identification data with central and extended features, thereby improving the identification efficiency and anti-tampering verification capability of the traceability identifier.

[0015] 2. This invention sets up an activation function and a filter neural network model after the important information core. First, it performs activation judgment on the integrity of the information core, the integrity of the video index, the matching degree of nodes, and the consistency of the summary. Then, it places the core information in the central area through elliptical segmentation, and compresses the low-level information into the narrow areas on both sides and merges them into the filter information core. This allows the core information to be learned and verified first, while the rest of the information is compressed and retained. Combined with the full-process video evidence chain, the video, data, and information core of the precious metal object can be verified synchronously during the weighing, packaging, warehousing, and handover processes.

[0016] Furthermore, the present invention also has the following technical effects: By defining the input information unit Ii as structured data containing field name, field content, field category, node, acquisition time, data source, and field summary, this invention enables the unique codes, weights, batches, nodes, and video indexes of precious metal objects to no longer be stored as loose fields, but rather as computable, partitionable, and traceable information units that enter subsequent processing flows. The scoring function Si determines the core information and other information through information level Li, information area Ai, derivation speed Ri, and authenticity correlation Qi, ensuring that core data is prioritized for entry into the central receiving area, while auxiliary data is placed in peripheral receiving areas, thus reducing the possibility of core data being interfered with by low-level auxiliary data from the data entry point.

[0017] By inputting the coordinate positions, path lengths, and partitioning results from the imported template matrix Min into the upscaling receiving module, the module can perform hierarchical feature extraction based on the information source location, information level, export time, and authenticity correlation. This ensures that the Min matrix not only displays the structure but also directly influences subsequent feature generation. The upscaling receiving module further integrates matrix transpose and Min_up to form a crucial information kernel K, enabling both the original partitioning information and the upscaling feature information to participate in anti-counterfeiting identification, thereby improving the ability to identify when traceability identifiers are copied or partially tampered with.

[0018] By setting Qc, Qv, Qn, Qh, and the anomaly factor E, the activated connection module can comprehensively judge the integrity of core fields, video index integrity, node business matching, and summary consistency before the important information core enters the filter neural network model, avoiding misjudgments caused by incomplete or abnormal data entering the subsequent model. The filter neural network model prioritizes learning core information through the central region of the ellipse, compresses and retains auxiliary information through the flanking regions, and forms the filter information core K' through merging and iterative processing, reducing the occupation of redundant information while retaining the auxiliary basis required for video evidence chain verification. Finally, the filter information core K', together with the node business data, video hash, and previous node summary, forms a chain-like evidence structure, ensuring that any modification to the video or business data of any node can be detected by subsequent summary verification, thereby achieving anti-tampering identification and node consistency verification in the video evidence traceability of the precious metal supply chain. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall structure of a precious metal supply chain anti-counterfeiting and traceability system based on full-process video evidence storage, according to a specific embodiment of the present invention. Figure 2 This is a schematic diagram of the overall process of a precious metal supply chain anti-counterfeiting and traceability method based on full-process video evidence storage, according to a specific embodiment of the present invention. Figure 3 This is a schematic diagram of the inverted residual three-segment information receiving structure according to a specific embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the classification and arrangement of core information and other information in the import receiving module according to a specific embodiment of the present invention. Figure 5 This is a schematic diagram of the activation function and filter neural network model processing flow according to a specific embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the generation and verification results of a full-process video evidence storage chain according to a specific embodiment of the present invention. Detailed Implementation

[0020] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0021] like Figures 1 to 6 Example: This embodiment provides a precious metal video evidence preservation and traceability system and method based on information kernel reconstruction and filter learning. It is applicable to anti-counterfeiting traceability of precious metal raw materials, semi-finished products, finished products, and precious metal packaging during the weighing, sorting, label binding, packaging, warehousing, outbound, handover, and querying processes for gold, platinum, silver, palladium, and other precious metals. Through unique traceability identifiers, full-process video acquisition, node data binding, information kernel construction, filter neural network processing, and chain-like evidence preservation, this system enables every key operation of precious metal objects in the supply chain to be recorded by video, bound to data, and subsequently verified.

