Green-passing vehicle AI auditing and checking method and system based on multi-dimensional feature fusion

Abstract

The present application relates to the technical field of highway operation management, in particular to a green pass vehicle AI auditing and checking method and system based on multi-dimensional feature fusion; the method comprises the following steps: collecting multi-dimensional data including green pass checking data, overload weighing data, transaction flow data, license plate data, driving license data and agricultural product price data; when the green pass vehicle enters the exit lane, parallel analysis is performed on eleven risk dimensions including non-compliance list matching, regional collaborative investigation, weight fluctuation, production and sales location judgment, suspicious reverse goods, long distance low value, cross-provincial large amount, low frequency vehicle, low flatbed truck, volume and mass analysis, and vehicle cargo state monitoring, and a warning is generated; the present application solves the problems of single dimension, separation of in-process and post-process, and inability to accurately monitor the state of goods in transit in the prior art, and simultaneously improves the checking efficiency and accuracy and significantly increases the toll revenue.

Description

Technical Field

[0001] This invention relates to the field of highway operation and management technology, specifically to an AI-based audit and inspection method and system for green channel vehicles based on multi-dimensional feature fusion. Background Technology

[0002] With the continuous development of highway toll operation and management models, the implementation of the green channel (hereinafter referred to as green channel) policy has facilitated the transportation of fresh agricultural products. However, at the same time, the phenomenon of some criminals using the green channel policy to evade tolls by impersonating green channel vehicles or mixing non-green channel goods has become increasingly serious. According to statistics, the annual free toll amount for green channel vehicles on the provincial highway sections under the jurisdiction of a certain provincial highway company is about 948 million yuan, indicating huge potential for plugging loopholes and increasing revenue. Currently, the inspection of green channel vehicles on highways mainly adopts the traditional model of checking every vehicle and manually opening and verifying the cargo box. This model has exposed many problems in the face of the increasing volume of green channel vehicles: First, the inspection efficiency is low, with an average inspection time of about 5 minutes per vehicle, which can easily cause congestion at toll stations during peak hours; second, the accuracy of the inspection is greatly affected by factors such as the experience and understanding of policies of the inspectors, making it difficult to ensure the consistency of inspection standards; third, there is a lack of systematic audit and analysis methods after the event, making it difficult to detect inspection loopholes in a timely manner. In existing technologies, some solutions attempt to improve green channel inspection work using single technological means. For example, some solutions use X-ray detection equipment to scan and image the carriages, and determine the type of goods through image analysis. However, these solutions only target the on-site inspection stage and do not involve the fusion analysis of multi-dimensional data, and the equipment cost is high. Other solutions use weighing and volume detection systems to measure the physical data of vehicles and determine the compliance of goods by calculating the density of goods, such as a green channel goods inspection method based on equivalent density and big data analysis. However, this method only judges from the single dimension of density and lacks the fusion analysis of multi-dimensional features such as vehicle passage behavior, historical inspection records, and production and sales information. There is also a method based on... The big data analysis and judgment method for green channel vehicle inspection combines green channel inspection data with toll vehicle passage data for data analysis. However, this method focuses on post-event analysis and judgment, lacking real-time risk reminders and auxiliary inspection functions during the event, and cannot provide timely and effective decision support for inspectors at the inspection site. In addition, the patented method, device, equipment and storage medium for green channel inspection of highway freight vehicles (CN119106045A) obtains green channel inspection results by comparing list data, highway passage behavior data and historical green channel inspection results data. However, the data dimensions of this solution are limited, and it does not involve key dimensions such as judgment of production and sales location information, analysis of the reasonableness of cargo value and transportation distance, and regional collaborative investigation.

[0003] Furthermore, existing technologies are almost entirely lacking in monitoring potential cheating behaviors by green channel vehicles during their stops at highway service areas, such as cargo swapping (i.e., unloading compliant green channel goods and replacing them with non-compliant goods, or mixing and disguising goods) and watering to preserve freshness (watering goods to increase weight or disguise freshness). Some solutions rely solely on the weight difference between entry and exit points, but watering can also increase weight, making it prone to misjudgment. On the other hand, simply relying on whether the duration of the stop at the service area is abnormal cannot distinguish between normal rest, vehicle inspection, and genuine cheating. In view of this, in order to overcome the above-mentioned technical problems, the present invention proposes an AI audit and inspection method and system for green channel vehicles based on multi-dimensional feature fusion. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-based audit and inspection method and system for green channel vehicles based on multi-dimensional feature fusion, in order to solve the problems of single data dimension, separation between in-process inspection and post-process audit, lack of regional collaborative investigation capability, and inability to accurately monitor changes in the status of goods on the road in existing technologies, thereby achieving simultaneous improvement in the efficiency and accuracy of green channel vehicle inspection and effectively curbing the behavior of counterfeit green channel vehicles evading tolls.

