Customs clearance management method and system for port gate and computer program product

Abstract

The embodiment of the specification discloses a customs clearance management method and system for a port gate and a computer program product. The method comprises the following steps: pre-connecting multiple business systems, performing field mapping and structured fusion based on multi-dimensional feature data maintained by each business system, and generating a corresponding preset passing feature set for each reserved vehicle; when a target vehicle enters a preset long-distance detection area, performing first information collection and matching the target preset passing feature set, performing preliminary verification and determining a customs clearance mode; in the case that the customs clearance mode is a non-sensing passing mode, when the target vehicle enters a gate passage, at least two different sensing modes are used to collect real-time features in parallel for consistency verification; and a passing control instruction is sent to a passage control device based on the verification result.

Description

Technical Field

[0001] This specification relates to the field of intelligent transportation and security control technology in ports, specifically to a customs clearance control method, system, and computer program product for port gates. Background Technology

[0002] As a logistics hub and regulatory area, a port's gates are the only physical passage for vehicles to enter and exit, simultaneously fulfilling the dual functions of traffic efficiency and safety control. The basic principle of port gate traffic control is to acquire vehicle identity information through sensing devices, compare it with the registered data in the business system, and release the vehicle after successful verification by the gate control equipment.

[0003] Existing port access control schemes generally employ a post-arrival serial processing architecture. When a vehicle arrives at the gate, the driver swipes a card or scans a code to trigger the system. The system then connects to independently operating business systems such as the vehicle management office, port reservation system, and freight management system, initiating data query requests one by one to obtain registration information for the vehicle, personnel, and cargo. Because these business systems were built by different vendors at different times, using different communication protocols and data standards, they form a physically isolated "siloed" architecture, resulting in significant network transmission delays for cross-system queries. After obtaining the registration information, manual verification of documents and physical goods is performed item by item before the gate is finally opened for passage. Each verification step is executed strictly sequentially, with the next step only proceeding after the previous one is completed.

[0004] However, in non-stop passage scenarios, the physical time window for a vehicle to pass through a toll gate is extremely limited. Typically, a vehicle at normal speed takes only a few seconds to pass through a 10-50 meter toll gate. Under the aforementioned serial architecture, if data from multiple heterogeneous systems needs to be acquired, the time spent on serial network requests can easily exhaust this physical time window. Further multi-dimensional identity verification takes far longer than the vehicle's travel time. Therefore, even with faster single-point identification devices, the serial dependency between data acquisition and verification processes cannot be broken, forcing vehicles to stop and wait for manual confirmation or a system response, making truly seamless passage difficult to achieve.

[0005] How to complete the acquisition of multi-source data and verification of multi-dimensional features within a limited physical passage time window, and achieve non-stop passage while ensuring safety, is a technical problem that needs further research in the field of port passage management. Summary of the Invention

[0006] The purpose of the embodiments in this specification is to provide a method, system and computer program product for customs clearance control at port gates, so as to solve the technical problem that the existing serial verification architecture cannot complete multi-dimensional identity consistency verification under the physical time constraint of non-stop passage.

[0007] According to a first aspect of this specification, a customs clearance control method for port gates is provided, comprising: Multiple business systems are pre-connected, and field mapping and structured fusion are performed based on the multi-dimensional feature data maintained by each business system to generate a corresponding pre-set passage feature set for each reserved vehicle. This pre-set passage feature set is a collection centered on a single clearance event, linking vehicle feature data, personnel feature data, cargo feature data, and credit assessment data from different business systems into a unified data structure, serving as the comparison benchmark for subsequent verification.

[0008] When a target vehicle enters the preset long-range detection area, the system performs an initial information collection on the vehicle and matches it with a corresponding preset target passage feature set based on the collected preliminary identity information. The preset long-range detection area is a predetermined distance range upstream of the gate channel, designed to provide the system with sufficient time to complete data matching and mode decision-making before the vehicle reaches the gate channel. The preliminary identity information is a unique identifier for the target vehicle's passage event, such as a license plate number or vehicle-mounted electronic tag number. The target vehicle undergoes preliminary verification based on the preset target passage feature set. If the preliminary verification is successful, the passage mode for the target vehicle is determined based on the credit assessment data contained in the preset target passage feature set. The passage mode indicates the appropriate verification depth and passage method for the vehicle.

[0009] When the passage mode is a seamless passage mode, when the target vehicle enters the gate channel, a second information collection is performed on the target vehicle and driver. This second information collection employs at least two different perception modalities to collect real-time features in parallel. The real-time features are then compared with the corresponding dimensions of preset features in the target's preset passage feature set for consistency verification. The at least two different perception modalities refer to perception methods based on different physical principles, such as optical imaging, radio frequency sensing, and biometric data collection. Each modality is executed in parallel over time, ensuring that the increase in verification dimensions does not lead to a linear increase in verification time.

[0010] Based on the result of the consistency verification, a passage control command is sent to the channel control device. The channel control device is a hardware device that performs physical passage control, such as a turnstile, barrier gate, or diversion guidance device.

[0011] According to a second aspect of this specification, a customs clearance control system for port gates is provided, comprising: The pre-aggregation unit is used to pre-connect to multiple business systems, perform field mapping and structured fusion based on the multi-dimensional feature data maintained by each business system, and generate a corresponding pre-set passage feature set for each reserved vehicle.

[0012] The long-range initial identification unit is used to perform the first information collection on the target vehicle when the target vehicle enters the preset long-range detection area, and match the corresponding target preset passage feature set according to the collected preliminary identity information; perform preliminary verification on the target vehicle according to the target preset passage feature set, and determine the passage mode of the target vehicle based on the credit assessment data contained in the target preset passage feature set if the preliminary verification is passed.

[0013] The multimodal travel verification unit is used to collect real-time features in parallel using at least two different perception modalities when the target vehicle enters the gate channel in the case of the non-intrusive passage mode. The real-time features are then used to verify the consistency between the real-time features and the corresponding dimensions of the preset features in the target preset passage feature set.

[0014] The channel control unit is used to send access control commands to the channel control device based on the result of the consistency verification.

[0015] According to a third aspect of this specification, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the steps of the method as described in the first aspect of this specification.

[0016] Compared with related technologies, the embodiments in this specification can achieve the following technical effects: By pre-connecting to multiple business systems and performing field mapping and structured fusion to generate a pre-set passage feature set, the aggregation and alignment of multi-source heterogeneous data is completed before the vehicle arrives at the gate, eliminating the network delay of cross-system serial queries after arrival and gaining a time window for subsequent non-stop verification.

