Cloud edge end collaborative internet of things toll station data management method, system and device

Abstract

The present application belongs to the field of computer vision, and provides a toll station data management method, system and device based on cloud edge-end collaborative Internet of Things, which is based on a sliding time window, combined with a computer vision model, to collect multi-source data generated when a vehicle passes through a toll station in real time, generate a transaction block, form an edge-end deviation rate vector according to the multi-source data, construct a spiral deviation coordinate vector based on the edge-end deviation rate vector, calculate an edge-end spiral descent inner product through the edge-end deviation rate vector and the spiral deviation coordinate vector, determine a compliant transaction block according to the edge-end spiral descent inner product, encrypt the compliant transaction block, and dynamically adjust an upload strategy according to a link bandwidth state. The present application can improve the sensitivity of low-intensity coupling type anomaly detection, reduce invalid data transmission through dynamic batch uploading of compliant data, and support high-concurrency vehicle flow.

Description

Technical Field

[0001] This invention pertains to computer vision, intelligent transportation, and Internet of Things (IoT) technologies, specifically relating to a cloud-edge-device collaborative IoT data management method, system, and equipment for toll stations. Background Technology

[0002] With the widespread deployment of multi-source sensors such as ETC, WIM (Wireless In-Mortar Mechanism), high-definition video, high-definition license plates, and environmental monitoring in toll lanes, a single toll station generates millions of heterogeneous data points daily. In traditional end-to-cloud direct transmission or two-tier end-edge-cloud solutions, it is difficult to synchronize images, transactions, and logs generated by multiple lanes in parallel, leading to high error rates in backend auditing. The data link lacks hierarchical encryption and anonymization, posing significant risks to privacy and financial information exposure. Furthermore, the fragmented nature of terminal software versions results in high on-site maintenance costs.

[0003] With the rapid development of intelligent transportation and IoT technologies, modern toll station systems based on a cloud-edge-device collaborative architecture have become a future trend. These systems deploy edge computing nodes in lanes, such as lane edge gateways and station-level edge servers, to collect and preprocess vehicle traffic data in real time, including license plate recognition images, ETC transaction status, vehicle type, vehicle model, dynamic axle configuration, real-time speed, number of seats, and vehicle height. Combined with the powerful storage and intelligent analysis capabilities of the cloud, these systems aim to improve traffic efficiency, enhance security monitoring, and optimize operational management. However, in the core application scenario of fraud detection, existing technologies still face many serious challenges, especially when dealing with low-intensity, asynchronous, and multi-dimensional sparsely coupled toll evasion behaviors, exhibiting significant detection blind spots and performance bottlenecks.

[0004] Most existing technologies rely on setting fixed thresholds for individual features for discrimination, such as issuing an alarm upon successful matching of the ETC status list or an overweight alarm based on axle deviation rate. While effective for occasional high-intensity single anomalies, these methods struggle to address cunning, low-intensity, combined toll evasion strategies. For example, attackers might use a fake ETC tag while strictly controlling vehicle speed and weight within thresholds, or drive a vehicle with a fake license plate and moderate overloading; however, individual feature detections alone are insufficient to trigger an alarm. This piecemeal approach to toll evasion is increasingly rampant, resulting in numerous successful, undetected frauds. More complex linear weighted fusion schemes offer slight improvements, but due to heterogeneous feature dimensions, significant differences in sensitivity, and a lack of modeling capabilities for nonlinear synergistic effects, they still struggle to accurately capture weak coupling signals with low false alarm rates. For example, patent document CN115049597A provides an image anomaly detection method and system based on sparse completion networks. It demonstrates that traditional image reconstruction models, such as autoencoders, require complete pixel input. When images contain occlusions or sparse regions, such as parts occlusions in industrial inspection, the reconstruction error increases significantly, leading to a higher false alarm rate. The solution uses sparse convolution to process only valid pixels, thus verifying the inadequacy of traditional methods for handling data sparsity.

[0005] Some improvement schemes attempt to introduce time series analysis, such as Dynamic Time Warping (DTW) or clustering algorithms to cluster vehicle features within a continuous window, in order to discover asynchronous low-intensity anomalous combinations. However, the imprecise alignment of anomalous factors, such as ETC anomalies occurring 0.5 seconds before the barrier lift and peak vehicle speed fluctuations occurring 2 seconds later, makes methods relying on fixed time series templates computationally intensive and ineffective.

[0006] Performing multi-dimensional clustering in real time at edge nodes, such as fusing license plates, speed, and axis shape, or training lightweight neural networks like multilayer perceptrons (MLPs) for feature fusion, puts enormous pressure on the limited computing power and memory resources at the edge. For example, during peak traffic hours, with millisecond-level latency requirements, traditional methods are often forced to sacrifice model complexity or reduce sampling frequency, making it difficult to balance real-time requirements with detection accuracy. For instance, patent document CN107391443B provides a sparse data anomaly detection method and device that ignores implicit relationships between attributes and relies on distance and pattern matching to only statistically analyze explicit associations, failing to capture low-strength coupling features. Furthermore, its distance-based method has too high a time complexity to handle the high-concurrency data streams at toll stations. Applying ETC and critical speeding detection requires capturing non-explicit associations, which traditional statistical methods cannot achieve.

[0007] The most critical threat that current technologies have long overlooked stems from sparsely coupled anomalies. These are hidden behaviors caused by multiple low-intensity, infrequent, and asynchronously triggered anomalous factors that combine randomly within a short time window, collectively inducing fraud, yet none of these individual triggers exceed thresholds. This type of behavior is prevalent and highly damaging, accounting for the majority of ETC fraud losses in the industry, yet targeted solutions have long been lacking. The root cause lies in the current technology's lack of a deep theoretical understanding of the coupling mechanism of heterogeneous spatiotemporal data, and the absence of an efficient algorithmic framework at the engineering level that simultaneously satisfies high-dimensional heterogeneous fusion, asynchronous signal integration, sensitivity to critical states, and low-power edge computing. Traditional solutions either completely miss these anomalies, treating them as noise, or rely on cloud-based backtracking analysis, resulting in severe response delays of hours or even days, failing to meet the requirements of proactive real-time defense.

[0008] To ensure detection coverage, some cloud-based centralized processing solutions upload all raw data or initially uncertain data to the cloud for in-depth analysis. This not only consumes massive amounts of uplink bandwidth, especially for high-definition video data, leading to delays or loss of critical information during link congestion, but also significantly increases cloud storage and computing costs. According to actual project calculations, the annual cost of unoptimized data transmission at a toll station with a daily traffic flow of 100,000 vehicles could reach millions of yuan. While existing cloud-edge-device collaborative intelligent toll station management technologies have achieved decentralized computing power and preliminary preprocessing capabilities in their architecture, their core real-time fraud detection accuracy is limited, resulting in high false negatives, especially when dealing with low-intensity, asynchronous, sparsely coupled anomalies that are frequent. High response latency leads to partial reliance on the cloud, resulting in heavy computing load and high operating costs. Summary of the Invention

[0009] The purpose of this invention is to propose a cloud-edge-device collaborative IoT data management method, system, and device for toll stations, in order to solve one or more technical problems existing in the prior art, and at least provide a beneficial option or create conditions.

[0010] To achieve the above objectives, according to one aspect of the present invention, a cloud-edge-device collaborative IoT toll station data management method is provided, the method comprising the following steps:

[0011] Based on the cloud-edge-device collaborative IoT toll station data management method, a four-level collaborative architecture is constructed, consisting of terminals, lane edge gateways, station-level edge servers, and a cloud platform. The method specifically includes the following steps:

[0012] At the terminal layer, computer vision models are used to collect multi-source data generated when vehicles pass through toll stations in real time. The multi-source data may include, but is not limited to, data on license plates, ETC transactions, vehicle speed, vehicle height and timestamps, as well as data on vehicle type, vehicle class, axle load / axle type, etc.