[0022] In this embodiment, the system includes a traceability identifier construction module, an activation connection module, a filter neural network model, a video acquisition module, a node data acquisition module, a video evidence storage and association module, an evidence storage and processing module, and an anti-counterfeiting traceability query module. The traceability identifier construction module is used to establish a unique traceability identifier. The activation connection module is used to determine whether important information kernels enter the filter neural network model. The filter neural network model is used to perform elliptic segmentation and compression of the information kernels. The video acquisition module is used to acquire operation process videos of each flow node. The node data acquisition module is used to acquire weighing data, barcode scanning data, label data, and workstation data. The video evidence storage and association module is used to bind videos to data. The evidence storage and processing module is used to generate a complete video evidence storage chain. The anti-counterfeiting traceability query module is used to display the traceability results to the management end or user end.

[0023] In practical implementation, the traceability information generated by precious metal objects at each circulation node is broken down into several input information units, and the unique traceability identifier can be represented as:

[0024] in, Indicates the first One input information unit, This indicates the total number of input information units.

[0025] The iiith input information unit can be represented as:

[0026] in, Indicates the field name. Indicates the field content. Indicates the field category, Indicates the node to which it belongs. Indicates the collection time. Indicates the data source. Represents a field summary.

[0027] To determine the importance of different input information units, this embodiment sets an importance scoring function for each input information unit:

[0028] in, Indicates the first The importance score of each input information unit Indicates information level, Indicates the area of ​​information. Indicates the information export speed. This indicates the degree of correlation in the authenticity verification. , , , This represents the weighting coefficient.

[0029] The weighting coefficients satisfy:

[0030] In this embodiment, it can be set as follows:

[0031] Information area The calculation formula is determined based on the storage length or matrix coverage area occupied by the input information unit:

[0032] in, Indicates the first The character byte length, storage address length, or number of cells occupied by the encoding matrix for each input information unit. This indicates the preset maximum field length or maximum number of cells to be laid.

[0033] Information export speed The amount of information exported by the input information unit per unit time is determined by the following formula:

[0034] in, Indicates the first The time when each input information unit enters the import receiving module Indicates the first The time of each input information unit is exported by the import receiving module. To prevent extremely small constants with a denominator of zero.

[0035] To ensure that information export speed is within a uniform dimension, it is possible to... Normalization is performed:

[0036] in, This indicates the speed at which the normalized information is extracted. Indicates the minimum export speed. This indicates the maximum export speed.

[0037] Authenticity verification correlation The calculation formula is determined based on the correlation between the input information unit and the authenticity verification of the precious metal object:

[0038] in, Indicates the first Does each input information unit have a binding relationship with a unique code? This indicates whether it is related to weight or batch verification. This indicates whether it is related to the order of the flow nodes. This indicates whether it is related to a video segment index or a video hash.

[0039] When the When the importance score of an input information unit is greater than or equal to a preset core threshold, the input information unit is classified into the core information set:

[0040] When the When the importance score of an input information unit is less than a preset core threshold, the input information unit is classified into the remaining information set:

[0041] in, Represents a set of core information. Represents the remaining information set. This indicates the preset core threshold.

[0042] The import receiving module is configured as a rectangular information receiving structure, with the central area used to lay the core information set and the four corner areas used to lay the remaining information sets. The import template matrix can be represented as:

[0043] in, This indicates the import of a template matrix. The comprehensive encoded value representing the core information set. , , , These represent four auxiliary information partitions.

[0044] After the receiving module completes the information deployment, it generates radiating information extending from the central receiving area to the peripheral receiving areas. The radiation information of each input information unit can be represented as:

[0045] in, Indicates information level, Display information area, Indicates the information export speed. Display information export time, Indicates the information export path.

[0046] Information export time can be expressed as:

[0047] in, Indicates the start time of the current node.

[0048] If the first The position of each input information unit in the import template matrix is: The location of the central receiving area is Then the path length It can be represented as:

[0049] in, Indicates the first The path length from the partition to the central receiving area for each input information unit.

[0050] The up-dimensional receiving module sets an input gating function to control the order in which core information and other information enter:

[0051] in, 1 indicates that the first Each input information unit enters the core receiving area. 0 indicates the first Each input information unit enters the waiting buffer area. Indicates the time when core information was exported. This indicates the waiting time for the remaining information.

[0052] The up-dimensional receiving module takes the content, location, level, area, velocity, time, path, and realism correlation of the input information unit as input to form an up-dimensional input vector:

[0053] in, Indicates the first The upgraded input vector of each input information unit Indicates the first The numerical vector obtained by converting the content of each input information unit.