[0017] By performing the first information collection and matching the target's pre-set passage feature set when the target vehicle enters the preset long-distance detection area, and making preliminary verification, the data retrieval and mode decision are completed in advance by utilizing the physical travel time of the vehicle approaching the gate. This realizes the exchange of physical spatial distance for data preparation time, making non-stop decision-making possible.

[0018] By using at least two different sensing modalities to collect real-time features in parallel during the process of the target vehicle entering the gate channel for consistency verification, the positive correlation between dimensional growth and time consumption caused by traditional serial item-by-item verification is broken, enabling multi-dimensional deep verification to be completed in parallel within a very short physical passage window.

[0019] By directly sending access control commands to the channel control equipment based on the consistency verification results, the direct mapping from digital space verification results to physical space access control is realized, eliminating the time-consuming manual confirmation process, and thus enabling non-stop passage while ensuring multi-dimensional identity consistency verification. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0021] Figure 1 This is a schematic diagram illustrating the overall deployment scenario of the port gate clearance control system provided in the embodiments of this specification; Figure 2 This is the main flowchart of the customs clearance control method provided in the embodiments of this specification; Figure 3 This is a schematic diagram of the second information acquisition multimodal parallel verification provided in the embodiments of this specification; Figure 4 This is a schematic diagram illustrating the spatiotemporal alignment and fusion decision-making of multi-source security inspection data provided in the embodiments of this specification; Figure 5 This is a schematic diagram of the customs clearance control system provided in the embodiments of this specification; Figure 6 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this specification. Detailed Implementation

[0022] The embodiments of this specification are described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the technical solutions of this specification and do not constitute a limitation on the scope of protection of this specification.

[0023] The customs clearance control method provided in the embodiments of this specification can be deployed on a port gate customs clearance control system, and is executed collaboratively by edge computing nodes deployed at the gate site and a remote customs clearance control server. This method is applicable to scenarios such as port container transport gates and bulk cargo transport gates that require identity verification and security control of vehicles entering and leaving the port.

[0024] It should be noted that the collection and processing of driver facial features, license plate information, and other data involved in the embodiments of this specification are all carried out with the authorization of the relevant parties or in accordance with the provisions of laws and regulations, and are only used for the purpose of port gate passage control.

[0025] The main terms used in this specification are defined as follows: The pre-set passage feature set refers to a unified data structure centered on a single passage event, generated by aggregating data from multiple business systems based on passage event identifiers and after field mapping and structured fusion before a vehicle arrives at the gate. This data serves as the comparison benchmark for subsequent verification.

[0026] The target pre-set access feature set refers to the pre-set access feature set that corresponds to the target vehicle when the target vehicle actually arrives, which is determined by matching the preliminary identity information obtained during the first information collection with the pre-set access feature set that has been generated.

[0027] A clearance event identifier is an index key that uniquely identifies a specific clearance event, such as a combination of reservation number, license plate number and reservation time window, or vehicle electronic tag number.

[0028] Consistency verification refers to the process of comparing the features of the target object collected in real time through the sensing modality with the corresponding preset features in the preset access feature set to determine whether the two meet the preset consistency conditions.

[0029] Perception modality refers to information acquisition channels based on different physical principles, such as image acquisition based on optical imaging, face acquisition based on near-infrared or visible light, electronic tag reading based on radio frequency sensing, and seal recognition based on image recognition or radio frequency reading.

[0030] Traffic risk assessment value refers to a quantitative indicator that characterizes the overall risk level of a target vehicle, calculated based on credit assessment data from a pre-set traffic feature set.

[0031] The core perception modality refers to the perception modality that plays a key role in identity verification during consistency checks, and whose verification result must pass before it can be released, such as license plate comparison modality and face comparison modality.

[0032] Auxiliary perception modalities refer to perception modalities that perform supplementary verification functions in consistency checks, and whose verification results are allowed to have slight deviations, such as vehicle body color comparison modalities and seal comparison modalities. Those skilled in the art can classify core modalities and auxiliary modalities according to actual safety level requirements.

[0033] Physical inspection methods refer to technical means of conducting safety inspections on target vehicles and their cargo based on different physical principles. Examples include internal structure inspection based on X-ray or gamma-ray penetration imaging, density distribution inspection based on millimeter-wave reflection, gas composition inspection based on chemical sensing, radioactive material inspection based on radiation detection, chassis attachment inspection based on optical scanning, and temperature anomaly inspection based on infrared thermal imaging.

[0034] Cross-dimensional fusion refers to the process of jointly evaluating the judgment results from different detection dimensions (identity verification dimension, physical security inspection dimension, and historical credit dimension) and outputting a unified comprehensive risk assessment value.

[0035] Deviation refers to a quantitative measure of the degree of difference between real-time collected features and the corresponding dimensions of preset features in a preset common feature set, such as the Euclidean distance or cosine distance between feature vectors.

[0036] A clearance event refers to the complete passage process corresponding to a specific vehicle passage reservation, with a clearance event identifier (such as reservation number) as a unique index.

[0037] Seamless passage mode refers to a passage mode in which the target vehicle does not need to stop, the driver does not need to get out of the vehicle, and does not need to show any documents to complete all verification and be automatically released.

[0038] The rapid review mode refers to the common mode that only performs consistency checks on the core perception modality and is supplemented by manual spot checks.

[0039] The detailed inspection mode refers to the traffic mode that guides the target vehicle to the manual verification lane for a comprehensive inspection.

[0040] like Figure 1 As shown in the embodiments of this specification, the port gate clearance control system includes a preset remote detection area 101, a gate channel 102, an edge computing node 103, a clearance control server 104, and a channel control device 105.

[0041] The preset long-range detection area 101 is located upstream of the gate channel 102, with a physical distance of 200 to 500 meters from the gate channel 102. A telephoto camera 111 and an electronic tag long-range sensing device 112 are used to detect target vehicles entering the preset long-range detection area 101. The telephoto camera 111 can be deployed near the gate or at an appropriate location upstream of the preset long-range detection area 101, and its resolution and focal length must meet the requirement of clearly identifying the license plate characters of the target vehicle at a distance of 200 to 500 meters. The electronic tag long-range sensing device 112 is used to read the vehicle's onboard electronic tag (such as an ETC tag or RFID tag) when the target vehicle enters the preset long-range detection area 101, obtaining a clearance event identifier. The purpose of setting up the preset long-range detection area 101 is to allow the system sufficient time to complete data matching and clearance mode decision-making before the target vehicle reaches the gate channel 102. Taking a vehicle speed of 30km / h as an example, the travel time from the preset remote detection area 101 to the gate channel 102 is 24 to 60 seconds. This travel time is the available processing time for the system to perform the first information collection, feature set matching, preliminary verification and pattern decision.