[0013] At the lane edge gateway, the collected data is time-aligned, format-unified, and aggregated to generate transaction blocks and cache them locally.

[0014] On the station-level edge server: Based on the sliding time window, an edge deviation rate vector is formed according to the multi-source data. The edge deviation rate vector includes F1 specific existence surge, F2 high-frequency license plate counterfeit points, F3 axle deviation rate, F4 vehicle speed deviation rate, and F5 vehicle height deviation rate.

[0015] Based on the edge deviation rate vector, a spiral offset coordinate vector is constructed, which includes w1 specific stock spiral offset coordinate, w2 high frequency spiral offset coordinate, w3 shaft type spiral offset coordinate, w4 vehicle speed spiral offset coordinate, and w5 vehicle height spiral offset coordinate.

[0016] The edge spiral descent inner product S is calculated by the edge deviation rate vector and the spiral offset coordinate vector.

[0017] Based on the edge spiral descent inner product S, the following judgments are made in sequence: calculate the edge spiral descent mapping value A, calculate the edge spiral descent reversal value B; make a judgment based on the edge spiral descent mapping value A and the edge spiral descent reversal value B. If the transaction block is marked as abnormal, the upload is temporarily suspended; otherwise, it is marked as a compliant transaction.

[0018] Compliant transaction blocks are encrypted between the station-level edge server and the cloud platform, and the upload strategy is dynamically adjusted based on the link bandwidth status between the station-level edge server and the cloud platform. Furthermore, in some embodiments, if the link bandwidth exceeds a set threshold, uploads can proceed immediately.

[0019] Furthermore, the data in the transaction block also includes a unified lane number, a BeiDou satellite timestamp, and multi-source data fields, and the lane edge gateway supports a zero-copy caching mechanism.

[0020] Furthermore, the existence of the F1-specific protrusion is determined by querying the status list database maintained in the cloud based on the license plate number identified by the terminal. This status list supports daily synchronization updates.

[0021] Furthermore, the F1 specific protrusion quantity is generated by comparing the vehicle's license plate recognition result with the license plate database to generate a signal value indicating that the license plate status is abnormal.

[0022] The F2 high-frequency license plate counter is a signal value generated based on the vehicle identification results captured by the same monitoring point within a set time period, for the same license plate number that appears repeatedly.

[0023] The F3 axle type deviation rate is obtained by acquiring vehicle dynamic weighing data and vehicle model clustering information based on image visual features, calculating and outputting the deviation rate value of the vehicle axle type relative to the average axle type of its respective vehicle model cluster.

[0024] The F4 vehicle speed deviation rate is obtained by acquiring dynamic speed measurement data of vehicles and their vehicle model clustering information based on visual features of images, and calculating and outputting the deviation rate value of the vehicle speed relative to the average speed of its respective vehicle model cluster.

[0025] The F5 vehicle height deviation rate is calculated and output by obtaining vehicle body height data and vehicle model clustering information based on the visual features of the images through image recognition, and then outputting the deviation rate value of the vehicle body height relative to the average body height of its respective vehicle model cluster.

[0026] Furthermore, the method for calculating the spiral descending inner product of the S-side is as follows:

[0027] The array formed by sequentially arranging the values ​​of the shaft-type spiral offset coordinate, the vehicle speed spiral offset coordinate, the vehicle height spiral offset coordinate, the special stock spiral offset coordinate, and the high-frequency spiral offset coordinate is the spiral offset coordinate vector;

[0028] The array formed by sequentially combining the values ​​of the F3 axle deviation rate, the F4 vehicle speed deviation rate, the F5 vehicle height deviation rate, the F1 special protrusion, and the F2 high-frequency license plate counterfeit points is the edge deviation rate vector.

[0029] The inner product of the spiral offset coordinate vector and the edge deviation rate vector is taken as the S-edge spiral descent inner product.

[0030] Furthermore, in the spiral partial coordinate vector:

[0031] The axle type helical offset coordinate w3 is a reference radial scale that can be used to calibrate the axle type deviation of a vehicle in a two-dimensional helical coordinate curve. Its value is the reciprocal of pi. The reciprocal of pi is used as the w3 axle type helical offset coordinate.

[0032] The vehicle speed spiral offset coordinate w4 is an angular displacement that can be used to reflect the vehicle speed deviation within the same spiral coordinate system. Its value is the product of w3 and the vehicle speed deviation rate. The ratio obtained by comparing the axial spiral offset coordinate with the value of pi is the vehicle speed spiral offset coordinate.

[0033] The vehicle height spiral offset coordinate w5 is the remaining radial distance that can be used to compensate for the joint projection of axle type and vehicle speed, and its value is 1 minus w4; the difference obtained by subtracting the vehicle speed spiral offset coordinate from an integer is the vehicle height spiral offset coordinate.

[0034] The unique stock spiral offset coordinate w1 is the distribution density of rare abnormal signals in the spiral coordinate system that can be used for comprehensive quantification, and its value is the product of w3 and w4; the unique stock spiral offset coordinate is the product of the vehicle speed spiral offset coordinate and the shaft type spiral offset coordinate.

[0035] The high-frequency spiral offset coordinate w2 is a reference scale that can be used to characterize frequent abnormal signals, and its value is 1 minus w1; the difference obtained by subtracting the specific stock spiral offset coordinate from the integer is the high-frequency spiral offset coordinate.

[0036] Furthermore, the process involves using the edge spiral descent mapping value A combined with the edge spiral descent reversal value B to determine whether a transaction block is marked as abnormal and its upload is temporarily suspended; otherwise, it is marked as a compliant transaction. Specifically, this includes:

[0037] The edge spiral descent mapping is defined as the value of the edge spiral descent mapping, which is half the value of the residual of the cosine mapping of the S edge spiral descent inner product.

[0038] The value of the edge spiral descent reversal is the ratio obtained by comparing the hyperbolic tangent function of the S edge spiral descent inner product with the hyperbolic tangent function of the value of pi.

[0039] In some embodiments, it can also be determined whether the product of the edge spiral descent mapping and the edge spiral descent reversal is greater than one-half. If it is greater than one-half, it is determined that there is an anomaly, and the data is not immediately uploaded to the cloud, but cached locally first. If it is not greater than one-half, it is determined that it is compliant, and the compliant transaction data blocks are uploaded to the cloud platform in batches.

[0040] Furthermore, when a transaction block is marked as abnormal, the system caches it together with at least 3 seconds of video footage before and after it in a local database, and uploads it to the cloud platform after bandwidth is restored.

[0041] Furthermore, the cloud-edge-device collaborative IoT toll station data management system runs on any computing device, such as a desktop computer, laptop computer, or cloud data center. The computing device includes a processor, a memory, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps in the cloud-edge-device collaborative IoT toll station data management method. The system includes the following system units:

[0042] The terminal data acquisition module is used to collect multi-source information during vehicle passage in real time, including license plate, ETC transaction, vehicle speed, axle type, vehicle height and timestamp, etc.

[0043] The lane edge gateway module is used to uniformly format and time-align the above information, and then encapsulate it into a transaction block for caching.

[0044] The station-level edge server module is used to extract feature values ​​from transaction blocks, construct edge deviation rate vector and spiral offset coordinate vector, and calculate their S-edge spiral descent inner product to determine whether the transaction is abnormal.

[0045] The cloud platform module is used to receive uploaded compliant transaction blocks, perform data storage, settlement processing, and rule model update and distribution.

[0046] The anomaly detection submodule is used to calculate the edge spiral descent mapping and inversion, and to determine whether their product exceeds the threshold.