[0054] The output of the up-dimensional receiving module is the first One source-tracing feature:

[0055] in, Indicates the first One traceability feature, Represents a nonlinear mapping function. Represents the feature mapping weight matrix. Indicates the bias term. Indicates the first The comprehensive weight of each input information unit.

[0056] Overall weight It can be represented as:

[0057] in, This represents the weighting parameter.

[0058] The weight parameters satisfy:

[0059] Multiple source-tracing features are arranged according to their location and node order to form an upgraded feature matrix:

[0060] The amplified receiving module performs matrix transpose on the upgraded feature matrix:

[0061] in, Indicates the first The magnified information matrix formed per unit time, Indicates the first The transpose of the upgraded feature matrix for each unit time.

[0062] The amplified receiving module fuses the transposed, up-dimensional feature matrix with the size-adapted imported template matrix to form a core of important information:

[0063] in, Indicates important information core, Represents the normalization function. Indicates the first The weight received by the second transpose This represents the imported template matrix after size adaptation. This indicates that the imported template matrix retains the weights.

[0064] The activation module uses a gated activation function to determine whether important information kernels should enter the filter neural network model.

[0065] in, This indicates that the output value is activated.

[0066] The comprehensive judgment value in the activation function It can be represented as:

[0067] in, Indicates the completeness of core information. Indicates the completeness of the video segment index. Indicates the degree of matching between node business data. Indicates the consistency of the evidence digest. Indicates an abnormal factor.

[0068] When the activation output value is greater than or equal to the preset activation threshold, the important information kernel enters the filter neural network model:

[0069] When the activated output value is less than the preset activation threshold, the critical information core enters the abnormal verification process:

[0070] in, This represents a filter neural network model. This indicates an abnormal process that needs to be verified. This indicates the preset activation threshold.

[0071] Completeness of core information Calculated based on the completeness of the core fields:

[0072] in, Indicates the total number of preset core fields. This indicates the number of core fields that have been collected, have non-empty values, are in a valid format, and have passed the field summary validation.

[0073] Video clip index completeness Calculated based on the completeness of the video index:

[0074] in, This indicates the total number of video nodes that the current precious metal object should be bound to. This indicates that the video file, video segment index, and video hash already exist, and the number of consistent video nodes can be recalculated using the video hash.

[0075] Node business data matching degree It can be represented as:

[0076] in, This indicates the result of node order matching. This indicates the weight data matching results. This indicates the matching results of equipment and workstation information.

[0077] Consistency of evidence digest Determined based on the results of recalculation according to node summaries:

[0078] in, This represents the total number of node summaries that have been generated for the current precious metal object. This indicates the number of node summaries that match the summaries saved in the backend after recalculation.

[0079] Abnormal factors It can be represented as:

[0080] in, This indicates an anomaly such as a missing video or an inconsistent video hash. This indicates an abnormal node order. This indicates an anomaly in the digest chain. This indicates that the weight data is abnormal.

[0081] The filter neural network model divides the important information kernel into at least one elliptical segmentation region, the first... Each elliptical segmented region can be represented as:

[0082] in, Indicates the first Elliptical division regions Indicates the center of the ellipse. Represents the radius of the major axis of the ellipse. This represents the radius of the minor axis of the ellipse.

[0083] The central region of the ellipse can be represented as:

[0084] in, Indicates the first The central region of each elliptical segmentation region Represents the scaling factor for the central region, and satisfies:

[0085] The narrow regions on both sides of the ellipse can be represented as:

[0086] in, Indicates the first The two narrow areas are located within an elliptical segmented region.

[0087] The central learning layer prioritizes learning the core information features in the central region of the ellipse:

[0088] Among them, F Indicates the first The core information features of the central region of each ellipse Represents the central learning function, This indicates extraction by region mask. This indicates the mask in the central area.

[0089] The flank contraction layer extracts features from the remaining information within the narrow regions on both sides of the ellipse:

[0090] in, Indicates the first The remaining information features within the narrow areas on both sides of the ellipse This represents the flank learning function.

[0091] Spatial contraction is applied to the remaining information features on both sides:

[0092] in, This represents the remaining information features after shrinkage. Let represent the spatial contraction coefficient, and satisfy:

[0093] The information core merging layer merges the central core information features and the remaining information features after shrinkage on both sides to form the first core information core merging layer. Individual filter information core:

[0094] in, Indicates the first Each filter information core, This represents the feature merging function.