[0042] Gate channel 102 is the physical passage through which vehicles actually pass, typically ranging from 10 to 50 meters in length. A multimodal sensing device group is used to detect target vehicles and drivers entering gate channel 102, including a close-range image acquisition device 121, a face acquisition device 122, an RFID reader 123, and a seal recognition device 124. These devices can be deployed at appropriate locations inside, on either side of, or near the entrance of gate channel 102, as long as their effective detection range covers gate channel 102. The close-range image acquisition device 121 is used to acquire license plate information, vehicle model outline, and vehicle body color features under close-range conditions. The face acquisition device 122 is used to acquire the driver's facial image features as the vehicle passes. The RFID reader 123 is used to read vehicle-mounted electronic tag information at close range. The seal recognition device 124 is used to identify cargo seal markings and container numbers. All of the above sensing devices operate simultaneously during vehicle movement, with their acquisition windows overlapping on the timeline.

[0043] In addition, the security inspection equipment group 130 is used to conduct security inspections on target vehicles and their cargo entering the gate channel 102, including X-ray scanning devices, millimeter-wave radar, gas detection sensors, etc. The security inspection equipment group 130 can be deployed inside the gate channel 102, adjacent to the channel exit, or at an appropriate location along the channel. Its deployment location is spatially adjacent to or overlaps with the multimodal sensing equipment group, so that identity verification and physical security inspection can be completed simultaneously during a single vehicle passage.

[0044] Edge computing node 103 is deployed at the gate site and connects to various sensing devices within the preset long-range detection area 101 and gate passage 102 via wired or wireless means. Edge computing node 103 undertakes computational tasks with high real-time requirements, including feature set matching, preliminary verification, and pattern decision-making after the first information collection, as well as multimodal parallel consistency verification after the second information collection. Customs clearance control server 104 is deployed in a data center or cloud and undertakes computational tasks with relatively low real-time requirements, such as pre-generation of preset passage feature sets, model training, and updates. Edge computing node 103 and customs clearance control server 104 are connected via a dedicated network.

[0045] The access control device 105 is a hardware device that performs physical access control, such as a turnstile (raise / lower gate), a barrier gate, a diversion and guidance device, or an interception device. The access control device 105 receives access control commands sent by the edge computing node 103 and executes the corresponding physical actions.

[0046] Based on the above deployment scenario, the process of the customs clearance control method provided in the embodiments of this specification is as follows: Figure 2 As shown.

[0047] like Figure 2As shown in the embodiments of this specification, the customs clearance control method includes the following steps.

[0048] Step S201: Connect to multiple business systems in advance, perform field mapping and structured fusion based on the multi-dimensional feature data maintained by each business system, and generate a corresponding preset passage feature set for each reserved vehicle.

[0049] The main entity executing this step is the customs clearance control server 104. Before the vehicle actually arrives at the gate, the customs clearance control server 104 has already completed the aggregation and fusion of multi-source heterogeneous data. This pre-processing timing arrangement ensures that the edge computing node 103 does not need to initiate cross-system data queries within the limited time window of the vehicle's travel in subsequent steps, thereby separating the data acquisition delay from the passage time.

[0050] The generation timing of the pre-set passage feature set is linked to the reservation system. When a new vehicle reservation record is generated in the port reservation system, the customs control server 104 initiates the data aggregation process for that reserved vehicle. In a specific implementation scenario, the customs control server 104 initiates data aggregation 2 hours before the vehicle's scheduled arrival time to ensure that the feature set has been generated and distributed to the edge computing node 103 before the vehicle actually arrives.

[0051] Specifically, the customs clearance control server 104 obtains characteristic data associated with the reserved vehicle from the following five types of business systems: vehicle registration information from the vehicle management office database, including license plate number, registered vehicle model, registered color, vehicle identification number, etc.; reservation information from the port reservation system, including reservation number, reservation arrival time window, carrier information, etc.; cargo declaration information from the freight management system, including container number, seal number, declared cargo type, weight, etc.; driver registration information from the personnel database, including driver name, ID number, facial feature vector, professional qualification certificate number, etc.; and credit assessment information from the credit management system, including vehicle credit score, driver credit score, carrier credit score, historical violation records, etc.

[0052] Because the five types of business systems mentioned above were built by different suppliers at different times, there are differences in the naming, encoding format, and semantic definition of data fields in each system. For example, the unique vehicle identifier stored in the vehicle management office database as the "Vehicle Identification Number" field corresponds to the "License Plate Number" field in the port reservation system, and the "Transport Vehicle Number" field in the freight management system—all three point to the same physical entity but have different names and formats. To eliminate such semantic conflicts, the customs clearance control server 104 performs field mapping and format unification processing on the raw data returned by each system: using the reservation number as the customs clearance event identifier, it associates records in each system that point to the same customs clearance event; it establishes a mapping table for differences in field naming, unifying synonymous fields in different systems into standard field names; and it performs format conversion for differences in encoding formats (such as date format, ID card number encryption method, etc.).

[0053] After the above processing, the customs clearance control server 104 constructs a unified data structure centered on the customs clearance event for each reserved vehicle, namely, a pre-set passage feature set. This feature set includes at least the following dimensions of pre-set features: vehicle feature benchmarks (pre-set license plate, pre-set vehicle model, pre-set color, pre-set electronic tag number), personnel feature benchmarks (pre-set facial feature vector, pre-set document information), cargo feature benchmarks (pre-set container number, pre-set seal number, pre-set declared cargo type), and credit assessment data (vehicle credit score, driver credit score, carrier credit score, historical violation records).

[0054] To illustrate with a specific example: A red Dongfeng Tianlong KL container truck with license plate number Lu F·12345 is scheduled to arrive at the port gate at 14:00 on the same day. The driver is Zhang San (whose facial recognition has been registered). The container number is TCLU1234567, the seal number is SEAL-2026-001, the declared cargo is mechanical parts, and the vehicle's credit score is 92. The customs clearance control server 104 starts data aggregation at 12:00, retrieving the above information from five types of business systems. After field mapping and format unification, it generates a pre-set passage feature set for the vehicle and distributes it to the edge computing node 103.

[0055] In another embodiment, the customs clearance control server 104 does not initiate data aggregation immediately upon the creation of a reservation. Instead, it employs a combination of periodic batch synchronization and real-time incremental updates: basic data is synchronized from each business system in batches every preset time interval (e.g., 5 minutes). When a new reservation record is generated in the reservation system or an existing record is changed, a real-time incremental update for that record is triggered. This approach is suitable for port scenarios with a large volume of reservations, balancing system load and data timeliness.