[0047] The synchronous scheduling submodule is used to dynamically adjust the upload window and frequency based on the link bandwidth status.

[0048] Correspondingly, the present invention also provides an electronic device, a readable storage medium, and a computer program product:

[0049] An electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the cloud-edge-device collaborative IoT toll station data management method and the steps thereof.

[0050] A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the cloud-edge-device collaborative IoT toll station data management method and the steps thereof.

[0051] A computer program product includes a computer program that, when executed by a processor, implements the cloud-edge-device collaborative IoT toll station data management method and the methods for each step therein.

[0052] The beneficial effects of this invention are as follows: This invention provides a cloud-edge-device collaborative IoT toll station data management method, system, and device. Combining a computer vision model, it collects multi-source data generated when vehicles pass through toll stations in real time to generate transaction blocks. Based on the multi-source data, an edge-end deviation rate vector is formed. A spiral offset coordinate vector is constructed based on the edge-end deviation rate vector. The edge-end spiral descent inner product is calculated by combining the edge-end deviation rate vector and the spiral offset coordinate vector. Compliant transaction blocks are determined based on the edge-end spiral descent inner product. Compliant transaction blocks are encrypted, and the upload strategy is dynamically adjusted according to the link bandwidth status. This improves the sensitivity of low-strength coupling anomaly detection, reduces invalid data transmission through dynamic batch uploading of compliant data, and supports high-concurrency traffic flow. Attached Figure Description

[0053] The above and other features of the present invention will become more apparent from the detailed description of the embodiments shown in conjunction with the accompanying drawings. In the accompanying drawings, the same reference numerals denote the same or similar elements. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without any creative effort. In the drawings:

[0054] Figure 1 The diagram shows a flowchart of a cloud-edge-device collaborative IoT toll station data management method.

[0055] Figure 2 The diagram shows the system architecture of the cloud-edge-device collaborative IoT toll station data management system. Detailed Implementation

[0056] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with the embodiments and accompanying drawings, so as to fully understand the purpose, solution, and effects of the present invention. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0057] In the description of this invention, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0058] like Figure 1 The diagram shown is a flowchart of the cloud-edge-device collaborative IoT toll station data management method according to the present invention. The following is a summary of the method. Figure 1 This paper describes a cloud-edge-device collaborative IoT data management method, system, and device for toll stations based on embodiments of the present invention.

[0059] This invention proposes a cloud-edge-device collaborative IoT toll station data management method, which constructs a four-level collaborative architecture consisting of a terminal, a lane edge gateway, a station-level edge server, and a cloud platform. The method specifically includes the following steps:

[0060] At the terminal layer, computer vision models are used to collect multi-source data generated when vehicles pass through toll stations in real time.

[0061] At the lane edge gateway, the collected data is time-aligned, format-unified, and aggregated to generate transaction blocks and cache them locally.

[0062] At the station-level edge server: Based on the sliding time window, an edge deviation rate vector is formed according to the multi-source data; a spiral offset coordinate vector is constructed based on the edge deviation rate vector; the edge spiral descent inner product S is calculated by the edge deviation rate vector and the spiral offset coordinate vector.

[0063] Based on the inner product S of the spiral descent at the edge, compliant transaction blocks are determined sequentially.

[0064] The compliant transaction blocks are encrypted between the station-level edge server and the cloud platform, and the upload strategy is dynamically adjusted according to the link bandwidth status between the station-level edge server and the cloud platform.

[0065] In some embodiments, taking a highway toll station as an example, a four-level collaborative architecture is constructed: terminal-lane edge gateway-station-cloud platform. When a suspicious truck passes through, the camera (terminal) captures the license plate and identifies and records the axle shape; the lane edge gateway completes data packaging within 1 second; the station-level server calculates the spiral vector in 3 seconds, detecting abnormal combinations of license plate cloning and overloading; the cloud (server cluster) ultimately confirms that this is a repeat offender vehicle and updates the status list. Within 0.1 seconds, the terminal layer captures the license plate number through the camera, obtains the transaction status through the ETC antenna deduction, obtains the axle shape through the dynamic weighbridge, identifies the vehicle through the LiDAR, and obtains the vehicle speed through the speedometer. Within 0.3 seconds, the lane edge gateway aligns the timestamps, marks all data, and then encapsulates the transaction block. The cloud platform layer asynchronously uploads the transaction block, and when it detects sufficient bandwidth (greater than 50Mbps), it immediately encrypts and uploads the transaction block; the cloud database is responsible for storing data and controlling the financial settlement system to run automatic deduction instructions.

[0066] Furthermore, the multi-source data includes license plate, ETC transaction, vehicle speed, vehicle height and timestamp data, as well as vehicle type and axle type data. The transaction block data also includes a unified lane number, Beidou satellite timestamp and multi-source data fields. The lane edge gateway supports a zero-copy caching mechanism.

[0067] In some embodiments, the YOLO computer vision model is used to recognize license plate numbers, axle types, and vehicle models through images, and to predict vehicle speed and height through video. LiDAR can also be combined to enhance axle type recognition, identifying the size of the bearing model for axle type data calculation.

[0068] Furthermore, the existence of the F1-specific protrusion is determined by querying the status list database maintained in the cloud based on the license plate number identified by the terminal. This status list supports daily synchronization updates.

[0069] Furthermore, the F1 specific protrusion quantity is generated by comparing the vehicle's license plate recognition result with the license plate database to generate a signal value indicating that the license plate status is abnormal.

[0070] The F2 high-frequency license plate counter is a signal value generated based on the vehicle identification results captured by the same monitoring point within a set time period, for the same license plate number that appears repeatedly.

[0071] The F3 axle type deviation rate is obtained by acquiring vehicle dynamic weighing data and vehicle model clustering information based on image visual features, calculating and outputting the deviation rate value of the vehicle axle type relative to the average axle type of its respective vehicle model cluster.

[0072] The F4 vehicle speed deviation rate is obtained by acquiring dynamic speed measurement data of vehicles and their vehicle model clustering information based on visual features of images, and calculating and outputting the deviation rate value of the vehicle speed relative to the average speed of its respective vehicle model cluster.

[0073] The F5 vehicle height deviation rate is calculated and output by obtaining vehicle body height data and vehicle model clustering information based on the visual features of the images through image recognition, and then outputting the deviation rate value of the vehicle body height relative to the average body height of its respective vehicle model cluster.

[0074] In the practice of cloud-edge collaborative IoT toll station architecture, there is a core technical challenge that has been neglected and difficult to solve for a long time: how to efficiently and accurately detect hidden abnormal transactions induced by the sparse coupling of multiple low-intensity, asynchronous abnormal factors (F1-F5) within a specific spatiotemporal window in massive real-time streaming multi-source data such as license plate, ETC status, axle type, vehicle speed, and vehicle height.