[0005] To achieve the above objectives, the present invention provides the following technical solution: The AI-based audit and inspection method for green channel vehicles based on multi-dimensional feature fusion includes the following: S1, Data Aggregation: Collect multi-dimensional data from the highway toll system. The multi-dimensional data includes green channel inspection data, entrance overload weighing data, entrance and exit transaction data, gantry transaction data, gantry license plate data, vehicle registration data, and agricultural product market price data. S2, Real-time Multi-dimensional Inspection and Early Warning: When a green channel vehicle enters the exit lane of the toll station, the system retrieves the current multi-dimensional data of the vehicle in real time, analyzes it in parallel from multiple risk dimensions, and generates the analysis results of each risk dimension; when any risk dimension triggers the early warning condition, an early warning message is generated and pushed to the inspection terminal. The early warning message includes the risk dimension identifier and the description of the risk cause, prompting the inspection personnel to conduct key inspections. The multiple risk dimensions include: a non-compliance list matching dimension based on historical inspection data; a regional collaborative investigation dimension based on information sharing among multiple toll stations within the jurisdiction; an entry and exit weight fluctuation analysis dimension; a production and sales location rationality judgment dimension based on a production and sales location information database; a cargo diversion suspicion analysis dimension based on a comparison of standard passage time and actual travel time; a long-distance low-value cargo risk analysis dimension based on a comparison of total cargo value and toll amount; a cross-provincial large amount early warning dimension based on cross-provincial passage records and toll amount; a low-frequency vehicle analysis dimension based on the frequency of vehicle passage at the current toll station; a low-flatbed truck identification and early warning dimension based on vehicle type and historical inspection records; a vehicle and cargo volume and mass analysis dimension based on a reasonable range of total vehicle and cargo mass; and a cargo status monitoring and early warning dimension based on vehicle sensor signals. S3, AI-assisted inspection: During the inspection process, the inspection terminal collects images of the goods and uploads them to the AI ​​recognition module. The AI ​​recognition module uses a trained large model to recognize the images of the goods and outputs the name of the goods, the green channel catalog matching result, and the freshness assessment information. If the goods are not in the green channel catalog, the inspection terminal will pop up a prompt box to remind the inspection personnel to verify further. At the same time, the inspection terminal retrieves the trajectory photos of the vehicle at the entrance station and the gantry nodes it passes through from the system for the inspection personnel to compare the changes in the vehicle's appearance. S4, Post-event Audit and Analysis by Management: Perform multi-dimensional post-event analysis on the green channel vehicle data that has been inspected. The analysis dimensions include: analysis of changes in vehicle types before and after exiting the vehicle, analysis of the distribution patterns of non-compliant green channel vehicles, analysis of abnormal green channel exemption amounts at toll stations, analysis of differences in inspection indicators of inspection personnel, and dynamic management of the key inspection list database; when abnormal indicator data is found, an alarm is triggered to notify relevant management personnel to conduct a review. S5, Closed-loop feedback: The audit analysis results generated in S4 are fed back to the early warning rules in S2 in real time, and the early warning thresholds and weights of each risk dimension are dynamically adjusted to form a closed-loop linkage mechanism of in-process early warning, post-event audit, and rule optimization.

[0006] Preferably, the specific implementation of the on-board cargo status monitoring and early warning dimension in S2 is as follows: When a green channel vehicle enters the highway entrance, sensors are installed at the rear cargo bed door or key cargo loading points. These sensors can detect the door opening / closing status, cargo displacement status, or changes in the cargo bed's internal environment, and upload the status signals to the system in real-time or periodically. While the vehicle is stopped at a highway service area, the system receives and monitors the sensor signal status. If the sensor signal is normal and the status remains unchanged, only the vehicle's stop time is recorded and compared with the vehicle's standard travel time to determine if the stop time exceeds a reasonable fluctuation range. If the sensor... If the sensor signal is interrupted or the sensor emits an abnormal status signal indicating that the door has been opened or the goods have been moved, the system will automatically retrieve the monitoring video of the corresponding time period in the service area, perform proactive behavior analysis on the video, and identify the type of operation that the vehicle is performing. The operation type includes at least watering, cargo handling, and cargo exchange. Then, the system will generate an early warning message by combining the analyzed operation type and the corresponding video clip information, and send it to the terminal equipment of the highway management personnel. When the vehicle arrives at the exit toll station, the system will automatically push a reminder on the inspection terminal so that the inspection personnel can retrieve the relevant video clips for on-site comparison and confirmation.

[0007] This invention also provides an AI-based audit and inspection system for green channel vehicles based on multi-dimensional feature fusion, comprising: The data aggregation module interfaces with the highway toll collection system to collect and integrate multi-dimensional data. The real-time multi-dimensional inspection and early warning module is connected to the data aggregation module and is used to retrieve the current multi-dimensional data of the vehicle in real time, analyze it in parallel from multiple risk dimensions and generate early warning information. The multiple risk dimensions include the vehicle cargo status monitoring and early warning dimension. The real-time multi-dimensional inspection and early warning module is also connected to the sensor signal receiving device and video monitoring system deployed in the service area to obtain vehicle sensor status data and service area monitoring video. The AI-assisted inspection module includes an AI image recognition unit and a trajectory photo tracing unit. The AI ​​image recognition unit is used to receive cargo images collected by the inspection terminal and use a large AI model to identify cargo information. The trajectory photo tracing unit is used to retrieve trajectory photos of the vehicle's current journey from the gantry sign recognition system and push them to the inspection terminal. The quick inspection assistance module is used to automatically calculate the reasonable range of total vehicle and cargo weight based on the parameters of the vehicle registration certificate, and automatically display a recent photo of the vehicle on the inspection terminal. The management audit and analysis module is used to perform multi-dimensional post-event analysis on the green channel vehicle data that has been inspected and to trigger anomaly alarms. The closed-loop feedback module is connected to the in-process multi-dimensional inspection and early warning module and the management terminal audit and analysis module, respectively, and is used to feed back the audit and analysis results to the early warning rules and adjust the early warning threshold and weight. The existing system optimization module is used to realize real-time synchronization of inspection data, reminders of data to be inspected, one-click login, and operation interface optimization functions.