[0056] After the pre-set access feature set is generated, the edge computing node 103 enters a waiting state until a vehicle actually enters the pre-set long-distance detection area 101. When the vehicle arrives, the edge computing node 103 uses the preliminary identity information obtained from the first information collection as an index to locate the corresponding set of the current vehicle in the multiple pre-set access feature sets that have been received, thereby establishing a connection between the pre-constructed static data and the dynamically arriving physical vehicle.

[0057] Step S202: When the target vehicle enters the preset long-distance detection area, the first information collection is performed on the target vehicle, and the corresponding target preset passage feature set is matched according to the collected preliminary identity information; the target vehicle is preliminarily verified according to the target preset passage feature set, and if the preliminary verification is passed, the passage mode of the target vehicle is determined based on the credit assessment data contained in the target preset passage feature set.

[0058] The edge computing node 103 is the main entity executing this step. When the target vehicle enters the preset long-range detection area 101, the telephoto camera 111 and the electronic tag long-range sensing device 112 are triggered to operate. The edge computing node 103 acquires the license plate image of the target vehicle through the telephoto camera 111, extracts the license plate number through optical character recognition, and / or reads the vehicle's on-board electronic tag through the electronic tag long-range sensing device 112 to obtain the clearance event identifier. The aforementioned license plate number or clearance event identifier constitutes the preliminary identity information.

[0059] Edge computing node 103 uses the initial identity information as an index to search and match multiple pre-set access feature sets stored locally. If a match is successful, the pre-set access feature set is determined as the target pre-set access feature set for the current vehicle, and subsequent verification and decision-making are based on this target pre-set access feature set. If a match fails (e.g., the vehicle has not made a reservation), edge computing node 103 sends a guidance command to channel control device 105 to guide the vehicle to the manual verification channel.

[0060] After matching the target preset passage feature set, the edge computing node 103 performs preliminary verification on the target vehicle. The purpose of the preliminary verification is to confirm whether the current physical vehicle and the registered vehicle corresponding to the target preset passage feature set are the same vehicle, so as to prevent cloned vehicles from using the preset passage feature set of legitimate vehicles to pass through the gate.

[0061] Specifically, the preliminary verification includes the following two levels. Edge computing node 103 uses images acquired by telephoto camera 111 and employs visual recognition algorithms to extract the current vehicle model outline and body color features of the target vehicle, comparing them with preset vehicle models and colors in a preset traffic feature set. Since license plates are externally replaceable identifiers, license plate matching alone is insufficient to confirm the vehicle's true identity—a cloned vehicle can display the same license plate number as a legitimate vehicle, but its model and body color often do not match the registration information. Therefore, comparing the vehicle model and color constitutes a crucial step in preventing license plate cloning. When both the vehicle model and color match, the preliminary verification is considered successful.

[0062] Based on the initial verification, edge computing node 103 further calculates a traffic risk assessment value based on the credit assessment data contained in the target pre-set traffic feature set. The calculation of the traffic risk assessment value comprehensively considers multiple dimensions such as vehicle credit score, driver credit score, carrier credit score, and historical violation records. In one specific implementation, the traffic risk assessment value is obtained by weighted summation of the above dimensions, where historical violation records are used as a deduction item, and a preset score is deducted for each violation record.

[0063] Based on the range of the traffic risk assessment value, the edge computing node 103 determines the clearance mode of the target vehicle: when the traffic risk assessment value is lower than the first threshold, the clearance mode is determined to be the seamless passage mode, in which the target vehicle will be automatically released after undergoing multimodal parallel consistency verification in the gate channel 102; when the traffic risk assessment value is between the first threshold and the second threshold, the clearance mode is determined to be the rapid verification mode, in which only the consistency verification of the core perception mode is performed and supplemented by manual spot checks; when the traffic risk assessment value is higher than the second threshold, the clearance mode is determined to be the detailed inspection mode, and the edge computing node 103 sends a guidance instruction to the channel control device 105 to cause the target vehicle to enter the manual verification channel for a comprehensive inspection.

[0064] Continuing with the example of vehicle Lu F·12345: When the vehicle enters the preset long-range detection area 101, 300 meters from the gate, the telephoto camera 111 identifies the license plate number Lu F·12345, and the edge computing node 103 uses this as an index to match the corresponding target preset passage feature set. Subsequently, the visual recognition extracts the current vehicle model as Dongfeng Tianlong KL and the body color as red, which are consistent with the preset vehicle model and preset color in the target preset passage feature set, thus passing the initial verification. Credit assessment data shows that the vehicle's credit score is 92 points, with no historical violation records, and the calculated passage risk assessment value is below the first threshold, determining the passage mode as the seamless passage mode.

[0065] After the clearance mode is determined, the target vehicle continues to travel towards gate channel 102. During the journey from the preset long-range detection area 101 to the entrance of gate channel 102, the edge computing node 103 has completed all the computational tasks of data matching and mode decision-making, and the system switches to a waiting state of parallel verification in motion. When the target vehicle actually enters gate channel 102, the multimodal sensing device group deployed in the channel is triggered and begins to perform the second information collection.

[0066] Step S203: When the clearance mode is the contactless passage mode, when the target vehicle enters the gate channel, a second information collection is performed on the target vehicle and the driver. The second information collection uses at least two different perception modes to collect real-time features in parallel, and the real-time features are checked for consistency with the corresponding dimension of the preset passage feature set.

[0067] The execution entity for this step is edge computing node 103. Unlike step S202, which only performs license plate recognition and coarse vehicle color comparison under long-distance conditions, step S203 performs more refined multi-dimensional consistency verification on the target vehicle and driver under close-range conditions (typically 10 to 50 meters) at the gate channel 102. The close-range conditions enable the implementation of perception modalities such as face capture and seal recognition, which have high requirements for image resolution and acquisition angle.

[0068] The aforementioned at least two different perception modalities refer to perception methods based on different physical principles. In one specific implementation, the second information collection employs the following four perception modalities in parallel to construct a consistency verification system for the entire vehicle-person-goods chain.

[0069] At the vehicle level, the close-range image acquisition device 121 acquires the license plate information, vehicle model outline features, and body color features of the target vehicle, and compares them with the preset license plate, vehicle model, and color in the preset traffic feature set. The license plate, vehicle model, and color comparison here differs in purpose from the preliminary verification in step S202: the comparison in step S202 occurs at a distance of 200 to 500 meters, where the image resolution is limited, primarily used to quickly eliminate obvious suspicions of license plate counterfeiting; the comparison in step S203 occurs at a close distance of 10 to 50 meters, where the image resolution is significantly improved, enabling the detection of more subtle feature differences and constituting accurate vehicle identification.