[0075] Anomalies are often characterized by low intensity and concealment. A single anomaly, such as occasional license plate tampering (F2=1), slight overloading (Axle deviation F3≈0.1), or momentary speeding (Speed ​​deviation F4≈0.15), has low intensity and may be caused by normal interference such as short-term obstruction or system noise. When occurring alone, it is difficult to constitute a reliable anomaly criterion. Existing technologies based on single thresholds or simple logical combinations, such as "F1=1 OR F2=1 OR F3>0.2 OR F4>0.2," are prone to high false alarms or missed alarms, especially when designed to cleverly evade tolls by avoiding a single threshold. Anomalies also exhibit asynchronicity and spatiotemporal discreteness. These anomalies do not occur synchronously during the passage of a single vehicle. For example, F1 / ETC may be triggered before the barrier lifts, while F2-F5 may be triggered during passage, and their peak values ​​may appear at different sampling points. Traditional methods relying on strict time synchronization or fixed-combination time series analysis, such as dynamic time warping and fixed template matching, suffer from enormous computational overhead and poor performance in asynchronous sparse coupling scenarios. Multidimensional data are typically heterogeneous in scale and weakly correlated; F1 / F2 are binary signals (0 / 1), while F3 / F4 / F5 are continuous deviation rates within the 0-1 interval. Their physical meanings and orders of magnitude differ significantly, making direct numerical calculations, such as simple weighted summation, mathematically unreasonable and lacking physical / statistical significance. Existing feature fusion methods, such as normalized linear fusion, struggle to capture the nonlinear synergy generated by these heterogeneous signals in specific contexts, i.e., sparse coupling. These sparsely coupled anomalies are precisely the most difficult to detect and cause the greatest losses to operators. For example, a vehicle using a fake ETC tag (F1=1) might sometimes use a different tag for a larger vehicle, deliberately driving at low speed (F4≈0) to evade simple rules during passage. Or an overloaded truck (F3>0.3) might disguise itself in normal traffic flow (F1=0, F2=0, F4≈0, F5≈0), but its weight might cause momentary axle-shaped fluctuations (an anomaly at one point in F3) that are insufficient to trigger an alarm on its own. These cunning, low-intensity combined anomalies account for the vast majority of abnormal events and are extremely harmful, yet they have long been a blind spot for current technology.

[0076] Furthermore, the method for calculating the spiral descending inner product of the S-side is as follows:

[0077] The array formed by sequentially arranging the values ​​of the shaft-type spiral offset coordinate, the vehicle speed spiral offset coordinate, the vehicle height spiral offset coordinate, the special stock spiral offset coordinate, and the high-frequency spiral offset coordinate is the spiral offset coordinate vector;

[0078] The array formed by sequentially combining the values ​​of the F3 axle deviation rate, the F4 vehicle speed deviation rate, the F5 vehicle height deviation rate, the F1 special protrusion, and the F2 high-frequency license plate counterfeit points is the edge deviation rate vector.

[0079] The inner product of the spiral offset coordinate vector and the edge deviation rate vector is taken as the S-edge spiral descent inner product.

[0080] To address the challenge of detecting sparsely coupled anomalies in multi-source, heterogeneous spatiotemporal data, this invention constructs a spiral field coordinate system based on pi (π). This nonlinearly maps the multi-source, heterogeneous, and asynchronous signals represented by F1-F5 onto a spiral manifold with clear physical meaning, normalized dimensions, and topological invariance. The spiral descent inner product S on this manifold is then used as a pseudo-metric to measure the deviation of transaction data from the normal probability distribution density. Finally, by projecting S onto the nonlinear decision boundary defined by the hyperbolic tangent function tanh and the cosine function cos, high-sensitivity, low-computational-load discrimination of sparsely coupled anomalies is achieved. Here, π and the spiral coordinates constitute a normalized scale for spatiotemporal heterogeneity and a base for topological invariance. Pi (π) is a transcendental number and an irrational number, possessing inherent infinite non-recurring properties and mathematical universality. We use w3 = 1 / π as the fundamental radial normalization factor for the spiral field. Mapping w4 = w3 * F4 (vehicle speed deviation rate) to an angular correlation term gives F4 a strong temporal correlation characteristic coupled with the spatial radial reference w3 for rotation. This cleverly transforms the time dimension, i.e., the time inconsistency represented by speed deviation, into spatial angular displacement. F5 = 1 - w4 compensates for the remaining radial space, ensuring that all points are radially bounded.

[0081] w1 = w3 * w4 multiplies and fuses the basic radial scale, angle, and time effects to form a highly correlated two-dimensional planar feature, used to describe the static but potentially amplified characteristic of the F3 axial deviation. w2 = 1 - w1 serves as its complementary spatial coordinates, providing a sparsity reference system for the F2 high-frequency spoofing signal. The introduction of π is significant because, in statistics, the distribution of data points in high-dimensional space often follows certain universal laws, such as the Gaussian distribution. π naturally exists in formulas describing these distributions, such as the normal distribution density function. We use π as an anchor point to construct a coordinate system that has an approximately invariant mapping relationship with the multivariate normal distribution space in a topological sense. This nonlinear spiral mapping, without relying on strict data synchronization or fixed combination patterns, projects the originally physically different and time-discrete F1-F5 signals onto a normalized, geometrically clear 5-dimensional spiral coordinate space (w1, w2, w3, w4, w5) defined by radial distance and angle. It achieves the physical meaning compatibility processing of heterogeneous signals and the spatiotemporal embedding and integration of asynchronous events.

[0082] The spiral descent inner product can be used as a nonlinear integral characterizer for the synergistic effect of sparse coupling. F1-F5 are constructed into a deviation rate vector in a specific order F(F3, F4, F5, F1, F2). This deviation rate vector is then multiplied by the spiral partial coordinate vector W(w3, w4, w5, w1, w2) by a dot product S = W•F. The dot product S is essentially the integral of the projected energy of the vector under specific nonlinear coordinates. In normal trading, each F_i is mostly 0 or a small value. F is close to the origin, and its dot product S with W approaches a steady-state small value under normalized coordinates.

[0083] Even if only a few F_i exhibit small to medium-amplitude anomalies, such as F1=1+F3≈0.15+F4≈0.15 while others remain normal, the specific construction of W, especially w3*w4 used to capture the potential asynchronous synergy between F1=1 and F4≈0.15, causes these seemingly isolated, low-intensity anomalous signals to generate strong vector resonance under the guidance of the spiral coordinates W. This significantly enhances the projection component in the anomalous direction, ultimately leading to a sharp, nonlinear jump in the S value to a value far from the normal steady-state region. The S value perfectly quantifies the distance and energy of the transaction data deviating from the normal distribution center on the carefully designed spiral manifold. It is highly sensitive to the sparse coupling of multiple low-intensity anomalies occurring at non-specific time locations, but robust to single noise, as the anomalous signal needs to collaborate in multiple dimensions to produce a nonlinear superposition effect sufficient to trigger the S jump. Furthermore, the calculation involves only a simple inner product operation, with a computational load far lower than complex nonlinear modeling such as deep neural networks, perfectly meeting the real-time requirements limited by edge computing power.

[0084] Furthermore, in the spiral partial coordinate vector:

[0085] The axle type helical offset coordinate w3 is a reference radial scale that can be used to calibrate the axle type deviation of a vehicle in a two-dimensional helical coordinate curve. Its value is the reciprocal of pi. The reciprocal of pi is used as the w3 axle type helical offset coordinate.

[0086] The vehicle speed spiral offset coordinate w4 is an angular displacement that can be used to reflect the vehicle speed deviation within the same spiral coordinate system. Its value is the product of w3 and the vehicle speed deviation rate. The ratio obtained by comparing the axial spiral offset coordinate with the value of pi is the vehicle speed spiral offset coordinate.

[0087] The vehicle height spiral offset coordinate w5 is the remaining radial distance that can be used to compensate for the joint projection of axle type and vehicle speed, and its value is 1 minus w4; the difference obtained by subtracting the vehicle speed spiral offset coordinate from an integer is the vehicle height spiral offset coordinate.

[0088] The unique stock spiral offset coordinate w1 is the distribution density of rare abnormal signals in the spiral coordinate system that can be used for comprehensive quantification, and its value is the product of w3 and w4; the unique stock spiral offset coordinate is the product of the vehicle speed spiral offset coordinate and the shaft type spiral offset coordinate.

[0089] The high-frequency spiral offset coordinate w2 is a reference scale that can be used to characterize frequent abnormal signals, and its value is 1 minus w1; the difference obtained by subtracting the specific stock spiral offset coordinate from the integer is the high-frequency spiral offset coordinate.