[0008] The beneficial effects of this invention are as follows: 1. The invention integrates seven types of data and performs parallel analysis from at least eleven risk dimensions, enabling a comprehensive and multi-faceted assessment of the compliance of green channel vehicles, and significantly improving the comprehensiveness and accuracy of risk identification; 2. This invention constructs a closed-loop feedback mechanism of in-process early warning and post-process audit, enabling the system to have the self-optimization ability to become more accurate with use, breaking the situation of the separation between pre-event prevention and post-event audit in the existing technology; 3. This invention innovatively proposes a regional collaborative investigation mechanism, which effectively curbs the fare evasion behavior of criminals who try to evade fees by switching stations and retrying.

[0009] 4. This invention combines vehicle sensor signals with service area video behavior analysis, which can accurately determine whether abnormal cargo movement occurs during the vehicle's stay in the service area, and can finely distinguish between normal watering and preservation operations and cargo swapping cheating behavior, effectively eliminating misjudgments caused by relying solely on weight changes, and filling the gap in existing technology for truck behavior perception on the road. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of the overall functional architecture of the system of the present invention; Figure 2 This is a schematic diagram of the main flow of the method of the present invention; Figure 3 A schematic diagram of the parallel analysis process for multiple risk dimensions in multi-dimensional inspection and early warning during the event; Figure 4 A schematic diagram illustrating the multi-dimensional analysis dimensions for post-event audit analysis at the management end.

[0011] The attached diagram lists the components represented by each number as follows: Figure 1 The modules are as follows: 100 - Data aggregation module; 200 - In-process multi-dimensional inspection and early warning module; 300 - AI-assisted inspection module; 310 - AI image recognition unit; 320 - Trajectory photo backtracking unit; 400 - Rapid inspection assistance module; 500 - Management audit and analysis module; 600 - Closed-loop feedback module; 700 - Existing system optimization module; 800 - Vehicle sensor signal receiving unit; 900 - Service area video analysis unit. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Example 1: System Overall Architecture and Workflow Please see Figure 1 and Figure 2 This invention provides an AI-based audit and inspection system for green channel vehicles based on multi-dimensional feature fusion. The system is installed in the server cluster of the highway operation and management unit and interacts with the toll collection systems and inspection terminals (inspection APP and toll collection assistant software) of each toll station via a dedicated network. The system consists of a data aggregation module 100, a real-time multi-dimensional inspection and early warning module 200, an AI-assisted inspection module 300, a rapid inspection assistance module 400, a management-end audit and analysis module 500, a closed-loop feedback module 600, and an existing system optimization module 700, as well as an on-board sensor signal receiving unit 800 for receiving signals from on-board sensors and a service area video analysis unit 900 for analyzing service area monitoring videos.

[0014] The vehicle-mounted sensor signal receiving unit 800 is deployed at key locations along the highway service area to receive wireless signals from sensors carried by green channel vehicles entering the service area. The sensors can be uniquely identified electronic seals, door magnetic switch sensors, or vibration displacement sensors, which are installed and activated by staff or drivers when the vehicle enters the highway entrance. The preferred installation location is at the closing point of the cargo box door or on the surface of the cargo. The sensors monitor their own status in real time (such as whether the seal is intact, whether the door is open, and whether the cargo has undergone significant displacement), and upload the status data to the in-process multi-dimensional inspection and early warning module 200 through the service area signal receiving unit 800. The service area video analysis unit 900 connects to the existing high-definition surveillance cameras in the service area, and can automatically retrieve the video stream data of the corresponding camera based on the vehicle's stopping position, and use a behavior recognition model to analyze the video images to identify typical behaviors such as human movements, cargo movement, and water spraying.

[0015] The data aggregation module 100 is responsible for collecting multi-dimensional data from the highway toll collection system. Specifically, it obtains vehicle entry weighing data from the overload control system at each toll station, including information such as total vehicle and cargo weight, axle type, and number of axles; it obtains entry and exit transaction data from the toll lane system, including entry station information, exit station information, passage time, toll amount, and vehicle type information; it obtains gantry transaction data and gantry license plate recognition data from the gantry system, including the time when vehicles pass through each gantry, gantry location, and vehicle license plate recognition photos; it obtains historical green channel inspection data from the green channel integrated management platform; and it obtains vehicle registration data from the vehicle management system. In addition, the data aggregation module also periodically collects agricultural product market price data through external data interfaces. Vehicle sensor status data and video analysis results are also incorporated into the data aggregation module, providing a more complete data foundation for in-process multi-dimensional analysis.