[0070] In terms of personnel, the face capture device 122 acquires the driver's facial image features and compares them with the preset personnel features (i.e., pre-stored facial feature vectors) in the target preset passage feature set. The face capture device 122 is installed at a height of approximately 1.5 meters on the side of the gate passage 102, facing the driver's seat, and captures the driver's facial image during vehicle movement. Facial feature comparison uses feature vector cosine similarity calculation; when the similarity exceeds a preset threshold (e.g., 0.85), it is determined to be the same person.

[0071] At the electronic tag level, the radio frequency reader 123 reads the vehicle-mounted electronic tag information of the target vehicle at close range and compares it with the preset electronic tag number in the target's preset traffic feature set. The reliability of close-range radio frequency reading is much higher than that of long-range sensing, and it can eliminate the crosstalk of adjacent lane signals that may occur under long-range conditions.

[0072] At the document level, after facial recognition is successful, edge computing node 103 automatically retrieves the driver's electronic document information (including driver's license, vehicle registration certificate, and professional qualification certificate) from the backend personnel database, verifying the validity and status of each document. This retrieval process does not require the driver to present any physical documents; the system automatically completes the status query based on the document numbers in the pre-set target access feature set. If any document has expired or been revoked, the consistency check result for the corresponding dimension will be "failed."

[0073] At the cargo level, the seal identification device 124 acquires the cargo seal identifier and / or container number, and compares it with the pre-set cargo features (pre-set seal number and pre-set container number) in the target pre-set access feature set. The seal identification device 124 can read the number characters on the seal using image recognition or read the electronic seal tag using radio frequency identification.

[0074] like Figure 3 As shown, the spatial layout of the four sensing modalities within the gate channel 102 and the timing relationship of their acquisition are as follows: the near-field image acquisition device 121 is deployed above the channel entrance, the face acquisition device 122 is deployed on the side of the middle section of the channel, the radio frequency reading device 123 is deployed in the middle section of the channel, and the seal recognition device 124 is deployed on the side of the rear section of the channel. When the target vehicle passes through the 10-meter to 50-meter gate channel 102 at a speed of 30 km / h, the available acquisition time is approximately 1.2 to 6 seconds. The acquisition windows of the four sensing modalities overlap on the time axis—each device begins acquisition as soon as the vehicle enters its effective field of view, without waiting for the previous device to complete acquisition before starting. Therefore, the total time for multimodal parallel acquisition depends on the slowest single modality, rather than the sum of the response times of each modality. This parallel architecture ensures that the increase in verification dimensions does not lead to a linear increase in verification time.

[0075] It should be noted that, for ease of annotation in the accompanying drawings, Figure 3 The above-mentioned devices are referred to as close-range image acquisition device 121, face acquisition device 122, radio frequency reading device 123, and seal recognition device 124.

[0076] After the consistency verification is completed, the edge computing node 103 needs to make a pass decision based on the verification results of each sensing modality. The decision logic of this verification result can be configured into multiple modes according to the security level requirements.

[0077] In a more stringent decision-making mode, edge computing node 103 determines that the consistency check has passed only when all sensing modalities pass the consistency check, and sends a release command to channel control device 105. If any sensing modality fails the check, the consistency check is deemed to have failed, triggering a manual review process. This mode is suitable for scenarios with high security requirements.

[0078] In another, more flexible decision-making mode, the edge computing node 103 weights and fuses the consistency verification results of each perception modality according to preset weights to obtain a comprehensive verification score. When the comprehensive verification score exceeds a preset threshold, the consistency verification is deemed to have passed. This mode allows for slight deviations in individual non-critical modalities (such as minor color recognition shifts caused by changes in lighting), and passage is permitted as long as the comprehensive score still meets the standard. It is suitable for scenarios with high requirements for passage efficiency.

[0079] In another compromise decision-making model, edge computing node 103 divides the perception modality into core modality and auxiliary modality. The consistency verification result of the core modality (such as license plate comparison and face comparison) must pass; the verification results of the auxiliary modality (such as color comparison and seal comparison) are weighted and scored. When the weighted score exceeds the auxiliary threshold, the overall result is considered to pass. This model provides fault tolerance for slight deviations in the auxiliary modality caused by environmental interference, while ensuring that key identity verification cannot be bypassed.

[0080] Continuing with the example of vehicle Lu F·12345: After the vehicle entered gate channel 102 at a speed of approximately 30 km / h, the close-range image acquisition device 121 acquired the license plate Lu F·12345, vehicle type Dongfeng Tianlong KL, and color red, all of which are consistent with the target preset passage feature set; the face acquisition device 122 acquired the facial features of the driver Zhang San, and the similarity with the preset personnel features was 0.93, exceeding the threshold; the radio frequency reading device 123 read the electronic tag number, which is consistent with the preset electronic tag; the seal recognition device 124 recognized the seal number SEAL-2026-001 and container number TCLU1234567, both of which are consistent with the preset cargo features. All four modalities passed, and the consistency verification result was passed.

[0081] Simultaneously with identity consistency verification (see...) Figure 3 The system also performs physical security checks on the target vehicle. For example... Figure 4 As shown, during the process of the target vehicle entering the gate channel 102, the security inspection equipment group 130 performs parallel security scans on the target vehicle and its cargo using at least two different physical detection methods to acquire multi-source security inspection data. In one specific embodiment, the security inspection equipment group 130 includes the following physical detection methods: X-ray scanning device or gamma-ray scanning device (used for penetrating imaging to detect the internal structure of cargo and identify contraband and concealed items); millimeter-wave radar (used for detecting abnormal density distribution of cargo and identifying hidden layers); gas detection sensor (used for detecting volatile flammable and explosive substances or toxic gases); radiation detector (used for detecting radioactive substances); chassis scanning system (used for detecting whether personnel or items are hidden under the vehicle chassis); thermal imaging detection device (used for detecting abnormal temperatures, such as temperature deviations in cold chain cargo or human heat sources). The above-mentioned physical detection methods operate based on different physical principles, covering different detection dimensions such as internal structure, density distribution, chemical composition, radioactivity, external attachments, and temperature. The physical blind spots of a single mode can be complemented by other modes. Those skilled in the art can select at least two of them for combination according to actual security inspection needs.

[0082] Because the acquisition frequency, triggering time, and data format of each physical detection method are different, the security inspection data they output are asynchronous on the time axis. To associate the asynchronous data with the same target vehicle, the edge computing node 103 spatially aligns the asynchronous security inspection data acquired by each physical detection method with a unified timestamp and vehicle motion trajectory model. Specifically, taking the moment when the target vehicle passes through the entrance of the gate channel 102 as the zero point of time, and combining it with the real-time vehicle speed, the effective acquisition window of each sensor on the time axis is calculated. The data acquired by each sensor within this window is mapped to a unified vehicle coordinate system, thereby associating it with the same target vehicle entity.