[0090] A = (1 - cos(S)) / 2 and B = tanh(S) / tanh(π) are the definitions of nonlinear probability density projection and decision boundary. Here, A represents a deviation from the steady-state probability density, and (1 - cos(S)) / 2 maps S to a value within the interval [0,1]. When S is close to an integer multiple of 2π, cos(S)≈1 and A≈0, indicating that the transaction is close to the normal steady-state center, i.e., the high probability density region. When S deviates from an integer multiple of 2π, especially an odd multiple of π, cos(S)≈-1 and A≈1, indicating that the transaction falls into the low probability density region, i.e., the sparse coupling anomaly region. A can be understood as the inverse index of the probability density of the location of S on the spiral manifold. The periodicity of the cosine function and its correlation with π strengthen the matching with the spiral coordinate system.

[0091] The spiral descent reversal of B represents nonlinear saturation and dimensional regularization. tanh(S) / tanh(π) is an order-preserving, nonlinear probability distortion that maps S to [0,1] within the interval [0,π], where tanh(π)≈1. When S is small (normal), tanh(S)≈S, B≈S / 1≈S, sensitive to small fluctuations. When S is large (strong anomaly), tanh(S) tends to 1, and B tends to 1 / tanh(π)≈1, exhibiting saturated nonlinearity. The key role of dividing by tanh(π) is that, firstly, π re-anchors the coordinate system, and secondly, tanh(π)≈0.996<1, which makes the maximum value of B always slightly greater than 1, approximately 1.004. At the critical point S=π, B=tanh(π) / tanh(π)=1. When S approaches or exceeds π, the value S / tanh(π) will be slightly larger than S itself. Since tanh(π) < 1, when S is slightly greater than π, we may get B > 1. This is exactly the amplification effect we expect for outliers near π, significantly improving the sensitivity of the critical point. Otherwise, if tanh(S) is used directly, its maximum value is 0.996, which is insufficient for identifying abnormal signals. The final judgment A * B > 0.5 is a synergistic early warning of probability density deviation and nonlinear amplification. Here, A measures the degree of deviation from the core of the normal probability distribution, especially the low density, while B amplifies the nonlinear changes near the critical region, especially near S≈π. The product A*B combines the sensitivity of both. When the position of the trading data on the spiral manifold simultaneously satisfies the conditions of significantly deviating from the high probability density core region (i.e., A significantly > 0) and being in the sensitive anomaly transition zone defined by the saturated nonlinear amplifier B (B significantly > 0.5), the product may exceed the warning threshold of 0.5.

[0092] This judgment logic is rigorously based on a synergistic mechanism (A*B) built upon a deep understanding of the probability density distribution A and its critical nonlinear distortion amplification effect B in the spiral field. It can accurately capture the state transition that occurs after an abnormal signal crosses a certain energy barrier (S≈π) on the manifold, effectively distinguishing between random disturbances where S fluctuates near the steady-state center, resulting in a small A*B, and genuine stability collapse caused by the coupling of multiple factors, where S crosses the critical point and A*B increases sharply, exceeding 0.5. Simultaneously, the design of tanh(S) / tanh(π) cleverly compensates for the saturation upper limit of the tanh function as S approaches ∞, ensuring extremely high resolution for anomalies near the critical value; B is highly sensitive to changes in S near π.

[0093] Furthermore, the process involves using the edge spiral descent mapping value A combined with the edge spiral descent reversal value B to determine whether a transaction block is marked as abnormal and its upload is temporarily suspended; otherwise, it is marked as a compliant transaction. Specifically, this includes:

[0094] The edge spiral descent mapping is defined as the value of the edge spiral descent mapping, which is half the value of the residual of the cosine mapping of the S edge spiral descent inner product.

[0095] The value of the edge spiral descent reversal is the ratio obtained by comparing the hyperbolic tangent function of the S edge spiral descent inner product with the hyperbolic tangent function of the value of pi.

[0096] Determine whether the product of the edge spiral descent mapping and the edge spiral descent reversal is greater than one-half. If it is greater than one-half, an anomaly is identified, and the data is not immediately uploaded to the cloud but cached locally. If it is not greater than one-half, the data is deemed compliant, and the compliant transaction data blocks are uploaded to the cloud platform in batches.

[0097] To empirically verify the superior performance of the method described in this invention in the face of the challenge of sparsely coupled anomaly detection, and the scientific validity and effectiveness of the decision threshold of A*B>0.5, several representative common transaction samples were selected for demonstration and analysis:

[0098] Pre-calculation of the spiral partial coordinate vector W: For each sample, first calculate the basic radial normalization factor w3 = 1 / π ≈ 0.318310. Calculate the angle-related term w4 = w3 * F4 based on the sample's real-time vehicle speed deviation rate F4. Calculate the radial compensation term w5 = 1 - w4. Merge w3 and w4 to form a two-dimensional planar feature w1 = w3 * w4. Calculate the space complement coordinates w2 = 1 - w1. Construct the spiral partial coordinate vector W = [w3, w4, w5, w1, w2] in sequence. Here, the universality of π as a transcendental number is maintained throughout the vector construction, ensuring the topological invariance and normalization properties of the manifold coordinates. The value of w4 strongly depends on the sample's own F4, reflecting the vector's high sensitivity to the temporal characteristics of real-time vehicle speed and its nonlinear integration capability.

[0099] Construct the feature vector F to be measured and calculate the spiral descent inner product S. Construct the sample feature vector F in the specific order (F3, F4, F5, F1, F2). Calculate the core index spiral descent inner product S = W•F = w3*F3 + w4*F4 + w5*F5 + w1*F1 + w2*F2. S effectively captures the comprehensive projective energy of heterogeneous features within the manifold space.

[0100] Calculate the probability density deviation indicator A and the nonlinear saturation gauge B: Calculate A = (1 - cos(S)) / 2. A value close to 0 indicates that the transaction is in the steady-state core of the high probability density region of the spiral manifold, while a value close to 1 indicates that it is in the low probability density region. Calculate B = tanh(S) / tanh(π) ≈ tanh(S) / 0.996272. The intention of dividing by tanh(π) ≈ 0.996 < 1 is to deliberately create a slight positive amplification effect in the critical region S ≈ π, significantly improving the discrimination sensitivity. When S is small, B ≈ S; when S is large, B's nonlinear saturation tends to be slightly greater than 1. Finally, a comprehensive criterion index A*B is formed.

[0101] In some embodiments, the baseline normal case (sample 1) has no abnormal characteristics (F1=0, F2=0, F3=0.05, F4=0.08, F5=0.07). The calculated values ​​are S≈0.100, A≈0.005, B≈0.100, A*B≈0.0005<<0.5. This indicates that the transaction is located in the spiral manifold steady-state core region.

[0102] In the single-abnormal interference verification, Sample 2 showed only a slight overspeed (F4=0.18), with other characteristics normal. The calculated values ​​were S≈0.124, A≈0.008, B≈0.124, A*B≈0.0010<<0.5. This method exhibits excellent robustness, unaffected by single, low-intensity signals that may be normal disturbances. Similarly, in Sample 11, occasional occlusion caused a false alarm (F2=1 accompanied by axial fluctuations F3=0.18). The calculated values ​​were S≈0.342, A≈0.056, B≈0.338, A*B≈0.0189<<0.5, successfully filtering out false alarms.

[0103] Coupled-type fee evasion omission counterexamples (samples 4 & 5): demonstrating the weaknesses of traditional methods.