[0016] The in-process multi-dimensional inspection and early warning module 200 is the core module of the system that functions at the inspection site. When a green channel vehicle enters the toll station exit lane and triggers the green channel inspection process, the toll system sends the vehicle's license plate information to this system. Based on the license plate information, the in-process multi-dimensional inspection and early warning module 200 retrieves the full multi-dimensional data of the vehicle from the data aggregation module 100 in real time and initiates multiple risk dimension analysis sub-tasks in parallel. Each sub-task is calculated and judged independently. When the analysis result of any risk dimension sub-task triggers the preset early warning conditions, the module generates an early warning message. This early warning message includes the risk dimension identifier, a description of the risk cause, and suggested inspection priorities. The early warning message is pushed to the inspection terminal in real time through a dedicated data channel.

[0017] The AI-assisted inspection module 300 includes two sub-modules: an AI image recognition unit 310 and a trajectory photo backtracking unit 320. The Quick Inspection Assist Module 400 provides two core functions: automatic calculation of the reasonable range of total vehicle and cargo weight and automatic display of vehicle photos. The management-side audit and analysis module 500 performs post-event analysis on the green channel vehicle data that has been inspected from multiple dimensions to uncover abnormal patterns; The closed-loop feedback module 600 is responsible for establishing a data channel between in-process early warning and management-end audit analysis, enabling dynamic optimization of early warning rules; The existing system optimization module 700 optimizes the functions of the existing Green Channel Inspection APP and Green Channel Integrated Management Platform.

[0018] Example 2: Detailed Explanation of the Eleven Risk Dimensions of In-Process Multi-Dimensional Inspection and Early Warning Please see Figure 3In this embodiment, the eleven risk dimensions of the real-time multi-dimensional inspection and early warning module 200 are described in detail one by one; the first ten risk dimensions are consistent with the previously described implementation method, and are briefly restated here, while the eleventh dimension is newly added and described in detail.

[0019] First risk dimension – matching the non-compliance list database: (consistent with the original implementation, retaining the original description); The second risk dimension – regional collaborative investigation: For typical toll evasion behavior, the original toll station inspectors can submit vehicle information on the inspection APP, and the system will synchronize it to all toll station terminals in the jurisdiction in real time; once the vehicle triggers the green channel inspection process at any other toll station in the region, the system will automatically identify it and generate a high-priority alarm. The third risk dimension – inlet and outlet weight fluctuation analysis: The system automatically acquires vehicle inlet and outlet weighing data and calculates the absolute value of the difference between the two; when the difference exceeds a preset threshold (e.g., 2 tons), a weight anomaly warning is generated. The fourth risk dimension – judgment of the rationality of production and sales locations: The system constructs a database of production and sales locations for fresh agricultural products transported through green channels. When the vehicle's entrance and exit areas do not match the production and sales characteristics of the declared goods, an abnormal production and sales location warning is generated. The fifth risk dimension – suspected reselling analysis: The system constructs a standard travel time reference based on historical data, calculates the actual travel time, and generates a suspected reselling warning when the actual travel time exceeds the standard travel time by a certain percentage (e.g., 50%) and there is an abnormally long stay in the service area. The sixth risk dimension – risk analysis of low-value goods over long distances: The system dynamically obtains the price of goods, calculates the theoretical total value of goods based on their quality, compares it with the toll amount, and generates an early warning when the total value is less than 1.5 times the toll amount. The seventh risk dimension – cross-provincial large-amount vehicle warning: When a vehicle has a cross-provincial travel record or a single toll amount exceeds 1,500 yuan, a cross-provincial large-amount warning will be generated. The eighth risk dimension – low-frequency vehicle analysis at toll stations: Statistics on the number of times vehicles pass through the current toll station exit in the past month. If the number of times is less than 5, a low-frequency vehicle warning is generated, and historical high-frequency toll stations and typical cargo types are displayed simultaneously. Ninth Risk Dimension – Low Flatbed Truck Identification and Warning: When a vehicle is identified as a low flatbed semi-trailer and has relevant historical non-compliance records, a low flatbed truck warning is generated. The tenth risk dimension – vehicle and cargo volume and mass analysis: Based on vehicle parameters and cargo density reference, calculate the reasonable range of total vehicle and cargo mass. When the actual total vehicle and cargo mass exceeds the reasonable range by more than 20%, an early warning is generated. Eleventh Risk Dimension – Onboard Cargo Status Monitoring and Early Warning: This dimension is specifically designed to address potential cargo dumping fraud and misjudgment issues related to watering for preservation that may occur when green channel vehicles stop at highway service areas (or emergency parking lanes).

