Expressway charging auditing system and method based on micro-service architecture

Abstract

The invention discloses an expressway charging auditing system and method based on a micro-service architecture, the system comprises a data auditing subsystem and a data application subsystem, the two subsystems are different services, and each service adopts a single architecture mode; the data auditing subsystem comprises a data acquisition module and a data analysis module, the data acquisition module performs data acquisition, the data analysis module is used for constructing a toll evasion vehicle screening model and obtaining an abnormal screening result of a vehicle according to the acquired data, and the data application subsystem comprises a suspicion vehicle checking module and a toll evasion vehicle checking module. And the feature extraction module is used for constructing a multi-branch vehicle re-identification model, performing feature extraction on the suspicious vehicles and the vehicles in the retrieval vehicle library based on the toll evasion vehicle screening model, and then classifying the suspicious vehicles and the vehicles in the retrieval vehicle library according to the feature distance to obtain views of the suspicious vehicles under different cameras. According to the invention, automatic analysis and screening of mass pass transaction data can be realized, and the suspected vehicle screening accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of highway management, and in particular to a highway toll auditing system and method based on a microservice architecture. Background Art

[0002] At present, my country's highway toll auditing methods have made great progress, but there are still some problems and challenges.

[0003] The existing highway toll auditing methods have gradually developed from traditional purely manual auditing to automation and informationization. By introducing advanced computer technology and big data analysis technology, the highway toll auditing system can realize automated auditing and big data analysis. Compared with manual auditing, the auditing efficiency and accuracy have been greatly improved. Secondly, the application areas of highway toll auditing methods are also constantly expanding. In addition to toll auditing business, it can also be applied to road network operation monitoring, traffic safety management and other aspects. By introducing intelligent technology, real-time monitoring and early warning of the operation status of road network vehicles can be realized, improving the operation efficiency and service quality of the road network.

[0004] However, after the provincial border toll stations were abolished, under the "one network" toll collection model, the massive transaction data and the endless evasion methods have made the existing highway toll auditing methods face new problems and challenges. For example, the massive mixed transaction data has greatly reduced the screening efficiency and accuracy of the auditing system, and there are cases of false positives or omissions, which requires continuous improvement and optimization of the algorithm. In addition, the existing auditing algorithm has a single screening type. With the continuous iteration and upgrading of evasion methods, it is impossible to effectively cover a variety of evasion types, and evasion behavior is difficult to effectively curb. Finally, the existing highway toll auditing methods do not have a complete evidence chain mechanism in terms of evasion evidence chain collection, and the evasion recovery mainly relies on manual evidence collection, which is inefficient. In the future, highway toll auditing methods will continue to develop in the direction of automation and informatization, and continue to expand their application areas. At the same time, it is necessary to strengthen technology research and development and innovation, improve the accuracy and efficiency of the auditing system, and provide more comprehensive and efficient support for highway operation and management.

[0005] In the prior art, the highway toll audit system adopts a monolithic architecture, which is composed of multiple monolithic subsystems such as data acquisition subsystem, data storage subsystem, data analysis subsystem, data application subsystem and data display subsystem, and the systems are relatively independent. Figure 1 Provide detailed explanation.

[0006] 1. The data acquisition subsystem makes telemetry requests to the gantry equipment on the highway and the toll collection equipment in the toll station lanes through protocols such as RPC (Remote Procedure Call) or MQTT (Message Queue Telemetry Transport Protocol) to obtain vehicle passage records and transaction flow records.

[0007] 2. The data storage subsystem uses storage tools such as MySQL relational database and Hive database to persistently store the data collected by the data collection subsystem.

[0008] 3. The data analysis subsystem specifies data analysis strategies for different types of fee evasion, and initiates data query requests to the data storage subsystem through interface calls, etc. The data storage subsystem persists the query results in the data storage subsystem database.

[0009] 4. The data application subsystem applies the data screening results of the data analysis subsystem, including the initiation of audit work orders, the recovery of vehicle fee evasion, the connection with third-party platforms, and the provision of interface services for user interaction with the data display subsystem.

[0010] 5. The data display subsystem visualizes the functions of the data application subsystem, providing highway operation management users with visual operation pages and clear data display functions.