[0095] The core area of ​​the filter information satisfies:

[0096] The iterative processing layer sequentially performs center learning, flanking contraction, and merging into kernels on multiple elliptical segmented regions, ultimately obtaining a set of filter information kernels:

[0097] in, This represents the final set of filter information cores. This indicates the number of regions divided by the ellipse.

[0098] After the filter information core is generated, the unique traceability identifier, filter information core, node business data, and video clips are associated to form node evidence storage data:

[0099] in, Indicates the first Each node stores evidence data. Indicates a unique traceability identifier. Indicates filter information core, Indicates the first Individual node business data, Indicates the first Each node video clip.

[0100] No. The business data of each node can be represented as follows:

[0101] in, Indicates the node name. Indicates weight data, Indicates the operator. Indicates the equipment number. Indicates the packaging number. Indicates the node time.

[0102] To prevent subsequent tampering with node data or video data, the evidence processing module performs [further steps]. Summarize the data from each node. The notarized digest of a node can be represented as:

[0103] in, Indicates the first The evidence digest of each node, Represents the hash digest function. Indicates the first Each node time, This represents the evidence digest of the previous node. This indicates data concatenation.

[0104] The evidence summaries from multiple nodes are linked in a chain according to the supply chain flow sequence, forming a complete video evidence chain:

[0105] in, This indicates a complete video evidence chain. This indicates the number of evidence storage nodes.

[0106] During anti-counterfeiting and traceability queries, the system recalculates the node digest and compares it with the node digest stored in the background. The verification result can be expressed as:

[0107] in, 1 indicates that the verification passed. 0 indicates a verification error. This represents the node digest that is recalculated during the query. This indicates the filter information core saved in the background. This indicates the node business data stored in the background.

[0108] After the filter information is generated, the system enters the full video evidence storage process. Video capture terminals are set up at the weighing, sorting, label binding, packaging, warehousing, outbound, and handover points of the precious metal objects. The video capture terminals cover at least two of the following: the operating table, the precious metal objects, the containers, the electronic weighing equipment, the barcode scanning equipment, the label output device, and the packaging area, so that the physical status and business data of the precious metal objects at each node can be recorded synchronously.

[0109] The design concept of this embodiment is as follows: First, the unique traceability identifier is transformed from ordinary coding into an important information core with central information, corner information, narrow-domain extraction, and wide-domain amplification through an inverted residual three-segment information receiving structure; then, the information core that can enter the filter neural network model is selected through an activation function; subsequently, the core information is placed in the central region through an elliptical segmentation structure, and the remaining information is compressed to the narrow regions on both sides, and the central information is learned first and the information on both sides is merged in the filter neural network model to form a filter information core with an area smaller than or equal to the original elliptical region; finally, the filter information core is bound to the whole process video, weighing data, tag data, and node summary chain, so that the physical operation process, business data, and anti-counterfeiting information core in the precious metal supply chain correspond to each other.

[0110] Through the above implementation methods, the present invention enables precious metal objects to not only have a unique and searchable number during the supply chain process, but also an information core structure composed of core information, auxiliary information, video index, and node data. At the same time, the video segment of each node is bound to the information core and node data, and subsequent queries can simultaneously verify the consistency of the information core, the integrity of the video, and the integrity of the chain summary, thereby improving the reliability of anti-counterfeiting and traceability in the precious metal supply chain.

[0111] All embodiments of the present invention are within the scope of protection of this patent.