[0020] The specific implementation method is as follows: First, at highway entrance toll stations, when green channel vehicles apply for green channel eligibility, inspectors or drivers need to install or affix a uniquely numbered vehicle-mounted sensor to the critical closing point of the truck bed door, or use an electronic cable tie with sensing function to seal the door; the sensor has a built-in status detection circuit and wireless communication module, which can sense its own physical integrity (whether it has been cut or disassembled), the opening and closing status of the door, and optional internal vibration / displacement status in real time; the sensor records the vehicle's license plate information, and the binding relationship is uploaded to the system when activated; When a vehicle enters and stops at any service area along the highway, the signal receiving unit (such as a LoRa gateway, RFID reader, or Bluetooth beacon) set up in that service area will continuously or periodically receive status data packets broadcast by the vehicle's onboard sensors. Under normal circumstances, the sensor sends a normal status data packet every 30 seconds to 1 minute, which carries the sensor number, the current status code (such as seal intact, door closed) and battery information. The system performs the following logical judgments on the received signals: Scenario A: The sensor signal is normal and the status remains unchanged; the sensor remains online throughout the entire service area stay, and the status code always shows "seal intact" or "door closed"; this indicates that the cargo loading space was not opened during the stay, and the cargo was most likely not touched or swapped; in this case, the system only records the vehicle's stay duration in the service area; this stay duration is compared with the standard travel time for the vehicle's current journey: the standard travel time is obtained by retrieving the median of the actual travel time of the vehicle model in the past three months of valid travel records between the same origin and destination stations, and adding a certain percentage (such as 30%) as a reasonable fluctuation limit; if the total travel time caused by the actual stay duration does not exceed this reasonable limit, no warning is generated; if the total travel time exceeds the reasonable limit, a lower-level abnormal stay reminder is generated, prompting the inspection personnel to note that the vehicle may have abnormal time due to non-cheating reasons (such as long rest), but it is not directly judged as unloading cargo.

[0021] Scenario B: Sensor signal interruption or abnormal status code; If, during the stop at the service area, the system detects that the vehicle's sensor signal is continuously interrupted for more than a preset threshold time (e.g., 3 minutes), or the status code uploaded by the sensor changes to an abnormal code such as damaged seal, opened door, or vibration displacement exceeding the threshold, it indicates that the cargo compartment may have been opened or the sensor itself may have been deliberately damaged; at this time, the system immediately triggers the high-risk warning procedure. Upon triggering a high-risk warning, the system automatically executes the following linked operations: First, based on the vehicle's parking location information within the service area (which can be determined through sensor signal strength triangulation or by combining license plate recognition from service area checkpoint cameras), the system retrieves video recordings from all high-definition surveillance cameras covering the area during the corresponding time period. Second, the service area video analysis unit 900 uses a pre-set behavior analysis model to intelligently analyze the video footage. This behavior analysis model, trained on a large number of samples, can identify and classify the following key operational behaviors: Watering operation – The footage shows water pipes or buckets spraying liquid into the carriage; Cargo handling operations – The footage shows people moving goods out of the carriage or moving goods from outside into the carriage; (d) No operation – The scene shows the simultaneous movement of goods with different packaging and appearances being moved out and in; The third step involves the system generating a warning data package based on the identified operational behavior category (such as suspected cargo exchange), confidence score, and a 5-10 second key video clip. This data package is then pushed to the highway management personnel's audit terminal APP and the toll station inspection terminal that will be inspected. The warning package clearly indicates that the vehicle sensor is abnormal: the door is open, and the video analysis determines it to be [cargo exchange]. Please conduct a thorough on-site inspection.

[0022] When the vehicle arrives at the exit toll station and enters the green channel inspection process, the inspection personnel's terminal will automatically display an alert, and they can directly click to play the abnormal video clip. The inspection personnel, combining the alert reason pushed by the system with the video evidence, will conduct a targeted manual review of the goods inside the vehicle. For example, if the video shows watering, the inspection personnel can focus on checking whether the freshness of the goods matches the watering behavior and whether there is any attempt to increase weight with water. If the video shows cargo handling and exchange, the inspection personnel can verify whether the existing cargo types are consistent with the declared types at the entrance. In this way, this dimension can not only accurately identify traditional cargo-switching fraud, but also effectively distinguish between the potentially reasonable agricultural operation of watering for freshness and genuine cargo-switching fraud, avoiding misjudgments based solely on increased weight at the entrance and exit, and significantly improving the accuracy of the judgment.

[0023] In addition, to prevent false alarms caused by non-cheating reasons such as sensor power depletion or signal obstruction, the system has an adaptive fault tolerance mechanism: when the signal interruption duration does not exceed the set buffer time (such as 5 minutes), and the normal status signal is received again before the vehicle leaves the service area after the interruption, the system does not judge it as abnormal, but only records it as a communication interruption; only when there is a long interruption or a clear abnormal status code appears will video linkage analysis be activated to ensure the credibility of the warning.

[0024] Example 3: Detailed Explanation of Five Analytical Dimensions for Post-Event Audit Analysis at the Management End Please see Figure 4 This embodiment provides a detailed description of the five core analysis dimensions of the management audit analysis module 500 of the present invention. The content of each dimension is consistent with the original implementation method and will not be repeated here. These dimensions include: the first analysis dimension—analysis of changes in vehicle types before and after exiting the vehicle; the second analysis dimension—analysis of the multi-dimensional distribution pattern of non-compliant green channel vehicles; the third analysis dimension—analysis of green channel data at toll stations; the fourth analysis dimension—analysis of green channel data of inspection personnel; and the fifth analysis dimension—dynamic management of the key inspection list database.

[0025] Experimental content and experimental data To verify the practical effect of the technical solution of this invention, a three-month system setup and application experiment was conducted on a section of highway under the jurisdiction of a provincial expressway company. This section covers an operating mileage of approximately 1,200 kilometers, including 68 toll stations, with an average daily green channel vehicle traffic volume of approximately 1,100 vehicles. The experimental environment included two application servers and one database server located in the company's data center, as well as a total of 136 mobile terminals with the verification APP installed at each toll station and 68 sets of toll collection assistant desktop terminals. In addition, vehicle-mounted sensor signal receiving devices were deployed in 15 key service areas along the route, and the video analysis service in the corresponding areas was upgraded.