[0011] The inventors of this application found that the prior art has at least the following technical problems during the implementation of the present invention:

[0012] 1) Relying on manual identification, there are problems such as low accuracy, low efficiency, single means, large amount of data, high degree of dispersion, and difficulty in manual screening, resulting in a lack of evidence chain for evasion and collection; 2) It is difficult to curb evasion. Traditional evasion behaviors such as card replacement and large vehicle with small label are difficult to curb, and new evasion behaviors emerge in an endless stream, causing huge toll losses; 3) A complete toll audit system has not been established. The current audit business system of expressways is relatively independent, and there are information barriers between each business system. Functions cannot be interoperable, data cannot be shared, and there is a lack of channels to form a chain of evidence, making it difficult to form an overall audit system and form a closed loop. 4) The toll audit credit system is missing. Driving information has not been managed in black, white, and warning lists, which is inconvenient to regulate the driving and abnormal driving behavior of car owners. 5) The traditional toll audit system adopts a monolithic architecture, with a high degree of system coupling and is not easy to maintain. Summary of the invention

[0013] In view of the above defects or improvement needs of the prior art, the present invention provides a highway toll audit system based on a microservice architecture.

[0014] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0015] In a first aspect, a highway toll audit system based on a microservice architecture is provided, including: a data audit subsystem and a data application subsystem, wherein the data audit subsystem and the data application subsystem are different services, each service adopts a monolithic architecture, and the services are isolated from each other;

[0016] The data audit subsystem includes a data acquisition module and a data analysis module. The data acquisition module is used to collect data. The data analysis module is used to build a screening model for fare evasion vehicles, and obtain abnormal screening results of vehicles based on the data collected by the data acquisition module through the fare evasion vehicle screening model. The fare evasion vehicle screening model adopts an abnormal fare evasion classification network with a multi-attention mechanism, and the abnormal screening results of vehicles include abnormal types of suspicious vehicles and vehicle information;

[0017] The data application subsystem includes a suspected vehicle verification module, which is used to build a multi-branch vehicle re-identification model, and extract features of suspicious vehicles and vehicles in the search vehicle library based on the toll evasion vehicle screening model through the multi-branch vehicle re-identification model, and then classify the suspicious vehicles and vehicles in the search vehicle library based on feature distance to obtain views of the suspicious vehicles under different cameras.

[0018] In one embodiment, the toll evasion vehicle screening model includes a data fusion module, a first feature extraction module and a classification module, wherein the data fusion module is used to preprocess the input data, and the first feature extraction module performs feature extraction on the preprocessed data based on the self-attention ResNet backbone network, so that the model pays attention to the connection and hierarchical relationship between the three types of data to extract feature information related to toll evasion anomalies; the classification module predicts and classifies the results of feature extraction through a fully connected layer, and outputs the abnormal type of the suspicious vehicle, wherein the data collected by the data acquisition module is used as input data, including vehicle attribute information, toll station passage data and gantry data.

[0019] In one implementation, the self-attention mechanism adopted by the first feature extraction module is:

[0020]

[0021] Among them, A(f,w) is the attention function of the toll evasion vehicle screening model, f and w are the independent variables of the attention function, and f k is the feature tensor corresponding to channel k in the last layer of the network, w k is the feature tensor f k The corresponding weight, d is the scaling factor, and C is the total number of channels.

[0022] In one embodiment, the multi-branch vehicle re-identification model includes an image processing module, a second feature extraction module, a similarity measurement and reordering module, wherein the image processing module is used to perform dehazing processing on the image, the second feature extraction module is used to use a self-attention-based ResNet backbone network to extract features, and then use a linear fusion strategy to fuse the extracted global features and local features, and the similarity measurement and reordering module is used to use a distance function to measure the similarity between the features of suspicious vehicle samples and the features of vehicle samples in the retrieved vehicle library, and reorder according to the results of the similarity measurement, and output views of the suspicious vehicle under different cameras.