[0083] After spatial alignment, edge computing node 103 uses a pre-trained deep learning model to analyze security inspection data of various modalities. For X-ray scan images, the deep learning model identifies object categories in the image and marks the bounding boxes and category labels of suspicious areas on the image, outputting the contraband identification result; for millimeter-wave radar data, the model detects whether there are abnormally concentrated areas in the density distribution and outputs the density anomaly detection result. In one specific implementation, the deep learning model adopts a hybrid architecture of convolutional neural networks and Transformers. The convolutional part is used to extract local texture features of X-ray images, and the Transformer part is used to capture the global spatial relationships between objects. The two work together to improve the ability to identify occluded and overlapping objects.

[0084] When the confidence level of the deep learning model in determining a certain region is lower than a preset confidence threshold, the edge computing node 103 marks the region as a region to be reviewed and pushes the X-ray image of the marked suspicious region, along with relevant information from the vehicle's target preset traffic feature set (such as the carrier's historical violation records and declared cargo type), to the human review station. The human reviewer views the AI-marked suspicious region and vehicle historical information on a visual interface and makes a final judgment. This AI-assisted annotation mechanism reduces manual image judgment from full-image search to local confirmation, significantly reducing the cognitive load on humans and the risk of missed detections.

[0085] The security analysis results of the above modalities constitute the basis for determining the physical security dimension.

[0086] The aforementioned identity verification results and security check analysis results are not isolated judgments, but are gradually accumulated into a unified risk metric during the vehicle's movement. As the target vehicle enters the gate channel 102 from the preset remote detection area 101, the edge computing node 103 gradually accumulates risk scores for each dimension based on the consistency verification results at each stage and the security check analysis results for each modality. Specifically, in the remote detection area, the preliminary verification results generate the first risk score adjustment—no points are added if the vehicle model and color match, and a corresponding risk score is added if there is a slight deviation; within the gate channel, the consistency verification results for each modality generate subsequent risk score adjustments; the security check analysis results also generate corresponding risk score adjustments.

[0087] The accumulated risk score is then fused across dimensions with the credit assessment data contained in the target pre-set access feature set to output a comprehensive risk assessment value. This fusion process incorporates identity verification, physical security checks, and historical credit data into a unified assessment framework—for example, density anomalies detected in X-ray images may be normal on vehicles with high credit scores and declared goods as metal parts, but would trigger a higher risk score on vehicles with low credit scores and declared goods as ordinary daily necessities.

[0088] Based on the comprehensive risk assessment value, the edge computing node 103 sends the access control command corresponding to the comprehensive risk assessment value to the channel control device 105: when the comprehensive risk assessment value is lower than the third threshold, an automatic release command is sent; when the comprehensive risk assessment value is between the third threshold and the fourth threshold, a command to guide manual review is sent; when the comprehensive risk assessment value is higher than the fourth threshold, an interception alarm command is sent.

[0089] Taking the aforementioned vehicle, Lu F·12345, as an example: X-ray scans did not detect any contraband, millimeter-wave radar did not detect any density anomalies, and gas detection sensors did not trigger alarms. All modal security inspection analysis results were normal, and no additional risk score was added. The overall risk assessment value was at an extremely low level, and edge computing node 103 sent an automatic release command to channel control equipment 105.

[0090] In addition to the threshold-based tiered handling mentioned above, the system also has the ability to identify proactive attack-type abnormal behavior in real time. Its core logic lies in: not relying on a known threat list, but rather on the deviation detection between the expected state defined by the target's preset access feature set and the actual state obtained through real-time perception.

[0091] Specifically, edge computing node 103 performs cross-dimensional comparison between the real-time features obtained from the first and / or second information acquisition and the corresponding preset features in the target preset passage feature set, calculating the deviation degree of each dimension. When only a single dimension shows deviation, it may be due to sensor error or environmental interference (such as color recognition shift caused by changes in light), and the system does not immediately determine it as abnormal. However, when the deviation degree of at least two dimensions exceeds the preset deviation threshold simultaneously, the probability of multi-dimensional joint deviation is much lower than that of single-dimensional deviation, and the false alarm probability decreases exponentially. At this time, edge computing node 103 determines that the target vehicle has abnormal behavior, sends an interception command to the channel control device 105, and triggers a real-time alarm.

[0092] In one specific implementation, the aforementioned abnormal behaviors include the following types: abnormal vehicle identity, i.e., the license plate information does not match the preset vehicle type and / or preset color in the target preset traffic feature set, indicating possible license plate cloning; abnormal personnel identity, i.e., the driver's facial features are inconsistent with the preset personnel features in the target preset traffic feature set, indicating possible impersonation; abnormal cargo integrity, i.e., the actual container number or seal mark is inconsistent with the preset cargo features in the target preset traffic feature set, indicating possible cargo swapping; abnormal trajectory, i.e., the target vehicle's driving trajectory deviates from the preset path or the time spent in non-parking areas exceeds a preset threshold, indicating possible loitering reconnaissance or detour avoidance; abnormal following, i.e., the distance between the target vehicle and the vehicle in front is less than the preset safe distance, indicating possible following and breaking through checkpoints; abnormal personnel behavior, i.e., abnormal actions such as the driver or accompanying personnel delivering items or exchanging seats are detected through video analysis; and abnormal documents, i.e., after retrieving electronic documents, it is found that the documents have expired, been revoked, or do not match the registration information.

[0093] To illustrate with an example of a vehicle using a counterfeit license plate: A vehicle with license plate number Lu F·67890 enters a preset long-range detection area 101. Edge computing node 103 matches the corresponding target preset passage feature set (the registered vehicle is a red Dongfeng Tianlong KL). However, the current vehicle model extracted by visual recognition is a blue Jiefang J6, which does not match either the preset vehicle model or preset color. The deviation in both the vehicle model and color dimensions exceeds the threshold. In the subsequent consistency check within the gate channel 102, the driver's face is inconsistent with the preset personnel features (similarity is only 45%, far below the threshold), and the deviation in the personnel dimension also exceeds the threshold. With significant deviations in all three dimensions, edge computing node 103 determines that the vehicle is engaging in abnormal behavior of using a counterfeit license plate and impersonation. It immediately sends an interception command to the channel control device 105, the gate does not lift, and simultaneously triggers an audible and visual alarm and notifies security personnel.