[0026] Experiment 1: Experiment to Improve Inspection Efficiency Experimental Design and Data Acquisition The experiment was divided into two phases: the first phase (one month before system setup) collected baseline data, and the inspectors carried out green channel inspections according to the original workflow, recording the time taken for each inspection process; the second phase (three months after system setup) used the system of this invention to assist in inspections, and similarly recorded the time taken for the entire process; 32,146 and 98,475 valid inspection records were collected in the two phases, respectively.

[0027] 1.2 Experimental Results Before the system was set up, the average time for a single vehicle to complete the entire inspection process was 289 seconds (approximately 4 minutes and 49 seconds). After the system was set up, the average time for a single vehicle to complete the entire inspection process was reduced to 226 seconds (approximately 3 minutes and 46 seconds), improving inspection efficiency by 21.8%, achieving the project's preset goal of improving green channel inspection efficiency by more than 10%. Specific data on the changes in time consumption at each stage are as described above and are omitted here.

[0028] Data Analysis The main contribution to the improved inspection efficiency comes from the replacement of manual operations by information technology and the more targeted inspection by the early warning mechanism; this shows that the rapid inspection assistance module and AI-assisted inspection module of the present invention have played a significant role in practical applications.

[0029] Experiment 2: Experiment to Improve Inspection Accuracy Experimental Design and Data Acquisition One month before the system was set up, statistics were compiled on complaints and disputes involving drivers and passengers reported by each toll station due to the compliance of goods. Three months after the system was set up, similar complaints and disputes were continuously compiled. At the same time, inspection records from both phases were sampled and reviewed (2,000 records were randomly selected from each phase), and senior auditors independently judged and evaluated the accuracy of the inspection conclusions.

[0030] Experimental results Regarding complaints and disputes: There were 47 cases in the month before the system was implemented; in the three months after the system was implemented, the average monthly number of complaints and disputes dropped to 18, a decrease of 61.7%. The sampling review doubt rate dropped from 4.45% to 1.85%, a decrease of 58.4%.

[0031] Data Analysis The significant decrease in the number of complaints and disputes is attributed to the fact that AI-assisted cargo identification and multi-dimensional early warning information provide a scientific basis for inspection and judgment, making the inspection conclusions more convincing. The decrease in the doubt rate of sampling review verifies the positive role of system-assisted inspection in improving the accuracy of inspection conclusions.

[0032] Experiment 3: Verification Experiment on Increased Revenue Experimental Design and Data Acquisition The method of comparison before and after was adopted. The total amount of green channel toll exemption for the entire road segment in the previous 12 months was set as the baseline value, and the amount of green channel toll exemption in the following 3 months (converted to annualized data) was set as the comparison value.

[0033] Experimental results The base year's reduction amount was 948 million yuan; after the system was set up, the total reduction amount for the three months was 230.3 million yuan, with an annualized reduction amount of 921.2 million yuan. The annualized reduction amount decreased by 26.8 million yuan, which means an annual increase in revenue of about 26.8 million yuan, far exceeding the preset target.

[0034] Analysis of the contribution of the early warning mechanism: Within three months, the multi-dimensional inspection and early warning module triggered 38,462 early warnings, detecting 10,010 non-compliance issues, with an overall non-compliance hit rate of 26.0%. Detailed contribution data for each dimension are shown in the table below:

[0035] As can be seen from the newly added data on vehicle-mounted cargo status monitoring and early warning in the table above, this dimension triggered 3126 warnings during the experiment. Among them, 1824 were confirmed on-site to be due to actual cargo anomalies (including unexplained cargo dumping, smuggling, and unexplained abnormal opening of the cargo compartment), resulting in a non-compliance hit rate as high as 58.3%. This is second only to low-flatbed truck warnings (83.1%) among all dimensions. However, due to the smaller trigger base of low-flatbed truck warnings, the absolute number of cases detected by vehicle-mounted cargo status monitoring and early warning is more prominent. Of particular note is that among the 1824 confirmed non-compliance cases, There were 427 instances of sophisticated cheating involving adding weight by watering or exchanging small amounts of high-value non-green channel goods. These behaviors would be extremely difficult to detect using only traditional weight fluctuation analysis (the third dimension) or time anomaly analysis (the fifth dimension). For example, in a typical case of watering cheating, a vehicle added approximately 1.8 tons of weight to its cargo by watering it at a service area. The difference between the exit and entrance weights was 1.9 tons. Under the original 2-ton threshold, this would not have triggered a weight warning. However, the vehicle's sensors recorded the door opening, and video analysis captured the watering action lasting 20 minutes. The system accurately identified and triggered an alarm, preventing missed detection. The addition of this dimension resulted in a net increase of 1,824 non-compliant alerts (some overlap with traditional dimensions has been deduplicated), directly contributing approximately 4.1 million yuan in additional toll revenue (over three months), demonstrating the effectiveness and originality of the new solution.