[0023] In one implementation, the image processing module performs defogging on the image in the following manner:

[0024] MSRCR(x,y)=γ(x,y)×RMSR(x,y)+β

[0025] Among them, MSRCR(x,y) represents the image after color restoration, x and y represent the horizontal and vertical coordinates of the input image, RMSR(x,y) represents the estimated value of the reflection component after multi-scale Retinex processing, γ is the scaling factor used to adjust the scale of the data, β is the bias term used to restore the original distribution characteristics of the data, and the formula for multi-scale Retinex processing is:

[0026] RMSR(x,y)=R(x,y)×L(x,y)

[0027] Where R(x,y) represents the reflection component of the object, and L(x,y) represents the illumination component of the ambient light;

[0028] The linear fusion strategy adopted is:

[0029]

[0030] in represents the fused features, λ 1 ,λ 2 ∈[0,1] is a hyperparameter; f G is the global feature, f i is the local feature at position i, w i is the local feature f i The weight coefficient of , L represents the total number of local features.

[0031] In one embodiment, the data application subsystem further includes an automated fee evasion work order processing module for creating and initiating fee evasion work orders.

[0032] In one embodiment, the data application subsystem also includes a vehicle credit system construction module and a toll evasion type frequency statistics module, wherein the credit system construction module is used to establish a credit database for highway vehicles and owners, and to manage vehicles in black and white lists according to the number of toll evasions and toll evasion types; the toll evasion type frequency statistics module is used to count the frequency of various toll evasion types of vehicles.

[0033] In one embodiment, the data application subsystem includes a system management module for performing system authority management and verification, and authorizing and authenticating system user authority.

[0034] In one embodiment, it also includes a security defense subsystem and a visualization display layer, wherein the security defense subsystem is used to maintain system security and resist external network attacks, and the visualization display layer provides users with data visualization and system operation visualization functions.

[0035] Based on the same inventive concept, the second aspect of the present invention provides a highway toll audit method based on a microservice architecture, based on the system implementation described in the first aspect, the method includes:

[0036] The data collection module collects vehicle attribute information, toll station traffic data and gantry data;

[0037] A screening model for toll evasion vehicles is constructed through the data analysis module, and an abnormal screening result of the vehicle is obtained through the screening model for toll evasion vehicles based on the data collected by the data collection module, wherein the screening model for toll evasion vehicles adopts an abnormal toll evasion classification network with a multi-attention mechanism, and the abnormal screening result of the vehicle includes the abnormal type of the suspicious vehicle and the vehicle information;

[0038] A multi-branch vehicle re-identification model is constructed through the suspected vehicle verification module, and the multi-branch vehicle re-identification model is used to extract features of suspicious vehicles and vehicles in the search vehicle library based on the toll evasion vehicle screening model, and then the suspicious vehicles and vehicles in the search vehicle library are classified based on feature distance to obtain views of the suspicious vehicles under different cameras.

[0039] Compared with the prior art, the advantages and beneficial technical effects of the present invention are as follows:

[0040] The present invention proposes a highway toll auditing system based on a microservice architecture, in which a data auditing subsystem and a data application subsystem are different services respectively, and each service adopts a monolithic architecture. The services are isolated from each other, which greatly reduces the coupling between services and improves the maintainability and scalability of the system; the data auditing subsystem further includes a data acquisition module and a data analysis module, and a toll evasion vehicle screening model is constructed through the data analysis module, and the abnormal screening results of the vehicles are obtained based on the data collected by the data acquisition module. The system can quickly and efficiently perform data analysis and data mining on the massive traffic and transaction flow data of the highway, reduce time costs, overcome the difficulty of manual screening, and greatly improve the screening accuracy of suspected vehicles. Compared with the manual auditing method, the accuracy is improved to a certain extent, and the abnormal screening results of vehicles can cover the screening of various toll evasion types, and can realize the screening of most toll evasion abnormal data. The data application subsystem includes a suspected vehicle verification module, which is used to build a multi-branch vehicle re-identification model. The model extracts features of suspicious vehicles and vehicles in the search vehicle library based on the toll evasion vehicle screening model, and then classifies the suspicious vehicles and vehicles in the search vehicle library based on feature distance to obtain views of the suspicious vehicle under different cameras, thereby realizing cross-lens vehicle tracking. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 It is a framework diagram of the highway toll audit system in the prior art;

[0043] Figure 2 A framework diagram of a highway toll audit system based on a microservice architecture provided in an embodiment of the present invention;