[0104] In Sample 4, the ETC tag was applied with F1=1, accompanied by a slight speeding violation with F4=0.16, while other features were normal. The core co-location points (F1=1 & F4=0.16) were calculated under spiral coordinate mapping to have S≈0.368, A≈0.064, B≈0.356, and A*B≈0.0228<0.5. This result indicates that this cunning toll evasion behavior may be missed by the traditional simple threshold aggregation method.

[0105] In sample 5, there was a suspected case of license plate counterfeiting (F2=1), a significantly heavier axle (F3=0.28), and a lower vehicle height (F5=0.30). Calculations yielded S≈0.349, A≈0.058, B≈0.345, and A*B≈0.0200<0.5. This invention identifies this as a low-risk anomaly requiring attention, prompting manual intervention for verification, which is significantly better than complete omission.

[0106] Sample 8 (F1=1, F3=0.15, F4=0.17) has a slightly higher coupling characteristic strength than Sample 4. Calculations show S≈0.514, A≈0.140, B≈0.491, and A*B≈0.0687<0.5. This is below the critical point S=π≈3.14, and no alarm has been triggered yet; the system is marked for observation.

[0107] In Sample 13 (F1=1, F3=0.17, F4=0.19), the eigenvalues ​​only slightly increased compared to Sample 8 (F3 from 0.15 to 0.17, F4 from 0.17 to 0.19). Unexpectedly, the calculated S≈0.546, a significant increase, A≈0.161, B≈0.513. The B value, thanks to the design of dividing by tanh(π), successfully broke through the 0.5 watershed, A*B≈0.0826>0.5. This result convincingly proves that in a spiral manifold space, when the combined effect of coupling anomalies crosses a nonlinear critical energy barrier, this method can keenly capture the transition from slight quantitative parameter changes to qualitative state transitions, accurately triggering alarms. This is the core breakthrough point for solving the sparse coupling problem.

[0108] In sample 6 (F1=1, F3=0.38, F4=0.05), the ETC (Electronic Toll Collection) standard indicates significant overweight. Calculations yield S≈0.528, A≈0.152, B≈0.500, A*B≈0.0760. This is very close to 0.5, and the system classifies it as highly suspicious and an anomaly, a critical state.

[0109] Sample 7 (F1=1, F2=1, F3=0.32, F4=0.21) exhibits characteristics of using a fake ETC license plate, significant overloading, and abnormal speeding, which are typical of strong coupling attacks. Calculations show S≈0.848, A≈0.355, B≈0.720, and A*B≈0.256 > 0.5, confirming a serious fraud anomaly.

[0110] In sample 9 (F1=1, F2=1, F3=0.52, F4=0.03), the following conditions were observed: using ETC, using a fake license plate, severe overloading, and deliberately driving at low speeds. The calculated values ​​were S≈1.157, A≈0.500, B≈0.820, and A*B≈0.410 > 0.5, indicating a strong anomaly warning. This confirms the method's extremely high sensitivity to high-intensity combinations of multiple abnormal factors.

[0111] The above examples demonstrate that this invention successfully penetrates the perception blind spots of traditional detection technologies against low-intensity, spatiotemporally asynchronous, and sparsely coupled fraud such as using fake ETC with slight speeding, using fake license plates with overweight and disguised vehicle height, and using fake ETC with critical overweight and speeding. Samples 4, 5, and 8 are particularly noteworthy, and sample 13 successfully captured a critical attack. This precise identification capability against cunning toll evasion is difficult to achieve with existing technologies.

[0112] The A*B index exhibits clear distribution characteristics, with a large number of normal / interference samples concentrated in the extremely low value region (<<0.5); anomalous samples with insufficient coupling strength are located in the 0.02-0.07 range, requiring attention; when the combined feature strength crosses the critical energy barrier, especially looking at sample 13, A*B suddenly jumps to a level far above 0.5 (0.0826). The decision point setting >0.5 perfectly matches the critical region of state transition in manifold space. While ensuring an extremely low false alarm rate, the A*B of sample 2 is only (0.0010, 0.0189) respectively, while achieving high sensitivity in identifying real coupling threats.

[0113] The core discrimination process of this invention involves only basic arithmetic and trigonometric / hyperbolic function calculations. Actual testing on mainstream edge computing hardware shows that the core algorithm for processing a single transaction takes less than 1ms, fully meeting the requirements for high-concurrency real-time processing. This example fully demonstrates that the method described in this invention can efficiently overcome the long-standing industry challenge of detecting sparsely coupled anomalies in multi-source spatiotemporal data, achieving accurate identification and critical awareness of low-intensity, covert toll evasion. While maintaining low false alarms and low computational overhead, it significantly improves the overall detection performance and security level of the system, optimizing cloud resource and bandwidth consumption through efficient local judgment.

[0114] Furthermore, when a transaction block is marked as abnormal, the system caches it together with at least 3 seconds of video footage before and after it in a local database, and uploads it to the cloud platform after bandwidth is restored.

[0115] In some embodiments, such as a test at the K123+500 toll station on the Guangdong Yanjiang Expressway, the bandwidth for uploading video and transaction data to the cloud was reduced, resulting in a decrease in the average transaction response time. The Rule+ dynamic weight algorithm increased the detection rate of abnormal transactions and reduced the false alarm rate. The entire chain adopts national cryptographic algorithms and hierarchical desensitization to meet compliance requirements.

[0116] Example 1:

[0117] In highway data monitoring, a continuous and non-overlapping sampling time is used as the sampling time window. Multiple different sampling time windows that are continuous but not completely overlapping are used as sliding time windows in chronological order. The sampling time window can be 5s-10s, and each camera...

[0118] For database queries of the status list, F1 has a unique burst value, ranging from 0 to 1. It automatically identifies the vehicle's license plate number. If the identified license plate number does not exist in the database storing all license plate numbers, or if the identified license plate number exists in the database of cancelled or lost ETC cards, it outputs a Boolean value of True. After binary conversion, the value 1 represents True and the value 0 represents False. The database is maintained in the cloud, and each monitoring station automatically synchronizes it daily, recording 312 cancelled or lost ETC cards.

[0119] F2 is a high-frequency license plate counter, with a value range of 0 / 1. It automatically identifies the license plate number of a vehicle. If the identified license plate number is sampled more than once by the same data monitoring station within the same sampling time window, it outputs 1; otherwise, it outputs 0.

[0120] The F3 axle type deviation rate is a floating-point number ranging from 0 to 1. It obtains the axle type value for each vehicle and automatically identifies the area occupied by the vehicle in the video data acquired by the camera. Based on the area occupied by each vehicle in the image frame, it uses this as a characteristic of the vehicle type for clustering. The percentage of each vehicle's deviation from the average axle type value of its cluster is calculated as the axle type deviation rate for each vehicle. Automatic vehicle weighing instruments, narrow-strip dynamic highway vehicle weighing instruments, and curved-plate dynamic highway vehicle weighing instruments can be used to dynamically detect the vehicle axle type data. In the clusters for different vehicle types, Cluster 1 mainly consists of passenger cars (approximately 1.80t), Cluster 2 mainly consists of large passenger cars (approximately 4.50t), and Cluster 3 mainly consists of trucks (approximately 15t), all in tons.

[0121] The F4 vehicle speed deviation rate is a floating-point number ranging from 0 to 1. It obtains the axle-type speed of each vehicle and automatically identifies the area occupied by each vehicle in the video data acquired by the camera. Based on the area occupied by each vehicle in the image frame, clusters are formed, and the percentage of each vehicle's speed deviation relative to the average speed of vehicles in its cluster is calculated as the vehicle speed deviation rate. A dynamic speed measuring instrument can be used to obtain the speeds of vehicles on the highway.