[0094] Furthermore, this specification also includes an emergency handling mechanism. When the following situations occur during the consistency verification process, the system automatically switches to the manual verification channel: the target vehicle fails to match the preset passage feature set (i.e., a vehicle without a reservation); a single sensing modality cannot complete data collection due to equipment failure or environmental interference; the AI ​​model's confidence level is lower than the preset confidence threshold and cannot pass cross-validation with other modalities; the passage risk assessment value falls within the range corresponding to the rapid review mode. In the manual verification channel, security personnel view all the information collected by the system and the AI ​​annotation results through a visual monitoring interface, make a manual judgment, and then manually control the channel control device 105 to either allow passage or intercept. This emergency handling mechanism ensures that the system still has the capability to handle security situations that cannot be covered by the automated process.

[0095] Step S204: Based on the result of the consistency check, send a passage control command to the channel control device.

[0096] The edge computing node 103 is the main entity executing this step. The edge computing node 103 converts the consistency verification result (and optional security inspection fusion decision result and abnormal behavior identification result) into physical control commands that can be executed by the channel control device 105. After receiving the command, the channel control device 105 executes the corresponding physical action: raise the barrier to allow passage, keep it closed, guide to the designated channel, or trigger the interception device.

[0097] Continuing with the example of vehicle Lu F·12345: all consistency checks passed, the overall risk assessment value was extremely low, and no abnormal behavior was triggered. The edge computing node 103 sent a gate-raising command to the channel control device 105, the gate automatically raised, and the vehicle passed through without being noticed. From the time the vehicle entered the preset long-distance detection area 101 (300 meters from the gate) to the time the gate raised and allowed passage, the entire process took approximately 28 seconds. During this time, the driver did not need to stop, get out of the vehicle, or show any documents.

[0098] This specification also provides an embodiment of a customs clearance control system for port gates. For example... Figure 5 As shown, this system can be deployed in the aforementioned edge computing node 103 and customs clearance management server 104, serving as the main body for executing the aforementioned method steps. The system includes the following units.

[0099] The pre-aggregation unit 501 is used to pre-connect to multiple business systems, perform field mapping and structured fusion based on the multi-dimensional feature data maintained by each business system, and generate a corresponding pre-set passage feature set for each reserved vehicle.

[0100] The remote initial identification unit 502 is used to perform the first information collection on the target vehicle when the target vehicle enters the preset remote detection area, and match the corresponding target preset passage feature set according to the collected preliminary identity information; perform preliminary verification on the target vehicle according to the target preset passage feature set, and determine the passage mode of the target vehicle based on the credit assessment data contained in the target preset passage feature set if the preliminary verification is passed.

[0101] The multimodal travel verification unit 503 is used to collect real-time features in parallel through at least two different perception modalities when the target vehicle enters the gate channel in the case of the non-contact passage mode. The real-time features are then used to verify the consistency between the real-time features and the corresponding dimensions of the target preset passage feature set.

[0102] The channel control unit 504 is used to send access control commands to the channel control device based on the result of the consistency check.

[0103] Optionally, the pre-aggregation unit 501 is specifically used to: obtain vehicle registration information, reservation information, cargo declaration information, driver registration information, and credit assessment information from the vehicle management office database, port reservation system, freight management system, personnel database, and credit management system, respectively; map the fields and unify the formats of records in the above information that point to the same customs clearance event; eliminate semantic conflicts between heterogeneous data; and construct a unified data structure centered on the customs clearance event.

[0104] Optionally, the remote initial recognition unit 502 is specifically used for: obtaining the current vehicle model and body color features of the target vehicle through visual recognition, comparing them with the preset vehicle model and preset color in the target preset passage feature set, and determining that the preliminary verification is passed when the comparison is consistent; calculating the passage risk assessment value based on credit assessment data, and determining the passage mode as seamless passage mode, rapid review mode or detailed inspection mode based on the comparison results of the passage risk assessment value with the first threshold and the second threshold.

[0105] Optionally, the multimodal travel verification unit 503 is specifically used for: comparing license plate information, vehicle model features and vehicle body color features obtained through an image acquisition device; comparing driver's facial image features obtained through a face acquisition device; comparing vehicle electronic tag information obtained through an RFID reader; and comparing cargo seal identification and / or container number obtained through a seal recognition device.

[0106] Optionally, the channel control unit 504 is specifically used to: send a release command when all sensing modal verifications pass, or send a release command based on a comprehensive score after weighted fusion of the verification results of each modality, or send a release command when the core modality passes and the auxiliary modality score meets the standard.

[0107] Optionally, the system also includes: a security inspection scanning unit 505, used to perform parallel security inspection scanning of the target vehicle and its cargo using at least two different physical detection methods, spatially align asynchronous security inspection data using a unified timestamp and vehicle motion trajectory model, analyze security inspection data of each modality using a pre-trained deep learning model, and output security inspection analysis results of each modality.

[0108] Optionally, the system also includes a risk calculation unit 506, which is used to gradually accumulate risk scores of each dimension based on the consistency verification results of each stage and the security inspection analysis results of each modality during the movement of the target vehicle, integrate the accumulated risk scores with credit assessment data across dimensions to output a comprehensive risk assessment value, and drive differentiated passage handling based on the comprehensive risk assessment value.

[0109] Optionally, the system also includes: an anomaly interception unit 507, which is used to perform cross-dimensional cross-comparison between real-time features and preset features of corresponding dimensions in the target preset access feature set and calculate the deviation degree of each dimension. When the deviation degree of at least two dimensions exceeds the preset deviation threshold, it is determined that there is abnormal behavior, and an interception command is sent to the channel control device and a real-time alarm is triggered.

[0110] The specific functional implementation, interaction logic, and technical effects of each of the above units are consistent with the descriptions in the aforementioned method embodiments, and will not be repeated here.

[0111] This specification also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the aforementioned customs clearance control method.

[0112] This specification also provides an electronic device through its embodiments. For example... Figure 6As shown, the electronic device includes a processor 601, a memory 602, and a communication interface 603, which are connected via a bus 604. The memory 602 stores a computer program, which the processor 601 executes to implement the steps of the aforementioned customs clearance control method. The communication interface 603 is used for data communication with various sensing devices and channel control devices.

[0113] As can be seen from the above embodiments, the customs clearance control method provided in this specification generates a pre-set passage feature set by pre-connecting multiple business systems, shifting the acquisition and fusion of multi-source heterogeneous data from after the vehicle arrives at the gate to before arrival, thus eliminating the network latency of cross-system serial queries after arrival. By performing the first information collection and completing the pattern decision in a preset long-distance detection area, and using the physical travel time of the vehicle approaching the gate as the processing window for data matching and decision-making, the method achieves the goal of trading physical spatial distance for data preparation time. By using at least two different perception modalities to collect real-time features in parallel for consistency verification during the passage through the gate channel, the positive correlation constraint between dimensional growth and time consumption in traditional serial item-by-item verification is broken, enabling multi-dimensional deep verification to be completed within a very short physical passage window. By directly sending passage control commands to the channel control equipment based on the consistency verification results, the direct mapping from digital spatial verification results to physical spatial passage control is realized, achieving non-stop passage of vehicles while ensuring multi-dimensional identity consistency verification.