[0112] The embodiments of the present invention are given for the purposes of illustration and description. Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A precious metal video storage traceability method based on information kernel reconstruction and filter net learning, characterized in that, Includes the following steps: S1. Establish a unique traceability identifier for precious metal objects that enter the supply chain. The precious metal objects include precious metal raw materials, precious metal semi-finished products, precious metal finished products, or packaging containing precious metal finished products. S2. Construct the unique traceability identifier into an inverted residual three-segment information receiving structure. The inverted residual three-segment information receiving structure includes an import receiving module, an up-dimensional receiving module, and an amplification receiving module connected in sequence. The information carrying area of ​​the import receiving module and the amplification receiving module is greater than the information carrying area of ​​the up-dimensional receiving module, so that the unique traceability identifier forms an information processing structure that is wide at both ends and narrow in the middle. S3. Receive the initial traceability information of the precious metal object through the import receiving module, and classify and lay out the initial traceability information according to information category, information importance level and information area, so that the core information is arranged in the central receiving area of ​​the import receiving module, and the remaining information is arranged in the corner receiving areas of the import receiving module. S4. Receive the radiation information output by the import receiving module through the up-dimensional receiving module, and perform hierarchical feature extraction on the import information based on at least one of the parameters of information level, information area, information export speed and information export time of the radiation information to form source tracing feature information. S5. When performing hierarchical feature extraction in the up-dimensional receiving module, core information is received first, and other information is in a waiting state before entering the up-dimensional receiving module. After the core information is exported by the up-dimensional receiving module, the other information enters the core receiving area of ​​the up-dimensional receiving module after a period of time corresponding to the core information export process, so as to replace the storage location of the previous core information. S6. The amplified receiving module receives the source traceability feature information output by the up-dimensional receiving module, and performs matrix transpose processing on the source traceability feature information within an equal unit time, so that the amplified receiving module receives the information exported by the up-dimensional receiving module from different information angles, forming an important information core for rapid identification. S7. The important information kernel is activated by an activation function. When the important information kernel meets the preset activation conditions, the important information kernel is input into the filter neural network model. S8. The important information kernel is elliptical segmented by the filter neural network model, so that the features of the core information are placed at the center of the elliptical segmentation region, and the remaining information with a lower information level than the core information is compressed into the narrow areas on both sides of the elliptical segmentation region. S9. The filter neural network model prioritizes learning the core information at the center of the elliptical segmentation region and merges the remaining information in the narrow areas on both sides of the elliptical segmentation region to form a filter information kernel with an area smaller than or equal to the area of ​​the original elliptical segmentation region. Then, the center learning and merging process is repeated for the next elliptical segmentation region. S10. Collect operation process videos at the nodes of weighing, sorting, label binding, packaging, warehousing, warehousing and handover of precious metal objects, and associate the filter information core, unique traceability identifier, node business data and operation process videos to form node evidence data. Then, form a full video evidence chain according to the supply chain flow sequence, and provide it to the query terminal for anti-counterfeiting traceability query.

2. The method for tracing and storing evidence of precious metal videos based on information kernel reconstruction and filter learning according to claim 1, characterized in that: In the inverted residual three-segment information receiving structure, the import receiving module is used to perform wide-domain classification and carrying of the initial source information, the up-dimensional receiving module is used to extract narrow-domain features from the information output by the import receiving module, and the amplification receiving module is used to perform wide-domain amplification and reception of the source feature information output by the up-dimensional receiving module. Among them, the information carrying area of ​​the import receiving module and the amplification receiving module is the same or similar, while the information carrying area of ​​the up-dimensional receiving module is smaller than that of the import receiving module and the amplification receiving module, so that the information is transmitted within the unique traceability identifier in the manner of wide-domain import, narrow-domain extraction, and wide-domain amplification.

3. The method for tracing and storing evidence of precious metal videos based on information kernel reconstruction and filter learning according to claim 1, characterized in that: The import receiving module is configured as a rectangular information receiving structure. The central area of ​​the rectangular information receiving structure is used to lay the core information of the precious metal object, and the four corner areas of the rectangular information receiving structure are used to lay the remaining information of the precious metal object. The core information includes at least one of the following: the unique code of the precious metal object, the precious metal category, the batch number, the weight information, and the current circulation node information; The remaining information includes at least one of the following: operator information, operation time information, packaging information, equipment number information, workstation information, logistics information, and video clip index information.

4. The method for tracing and storing evidence of precious metal videos based on information kernel reconstruction and filter learning according to claim 1, characterized in that: The radiation information is an information output sequence that is imported into the receiving module and extends from the central receiving area to the corner receiving areas; The radiation information includes at least the information level, information area, information export path, information export speed, and information export time generated when the core information expands to the remaining information; The upscaling receiving module performs hierarchical feature extraction on the imported information based on different information levels, information areas, information export speeds, or information export times in the radiation information, rather than performing upscaling on the imported information only through a single path.

5. The method for tracing and storing evidence of precious metal videos based on information kernel reconstruction and filter learning according to claim 1, characterized in that: When core information and other information arrive at the up-dimensional receiving module at the same time, the up-dimensional receiving module only allows the core information to enter the core receiving area, while the other information enters the waiting buffer area. Once the core information has completed feature extraction and been exported by the up-dimensional receiving module, the remaining information in the waiting buffer will have an equal waiting time based on the export time of the core information. After the equal waiting time has elapsed, the remaining information enters the core receiving area of ​​the up-dimensional receiving module and occupies the original storage location of the core information, so that the remaining information can obtain the core receiving weight corresponding to the core information in the subsequent feature extraction process.