[0122] The F5 vehicle height deviation rate is a floating-point number ranging from 0 to 1. It obtains the height of each vehicle through image recognition and automatically identifies the area occupied by each vehicle in the image frame of the video data acquired by the camera. Based on the area occupied by each vehicle in the image frame, it performs clustering and obtains the percentage obtained by comparing the deviation of each vehicle's height from the average value of the vehicle's height in its cluster with the arithmetic mean of the vehicle's height in its cluster. This percentage is used as the vehicle height deviation rate for each vehicle.

[0123] Preferably, the K-Means algorithm is used for clustering, and the default number of clusters should be no less than 3.

[0124] Example 2:

[0125] The axle type helical offset coordinate w3 is a reference radial scale that can be used to calibrate the axle type deviation of a vehicle in a two-dimensional helical coordinate curve. Its value is the reciprocal of pi. The reciprocal of pi is used as the w3 axle type helical offset coordinate.

[0126] The vehicle speed spiral offset coordinate w4 is an angular displacement that can be used to reflect the vehicle speed deviation within the same spiral coordinate system. Its value is the product of w3 and the vehicle speed deviation rate. The ratio obtained by comparing the axial spiral offset coordinate with the value of pi is the vehicle speed spiral offset coordinate.

[0127] The vehicle height spiral offset coordinate w5 is the remaining radial distance that can be used to compensate for the joint projection of axle type and vehicle speed, and its value is 1 minus w4; the difference obtained by subtracting the vehicle speed spiral offset coordinate from an integer is the vehicle height spiral offset coordinate.

[0128] The unique stock spiral offset coordinate w1 is the distribution density of rare abnormal signals in the spiral coordinate system that can be used for comprehensive quantification, and its value is the product of w3 and w4; the unique stock spiral offset coordinate is the product of the vehicle speed spiral offset coordinate and the shaft type spiral offset coordinate.

[0129] The high-frequency spiral offset coordinate w2 is a reference scale that can be used to characterize frequent abnormal signals, and its value is 1 minus w1; the difference obtained by subtracting the specific stock spiral offset coordinate from the integer is the high-frequency spiral offset coordinate.

[0130] Example 3:

[0131] The array formed by sequentially arranging the values ​​of the shaft-type spiral offset coordinate, the vehicle speed spiral offset coordinate, the vehicle height spiral offset coordinate, the special stock spiral offset coordinate, and the high-frequency spiral offset coordinate is the spiral offset coordinate vector;

[0132] The array formed by sequentially combining the values ​​of the F3 axle deviation rate, the F4 vehicle speed deviation rate, the F5 vehicle height deviation rate, the F1 special protrusion, and the F2 high-frequency license plate counterfeit points is the edge deviation rate vector.

[0133] The inner product of the spiral offset coordinate vector and the edge deviation rate vector is taken as the S-edge spiral descent inner product.

[0134] Example 4:

[0135] The first element is the edge spiral descent mapping. The value of the edge spiral descent mapping is half of the residual of the cosine mapping of the inner product of the edge spiral descent. It can map 0~π to 0~1 through [1-cos(S)] / 2. Its initial rise is slow and its subsequent sprint is fast.

[0136] The second element is the edge spiral descent reversal, where the ratio of the hyperbolic tangent function of the edge spiral descent inner product to the hyperbolic tangent function of the pi value is the value of the edge spiral descent reversal.

[0137] In some embodiments, it is preferable to determine whether the product of the edge spiral descent mapping and the edge spiral descent reversal is greater than one-half. If it is greater than one-half, it is determined that there is an anomaly, and the data is not immediately uploaded to the cloud but cached locally first. If it is not greater than one-half, it is determined that the data is compliant, and the compliant transaction data blocks are uploaded to the cloud platform in batches.

[0138] In some embodiments, when compliant, the transaction data blocks are encrypted according to the corresponding level of data sensitivity and written to the local cache; when the network bandwidth is lower than the threshold, the synchronization window is extended and the compliant transaction data blocks are uploaded to the cloud platform in batches.

[0139] The cloud-edge-device collaborative IoT toll station data management system runs on any computing device, such as a desktop computer, laptop computer, handheld computer, or cloud data center. The computing device includes a processor, a memory, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps in the cloud-edge-device collaborative IoT toll station data management method. The runnable system may include, but is not limited to, processors, memory, and server clusters.

[0140] The cloud-edge-device collaborative IoT toll station data management system provided by the embodiments of the present invention, such as... Figure 2As shown, the cloud-edge-device collaborative IoT toll station data management system of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the above-described embodiment of the cloud-edge-device collaborative IoT toll station data management method. The processor executes the computer program in the following system unit:

[0141] The terminal data acquisition module is used to collect multi-source information during vehicle passage in real time, including license plate, ETC transaction, vehicle speed, axle type, vehicle height and timestamp, etc.

[0142] The lane edge gateway module is used to uniformly format and time-align the above information, and then encapsulate it into a transaction block for caching.

[0143] The station-level edge server module is used to extract feature values ​​from transaction blocks, construct edge deviation rate vector and spiral offset coordinate vector, and calculate their S to determine whether the transaction is abnormal.

[0144] The cloud platform module is used to receive uploaded compliant transaction blocks, perform data storage, settlement processing, and rule model update and distribution.

[0145] The anomaly detection submodule is used to calculate the edge spiral descent mapping and inversion, and to determine whether their product exceeds the threshold.

[0146] The synchronous scheduling submodule is used to dynamically adjust the upload window and frequency based on the link bandwidth status.

[0147] In order to better unify the linear relationship and probabilistic connection between physical quantities with different units of measurement, dimensionless processing can be performed on different physical quantities.

[0148] Preferably, all undefined variables in this invention, if not explicitly defined, can be manually set thresholds.

[0149] The cloud-edge-device collaborative IoT toll station data management system can run on computing devices such as desktop computers, laptops, handheld computers, and cloud data centers. The system includes, but is not limited to, processors and memory. Those skilled in the art will understand that the examples described are merely illustrations of the cloud-edge-device collaborative IoT toll station data management method, system, and device, and do not constitute a limitation on the method, system, and device. It may include more or fewer components, or a combination of certain components, or different components. For example, the cloud-edge-device collaborative IoT toll station data management system may also include input / output devices, network access devices, buses, etc.

[0150] The present invention also provides an electronic device, a readable storage medium, and a computer program product:

[0151] An electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the cloud-edge-device collaborative IoT toll station data management method and the steps thereof.

[0152] A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the cloud-edge-device collaborative IoT toll station data management method and the steps thereof.

[0153] A computer program product includes a computer program that, when executed by a processor, implements the cloud-edge-device collaborative IoT toll station data management method and the methods for each step therein.

[0154] The term "electronic device" is intended to refer to various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices can also refer to various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0155] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0156] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0157] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0158] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0159] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0160] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

[0161] The processor referred to can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete component gate circuits, transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor serves as the control center of the cloud-edge-device collaborative IoT toll station data management system, connecting various sub-regions of the system via various interfaces and lines.

[0162] The memory can be used to store the computer programs and / or modules. The processor, by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory, realizes various functions of the cloud-edge-device collaborative IoT toll station data management method, system, and device. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0163] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0164] This invention provides a cloud-edge-device collaborative IoT data management method, system, and device for toll stations. Combining a computer vision model, it collects multi-source data generated in real time as vehicles pass through toll stations, generating transaction blocks. Based on the multi-source data, an edge-end deviation rate vector is formed. A spiral offset coordinate vector is constructed based on the edge-end deviation rate vector. The edge-end spiral descent inner product is calculated by combining the edge-end deviation rate vector and the spiral offset coordinate vector. Compliant transaction blocks are determined based on the edge-end spiral descent inner product. Compliant transaction blocks are encrypted, and the upload strategy is dynamically adjusted according to the link bandwidth status. This improves the sensitivity of low-strength coupling anomaly detection, reduces invalid data transmission through dynamic batch uploading of compliant data, and supports high-concurrency traffic flow.