[0036] Experiment 4: Verification Experiment on the Anti-False Judgment Effect of Onboard Cargo Status Monitoring and Early Warning Experimental Design and Data Acquisition To independently verify the performance of the eleventh risk dimension, particularly its accuracy in distinguishing between normal watering and preservation methods and cheating, a controlled experiment was designed. Three hundred volunteer green channel vehicles were selected, sensors were installed in their cargo compartments, and their consent was obtained to conduct simulated behavior tests in service areas. The simulated behaviors were divided into four groups: Normal parking and rest, without opening the compartment door or moving the goods (100 times). Simulates normal watering, cooling and preservation of fruits and vegetables by opening the door and watering (80 times). Simulate the act of transferring goods to cheat, by moving out some of the original goods and putting them into similar-looking packaging boxes (80 times). (D) Open the compartment door for cleaning and other non-cargo changes (40 times); all simulated behaviors were carried out in the selected 3 service areas and under system operation; at the same time, as a control, the traditional judgment rules based solely on weight difference (threshold 2 tons) and time anomaly (50% over standard travel) were used to make early warning judgments for the same batch of simulated events; the number of correct warnings, false alarms and missed warnings of the two methods were recorded.

[0037] Experimental results The experimental data are summarized in the table below:

[0038] Data Analysis The experimental results clearly show that traditional methods relying solely on weight and time are severely inadequate when dealing with complex service area behaviors: For Group B's watering and preservation activities, the false negative rate was as high as 81.3% (65 / 80), because the weight gain from these watering activities often did not reach the fixed threshold of 2 tons; for Group C's unloading and cheating activities, the false negative rate was as high as 85.0% (68 / 80), because cheaters often control the loading and unloading volume to avoid significant weight fluctuations and control the dwell time; simultaneously, the traditional method generated 8 false alarms for Group A's normal rest. The method of this invention directly detects the sensor status... By sensing the opening of the cargo door and combining it with video behavior analysis, the system achieved 100% accurate identification and differentiation of watering behavior in Group B and unloading behavior in Group C, with zero false alarms for normal rest scenarios. The only two missed alarms occurred in the cleaning and tidying scenario in Group D, where the person was in the camera's blind spot and the system could not obtain effective video footage, but the abnormal sensor signal was recorded, and the system still triggered a low-level alert. The overall accuracy rate reached 99.3%, which fully demonstrates that the newly added dimension of vehicle cargo status monitoring and early warning in this invention has extremely high practical value and creatively solves the industry pain point of misjudging watering and preservation.

[0039] This invention integrates and analyzes multi-dimensional data from green channel inspections, entrance overload weighing data, entry and exit transaction data, gantry transaction data, gantry license plate recognition data, vehicle registration data, agricultural product market price data, and newly added vehicle sensor status data and service area video analysis data. Combined with AI's large-scale model for cargo identification, it provides on-site inspection reminders, rapid inspections, assisted inspections, and management-end audit analysis. Notably, it innovatively incorporates key aspects such as precise perception and judgment of truck service area behavior, constructing a complete multi-dimensional audit and inspection method and system for green channel vehicles. After effective application, the system has not only significantly improved inspection efficiency and reduced disputes and complaints, resulting in substantial revenue increases, but also overcome the long-standing industry challenges of monitoring the status of goods en route and misjudging watering and preservation issues. This embodies the comprehensive management goals of ensuring safety and smooth traffic flow, maximizing cargo collection, and improving quality and reducing costs.

[0040] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A green channel vehicle AI audit and inspection method based on multi-dimensional feature fusion, characterized in that, Including the following: S1, collect multi-dimensional data from the highway toll system. The multi-dimensional data includes at least green channel inspection data, overload control weighing data, transaction flow data, gantry identification data, vehicle parameter data, and agricultural product price data. S2, when a green channel vehicle enters the exit lane, it retrieves the current multi-dimensional data of the vehicle in real time and analyzes it in parallel from multiple risk dimensions, including at least the monitoring and early warning of the status of the cargo on board. When any dimension triggers the early warning condition, it generates an early warning message with risk dimension identifier and cause description and pushes it to the inspection terminal. S3, During the inspection process, the inspection terminal collects images of the goods and uploads them to the AI ​​recognition module for goods identification and catalog matching. It also issues a prompt when the goods do not conform to the green channel catalog. At the same time, the inspection terminal retrieves photos of the vehicle's journey for the inspection personnel to compare. S4. Perform multi-dimensional post-inspection audit analysis on the vehicle data that has been inspected. The analysis includes at least vehicle type change analysis, non-compliance distribution pattern analysis and inspection personnel indicator difference analysis. When abnormal indicator data is found, an alarm is triggered. S5 feeds back the audit results generated in S4 to the early warning rules in S2, dynamically adjusting the early warning thresholds and weights for each risk dimension.

2. The AI-based auditing and inspection method for green channel vehicles based on multi-dimensional feature fusion as described in claim 1, characterized in that, The multiple risk dimensions in S2 include: non-compliance list database matching dimension, regional collaborative investigation dimension, import and export weight fluctuation analysis dimension, production and sales location rationality judgment dimension, suspected reselling analysis dimension, long-distance low-value cargo risk analysis dimension, cross-provincial large amount early warning dimension, toll station low-frequency vehicle analysis dimension, low flatbed truck identification early warning dimension, vehicle and cargo volume and quality analysis dimension, and vehicle cargo status monitoring early warning dimension.