Claims

1. A method for customs clearance control at port gates, characterized in that, include: It pre-connects to multiple business systems, performs field mapping and structured fusion based on the multi-dimensional feature data maintained by each business system, and generates a corresponding pre-set access feature set for each reserved vehicle; When a target vehicle enters a preset long-range detection area, the first information collection is performed on the target vehicle, and the corresponding target preset passage feature set is matched based on the collected preliminary identity information; the target vehicle is preliminarily verified based on the target preset passage feature set, and if the preliminary verification is passed, the passage mode of the target vehicle is determined based on the credit assessment data contained in the target preset passage feature set; When the passage mode is the seamless passage mode, when the target vehicle enters the gate channel, a second information collection is performed on the target vehicle and the driver. The second information collection uses at least two different perception modes to collect real-time features in parallel, and the real-time features are verified to be consistent with the corresponding dimension of the preset features in the target preset passage feature set. Based on the result of the consistency verification, a passage control command is sent to the channel control device.

2. The method according to claim 1, characterized in that, The process involves mapping and merging fields based on the multidimensional feature data maintained by each business system to generate a pre-set access feature set for each reserved vehicle, including: Vehicle registration information, reservation information, cargo declaration information, driver registration information, and credit assessment information are obtained from the vehicle management office database, port reservation system, freight management system, personnel database, and credit management system, respectively. Records pointing to the same clearance event in the above information are mapped by fields and formatted to eliminate semantic conflicts between heterogeneous data, and a unified data structure centered on the clearance event is constructed as the preset passage feature set for the reserved vehicle.

3. The method according to claim 1, characterized in that, The step of performing preliminary verification of the target vehicle based on the target preset passage feature set, and determining the passage mode of the target vehicle based on the credit assessment data contained in the target preset passage feature set if the preliminary verification is passed, includes: The current vehicle model and body color features of the target vehicle are obtained through visual recognition and compared with the preset vehicle model and preset color in the preset traffic feature set of the target. If the vehicle model and color match, the preliminary verification is deemed successful. The passage risk assessment value is calculated based on the credit assessment data contained in the target preset passage feature set; If the passage risk assessment value is lower than the first threshold, the passage mode is determined to be a seamless passage mode; If the passage risk assessment value is between the first threshold and the second threshold, the passage mode is determined to be the rapid review mode. In the rapid review mode, the consistency verification of the core perception mode is performed and supplemented by manual spot checks. If the passage risk assessment value is higher than the second threshold, the passage mode is determined to be the detailed inspection mode, and a guidance instruction is sent to the channel control device to cause the target vehicle to enter the manual verification channel.

4. The method according to claim 1, characterized in that, The second information acquisition employs at least two different perception modalities to acquire real-time features in parallel, and performs consistency verification between the real-time features and the corresponding dimensions of the preset access feature set, including: The license plate information, vehicle type characteristics, and body color characteristics of the target vehicle are acquired by an image acquisition device and compared with the preset license plate, preset vehicle type, and preset color in the preset traffic feature set of the target vehicle. The facial image features of the driver are acquired by a facial recognition device and compared with the preset personnel features in the target preset passage feature set; The vehicle-mounted electronic tag information of the target vehicle is obtained by a radio frequency reading device and compared with the preset electronic tags in the preset access feature set of the target vehicle. The cargo seal identification device obtains the cargo seal identifier and / or container number, and compares it with the pre-set cargo features in the target pre-set access feature set.

5. The method according to claim 4, characterized in that, The step of sending a passage control command to the channel control device based on the result of the consistency check includes: If the consistency check results of all sensing modalities pass, a release command is sent to the channel control device; or, The consistency verification results of each sensing modality are weighted and fused according to preset weights to obtain a comprehensive verification score. If the comprehensive verification score exceeds a preset threshold, a release command is sent to the channel control device; or, If the consistency verification result of the preset core sensing modality is passed and the weighted score of the auxiliary sensing modality exceeds the auxiliary threshold, a release command is sent to the channel control device.

6. The method according to claim 1, characterized in that, During the process of the target vehicle entering the gate channel, the method further includes: The target vehicle and its cargo are subjected to parallel security scans using at least two different physical detection methods to obtain multi-source security inspection data. The asynchronous security inspection data collected by each physical detection method are spatially aligned and associated with the target vehicle using a unified timestamp and vehicle motion trajectory model. The pre-trained deep learning model is used to analyze the security inspection data of each modality and output the security inspection analysis results of each modality.

7. The method according to claim 6, characterized in that, The method further includes: During the process of the target vehicle entering the gate channel from the preset long-distance detection area, based on the consistency verification results of each stage and the security inspection analysis results of each modality, the risk scores of each dimension are gradually accumulated, and the accumulated risk scores are fused with the credit assessment data contained in the target preset passage feature set across dimensions to output a comprehensive risk assessment value. The step of sending a passage control command to the channel control device based on the result of the consistency check includes: Based on the comprehensive risk assessment value, a passage control instruction corresponding to the comprehensive risk assessment value is sent to the channel control device. The passage control instruction includes automatic release, guidance for manual review, or interception alarm.

8. The method according to claim 1, characterized in that, The method further includes: The real-time features obtained from the first and / or second information collections are cross-compared with the corresponding preset features in the target preset access feature set, and the deviation of each dimension is calculated. If the deviation in at least two dimensions exceeds a preset deviation threshold, the target vehicle is determined to have abnormal behavior, and an interception command is sent to the channel control device and a real-time alarm is triggered.

9. A customs clearance control system for port gates, characterized in that, include: The pre-aggregation unit is used to pre-connect to multiple business systems, perform field mapping and structured fusion based on the multi-dimensional feature data maintained by each business system, and generate a corresponding pre-set passage feature set for each reserved vehicle. The long-range initial identification unit is used to perform the first information collection on the target vehicle when the target vehicle enters the preset long-range detection area, and match the corresponding target preset passage feature set according to the collected preliminary identity information; perform preliminary verification on the target vehicle according to the target preset passage feature set, and determine the passage mode of the target vehicle based on the credit assessment data contained in the target preset passage feature set if the preliminary verification is passed; A multimodal travel verification unit is used to collect real-time features in parallel using at least two different sensing modalities when the target vehicle enters the gate channel in the case where the passage mode is a contactless passage mode, and to verify the consistency of the real-time features with the corresponding dimension of the preset features in the target preset passage feature set. The channel control unit is used to send access control commands to the channel control device based on the result of the consistency verification.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 8.

Images