6. The method for tracing and storing evidence of precious metal videos based on information kernel reconstruction and filter learning according to claim 1, characterized in that: When the amplified receiving module receives the source traceability feature information output by the up-dimensional receiving module, it performs matrix transposition on the information matrix corresponding to the source traceability feature information with an equal unit time as the transposition period, so that the row information is converted into column information, or the column information is converted into row information. By receiving data from multiple angles after matrix transposition, the amplified receiving module forms an important information core with an expanded information area; The important information core retains the original classification information of the import receiving module and the feature extraction information of the upgrade receiving module, enabling the important information core to be used for both rapid identification and anti-tampering verification.

7. The method for tracing and storing evidence of precious metal videos based on information kernel reconstruction and filter learning according to claim 1, characterized in that: The activation function is used to connect the amplified receiving module and the filter neural network model; When the core information completeness, information level, information area, node business data matching degree, or video segment index completeness in the important information core meet the preset activation conditions, the activation function outputs an entry signal, allowing the important information core to enter the filter neural network model. When the critical information core does not meet the preset activation conditions, the activation function outputs a blocking signal, causing the critical information core to enter an abnormal pending verification state.

8. The method for tracing and storing evidence of precious metal videos based on information kernel reconstruction and filter learning according to claim 1, characterized in that: The filter neural network model includes an elliptical segmentation layer, a central learning layer, a flanking contraction layer, and an information kernel merging layer. The elliptical segmentation layer is used to divide the important information kernel into elliptical segmentation regions; The central learning layer is used to prioritize learning the core information features at the center of the elliptical segmentation region. The flank contraction layer is used to spatially contract the remaining information on both sides of the elliptical segmentation region whose information level is lower than that of the core information. The information core merging layer is used to merge the remaining information after the two sides have shrunk to form a filter information core, and the area of ​​the filter information core is less than or equal to the area of ​​the original elliptical segmented region.

9. A precious metal video evidence preservation and traceability system based on information kernel reconstruction and filter learning, characterized in that, include: The traceability identifier construction module is used to establish a unique traceability identifier for precious metal objects. The traceability identifier construction module includes an import receiving module, an up-dimensional receiving module, and an amplification receiving module, which together form an inverted residual three-segment information receiving structure. The activation connection module is used to determine, through the activation function, whether the important information kernel output by the amplified receiving module meets the conditions for entering the filter neural network model; The filter neural network model is used to perform elliptical segmentation, center information learning, bilateral information shrinkage, and information kernel merging on important information kernels to form filter information kernels. The video capture module is used to capture video of the operation process at the nodes of weighing, sorting, label binding, packaging, warehousing, warehousing and handover of precious metal objects; The node data acquisition module is used to collect node business data generated by barcode scanning devices, electronic weighing devices, label output devices, or back-end business systems. The video evidence association module is used to associate the unique traceability identifier, filter information core, node business data and operation process video to form node evidence data; The evidence processing module is used to link the node evidence data of different circulation nodes in a chain according to the circulation sequence of the precious metal object in the supply chain, forming a full-process video evidence chain. The anti-counterfeiting and traceability query module is used to output the circulation node information, video evidence data, node business data and filter information verification results of precious metal objects based on the unique traceability identifier.

10. A precious metal video evidence preservation and traceability system based on information kernel reconstruction and filter learning according to claim 9, characterized in that: The filter neural network model includes an ellipse segmentation unit, a center learning unit, a flanking contraction unit, a merging kernel unit, and a loop processing unit. The elliptical segmentation unit is used to divide the important information core into at least one elliptical segmentation region. The central learning unit is used to prioritize the extraction of core information features at the center of the elliptical segmentation region. The flank contraction unit is used to compress and place the remaining information in the narrow areas on both sides of the elliptical segmentation region. The merging and core-forming unit is used to merge the remaining information in the narrow areas on both sides to form a filter information core with an area smaller than or equal to the area of ​​the original elliptical segmented region; The loop processing unit is used to input the next elliptical segmentation region into the elliptical segmentation unit after the current elliptical segmentation region is processed, and to repeatedly perform center learning, flanking contraction and merging into a nucleus process.

Images