3. The AI-based auditing and inspection method for green channel vehicles based on multi-dimensional feature fusion according to claim 2, characterized in that, The specific dimensions of the vehicle-mounted cargo status monitoring and early warning system include: installing sensors in the rear cargo bed of green channel vehicles when they enter the highway entrance; these sensors monitor the opening and closing status of the cargo doors or the displacement status of the cargo; if the sensor signal is normal and the status remains unchanged while the vehicle is stopped at a service area, only the duration of the stop is recorded and an anomaly is judged; if the sensor signal is interrupted or an abnormal status signal is issued, the service area monitoring video is automatically retrieved for behavior analysis to identify the category of the operation behavior, which includes at least watering operations, cargo handling operations, and cargo exchange operations, and the analysis results and corresponding video clips are generated and pushed to the management personnel terminal and the inspection terminal.

4. The AI-based auditing and inspection method for green channel vehicles based on multi-dimensional feature fusion according to claim 1 or 2, characterized in that, The warning conditions for the multiple risk dimensions include: the difference between the inbound and outbound weights exceeds a preset weight threshold; the proportion of actual travel time exceeding the standard travel time exceeds a preset proportion threshold and the vehicle has an abnormal stop record in the service area; the total value of the goods is less than a preset multiple of the toll amount; the number of exit passages at the current toll station in the past month is less than a preset frequency threshold; and the actual total weight of the vehicle and goods is less than the lower limit of the reasonable range or more than the upper limit of the reasonable range by a preset percentage.

5. The AI-based audit and inspection method for green channel vehicles based on multi-dimensional feature fusion according to claim 1, characterized in that, The specific method of the regional collaborative investigation dimension in S2 is as follows: For vehicles that fail the green channel inspection and are suspected of toll evasion, the original toll station inspection personnel submit vehicle information including license plate number, reason for non-compliance and contact information. The system synchronizes this information to all toll station terminals in the jurisdiction in real time. When the vehicle triggers the green channel inspection process at other toll stations in the region, the system automatically identifies and generates a high-priority alarm. The alarm information includes the risk reason and contact information issued by the original toll station.

6. The AI-based audit and inspection method for green channel vehicles based on multi-dimensional feature fusion according to claim 1, characterized in that, The reasonable range of total vehicle and cargo mass in the S2 volumetric mass analysis dimension is calculated as follows: based on the total mass, curb weight and rated load capacity parameters recorded on the vehicle registration certificate, the sum of the curb weight and rated load capacity multiplied by the first coefficient is used as the reasonable lower limit of total vehicle and cargo mass, and the result of the total mass multiplied by the second coefficient is used as the reasonable upper limit of total vehicle and cargo mass, where the first coefficient is less than 1 and the second coefficient is greater than 1.

7. The AI-based auditing and inspection method for green channel vehicles based on multi-dimensional feature fusion according to claim 1, characterized in that, The S3 also includes: the inspection terminal automatically calculates and displays the reasonable range of total vehicle and cargo weight based on the vehicle registration certificate parameters, and automatically retrieves and displays the vehicle registration certificate photos, front photos, side photos and rear photos of the vehicle stored in the system within one week.

8. The AI-based audit and inspection method for green channel vehicles based on multi-dimensional feature fusion according to claim 1, characterized in that, The dynamic management of the key inspection list in S4 specifically involves: comprehensively considering the vehicle's historical violation frequency, violation severity, and recent inspection result trends, and adopting a hierarchical and classified management strategy to divide vehicles into different risk levels; supporting administrators to manually add or remove vehicles from the list and record operation logs; the list is synchronized in real time to S2 as a data source for the non-compliance list matching dimension.

9. A green channel vehicle AI audit and inspection system based on multi-dimensional feature fusion, characterized in that, include: The data aggregation module is used to collect and integrate multidimensional data, including green channel inspection data, overload control weighing data, transaction flow data, gantry identification data, vehicle parameter data, and agricultural product price data. The real-time multi-dimensional inspection and early warning module is connected to the data aggregation module and is used to retrieve the current multi-dimensional data of the vehicle in real time, analyze it in parallel from multiple risk dimensions including the monitoring and early warning of the status of the on-board cargo, and generate early warning information to be pushed to the inspection terminal; the real-time multi-dimensional inspection and early warning module is also connected to the on-board sensor signal receiving unit and the service area video analysis unit. The AI-assisted inspection module includes an AI image recognition unit and a trajectory photo tracing unit; The management audit and analysis module is used to perform multi-dimensional post-event analysis on the inspected vehicle data, including changes in vehicle type, non-compliance distribution patterns, and differences in inspector indicators, and to trigger anomaly alarms. The closed-loop feedback module is connected to the in-process multi-dimensional inspection and early warning module and the management audit and analysis module, respectively, and is used to feed back the audit results to the early warning rules and dynamically adjust the early warning thresholds and weights of each risk dimension.

10. The AI-based audit and inspection system for green channel vehicles based on multi-dimensional feature fusion as described in claim 9, characterized in that, It also includes a rapid inspection assistance module and an existing system optimization module; the rapid inspection assistance module is used to automatically calculate the reasonable range of total vehicle and cargo weight based on the vehicle registration certificate parameters and display it on the inspection terminal, as well as to automatically retrieve and display recent vehicle photos stored in the system; the existing system optimization module is used to realize real-time synchronization of inspection data, automatic reminder of data to be inspected, one-click login, and optimization of the operation interface.

Images