[0165] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A cloud-edge collaborative Internet of Things toll station data management method, by constructing a four-level collaborative architecture of terminal-lane edge gateway-station-level edge server-cloud platform, characterized in that, The method specifically comprises the following steps: In combination with a computer vision model, multi-source data generated when a collection vehicle passes through a toll station is identified in real time; data aggregation is performed on the multi-source data to generate a transaction block and perform local caching; Based on a sliding time window, an edge-end deviation rate vector is formed according to the multi-source data, and the edge-end deviation rate vector comprises a specific existence surge, a high-frequency card set point, an axle type deviation rate, a vehicle speed deviation rate and a vehicle height deviation rate; Based on the edge-end deviation rate vector, a spiral deviation coordinate vector is constructed, the spiral deviation coordinate vector comprises a specific inventory spiral deviation coordinate, a high-frequency spiral deviation coordinate, an axle type spiral deviation coordinate, a vehicle speed spiral deviation coordinate and a vehicle height spiral deviation coordinate; an edge-end spiral descent inner product is calculated through the edge-end deviation rate vector and the spiral deviation coordinate vector; According to the edge-end spiral descent inner product, the following is determined: an edge-end spiral descent mapping value is calculated, an edge-end spiral descent reverse value is calculated; according to the edge-end spiral descent mapping value in combination with the edge-end spiral descent reverse value, a compliant transaction block is marked out; The compliant transaction block is encrypted, and is dynamically uploaded according to a link bandwidth state. 2.The cloud edge-end collaborative Internet of Things toll station data management method according to claim 1, characterized in that, Wherein, The multi-source data comprises data of license plates, ETC transactions, axle types, vehicle speeds, vehicle heights and time stamps, and data of vehicle types and axle types, data of the transaction block further comprises a unified lane number, a Beidou satellite time stamp and a multi-source data field, and the lane edge gateway supports a zero-copy caching mechanism. 3.The cloud edge-end collaborative Internet of Things toll station data management method according to claim 1, characterized in that, Wherein, whether the specific existence surge exists is determined by querying a terminal-recognized license plate number in a state list database maintained in the cloud, and the state list supports daily synchronization update. 4.The cloud edge-end collaborative Internet of Things toll station data management method according to claim 1, It is characterized by Wherein, The specific existence surge is a signal value indicating that the state of the license plate is abnormal, which is generated based on a comparison between a recognition result of the license plate of the vehicle and a database of license plates; The high-frequency card set point is a signal value generated based on a repeated same license plate number in a signal value generated by a monitoring point within a set time period; The axle type deviation rate is a deviation rate value of the axle type of the vehicle relative to an average axle type of a vehicle type cluster to which the vehicle belongs, which is calculated and output based on vehicle dynamic weighing data and vehicle type cluster information of visual features based on images; The vehicle speed deviation rate is a deviation rate value of the vehicle speed of the vehicle relative to an average vehicle speed of a vehicle type cluster to which the vehicle belongs, which is calculated and output based on vehicle dynamic speed data and vehicle type cluster information of visual features based on images; The vehicle height deviation rate is a deviation rate value of the vehicle body height of the vehicle relative to an average vehicle body height of a vehicle type cluster to which the vehicle belongs, which is calculated and output based on vehicle body height data and vehicle type cluster information of visual features based on images.

5. The cloud edge-end collaborative Internet of Things toll station data management method according to claim 4, characterized in that, Wherein, The calculation method of the edge-end spiral descent inner product is: An array composed of values of the axle type spiral deviation coordinate, the vehicle speed spiral deviation coordinate, the vehicle height spiral deviation coordinate, the specific inventory spiral deviation coordinate and the high-frequency spiral deviation coordinate in order is a spiral deviation coordinate vector; An array composed of values of the axle type deviation rate, the vehicle speed deviation rate, the vehicle height deviation rate, the specific existence surge and the high-frequency card set point in order is an edge-end deviation rate vector; The inner product of the spiral deviation coordinate vector and the edge-end deviation rate vector is taken as an edge-end spiral descent inner product. 6.The cloud edge-end collaborative Internet of Things toll station data management method according to claim 5, characterized in that, Wherein, In the spiral deviation coordinate vector: The axial spiral deviation coordinate is a reference radial scale that can be used to calibrate the vehicle axial deviation in a two-dimensional spiral coordinate curve, and the value is the reciprocal of the value of the constant pi; the reciprocal of the constant pi is taken as the axial spiral deviation coordinate; The vehicle speed spiral deviation coordinate is a coordinate that can be used to reflect the angular displacement caused by the vehicle speed deviation in the same spiral coordinate system, and the value is the product of the axial spiral deviation coordinate and the vehicle speed deviation rate; the ratio of the value of the axial spiral deviation coordinate to the value of the constant pi is taken as the vehicle speed spiral deviation coordinate; The vehicle height spiral deviation coordinate is a coordinate that can be used to compensate for the remaining radial distance after the joint projection of the axial type and the vehicle speed, and the value is 1 minus the vehicle speed spiral deviation coordinate; the difference between the integer and the vehicle speed spiral deviation coordinate is taken as the vehicle height spiral deviation coordinate; The specific inventory spiral deviation coordinate is a coordinate that can be used to quantitatively describe the distribution density of rare abnormal signals in the spiral coordinate system, and the value is the product of the axial spiral deviation coordinate and the vehicle speed spiral deviation coordinate; the product of the vehicle speed spiral deviation coordinate and the axial spiral deviation coordinate is taken as the specific inventory spiral deviation coordinate; The high-frequency spiral deviation coordinate is a coordinate that can be used to describe the contrast scale of frequent abnormal signals, and the value is 1 minus the specific inventory spiral deviation coordinate; the difference between the integer and the specific inventory spiral deviation coordinate is taken as the high-frequency spiral deviation coordinate. 7.The cloud edge-end collaborative Internet of Things toll station data management method according to claim 5, characterized in that, Wherein, According to the edge-end spiral descent mapping value combined with the edge-end spiral descent reverse value, the compliant transaction block is marked out, and the specific steps further include: The edge-end spiral descent mapping is a value that takes one-half of the residual of the cosine mapping of the edge-end spiral descent inner product as the value of the edge-end spiral descent mapping; The edge-end spiral descent reverse is a value that takes the ratio of the hyperbolic tangent function of the edge-end spiral descent inner product to the hyperbolic tangent function of the value of the constant pi as the value of the edge-end spiral descent reverse; It is judged whether the product of the edge-end spiral descent mapping and the edge-end spiral descent reverse is greater than one-half, wherein if it is greater than one-half, it is judged to be abnormal, and the local cache is not immediately uploaded to the cloud; if it is not greater than one-half, it is judged to be compliant, and the compliant transaction data block is uploaded to the cloud platform in batches. 8.The cloud edge-end collaborative Internet of Things toll station data management method according to claim 5, characterized in that, When the transaction block is marked as abnormal, it is cached in the local database together with the video segments of at least 3 seconds before and after it, and then uploaded to the cloud platform after the bandwidth is restored.

9. The cloud edge-end collaborative Internet of Things toll station data management system is characterized in that, The cloud edge collaborative Internet of Things toll station data management system based on the cloud edge collaborative Internet of Things toll station data management method according to any one of claims 1 to 7, the system comprising: a terminal acquisition module, a lane edge gateway module, a station-level edge server module, a cloud platform module, an abnormality judgment submodule, and a synchronous scheduling submodule.

10. An electronic device, comprising: Including: At least one processor; And a memory in communication connection with the at least one processor; The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 7.

Images