[0044] Figure 3 It is a structural diagram of a MANet network model in an embodiment of the present invention;

[0045] Figure 4 It is a structural diagram of the Ve-RI model in an embodiment of the present invention;

[0046] Figure 5 is a framework diagram of a system management module in an embodiment of the present invention;

[0047] Figure 6 This is a framework diagram of the security defense subsystem in an embodiment of the present invention; DETAILED DESCRIPTION

[0048] The present invention proposes a highway toll audit system based on microservice architecture, including a data audit subsystem and a data application subsystem. The microservice architecture is different from the traditional monolithic architecture mode. The system is divided into multiple services according to domains and functions. Each service adopts a monolithic architecture mode. The services are isolated from each other, which greatly reduces the coupling between services and improves the maintainability and scalability of the system. The data audit subsystem includes a data acquisition module and a data analysis module. The data acquisition module collects highway vehicle traffic data and transaction flow data. The data analysis module completes the analysis and decision-making of the collected data, builds a toll evasion vehicle screening model, and realizes the automatic analysis and screening of massive toll transaction data. The data application subsystem comprehensively applies the data analyzed by the data audit subsystem, including suspicious vehicle information display, suspicious vehicle verification, automatic toll evasion work order processing, vehicle credit system, toll evasion type frequency statistics and system management modules.

[0049] The key points and main innovative points of the present invention include:

[0050] 1. The highway toll audit system based on microservice architecture integrates data collection, data analysis, data application, and data display functions, breaking the problems of system independence, communication barriers between systems, and the inability to close the audit business under the monolithic architecture of the traditional toll audit system.

[0051] 2. In the face of complex and diverse toll evasion behaviors, based on deep learning methods, combined with the toll evasion rules of abnormal toll evasion behaviors such as large vehicles with small logos, entry without exit, U / J driving, shielded traffic media, and fake green pass, the Data Fusion module is introduced to pre-process the original images, and a self-attention mechanism is designed to mine more fine-grained semantic information. A MANet network model is constructed to automatically analyze the toll evasion big data.

[0052] 3. Construct an annotated dataset AV Dataset based on manually verified abnormal data. The dataset contains 6,000 numerical data, covering more than 20 fields such as vehicle color, license plate, vehicle type, and travel time. Among them, there are 5,000 abnormal evasion data, covering more than five types of evasion, which are used for MANet network model training; there are 1,000 normal data, which are used for model evaluation.

[0053] 4. For vehicles suspected of evading tolls, the MSRCR defogging module is designed for image enhancement. The Ve-RI model is constructed by combining the feature fusion strategy and feature distance measurement function to achieve cross-border vehicle head tracking, extract vehicle driving trajectories, and improve the evidence of vehicle evasion.

[0054] 5. Integrate the fee audit system with the audit work order initiation to achieve cross-platform docking and automated work order processing of audit work orders.

[0055] 6. Build a vehicle credit system, evaluate vehicle credit based on historical toll evasion data, adopt blacklist and whitelist management, and effectively regulate the driving behavior of vehicle owners.

[0056] The technical solution of the present invention can solve the following technical problems:

[0057] 1) Solve the problem of low accuracy (about 20%) and low efficiency in the verification of toll evasion in high-speed toll audit business;

[0058] 2) Solve the problem that the traditional audit algorithm is single and cannot cope with the endless types of evasion, and the evasion behavior is difficult to curb; 3) Solve the problem that the traditional high-speed toll audit business system is not sound, lacks the chain of evidence for evasion recovery, and cannot effectively recover evasion; 4) Solve the problem that the traditional toll audit system lacks a credit system and cannot effectively regulate illegal vehicle driving and driving regulations. 5) Solve the problem that the system is prone to collapse and difficult to maintain under the monolithic architecture.

[0059] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] Embodiment 1

[0061] The present invention discloses a highway toll audit system based on a microservice architecture, comprising: a data audit subsystem and a data application subsystem, wherein the data audit subsystem and the data application subsystem are different services, each service adopts a monolithic architecture, and the services are isolated from each other;

[0062] The data audit subsystem includes a data acquisition module and a data analysis module. The data acquisition module is used to collect data. The data analysis module is used to build a screening model for fare evasion vehicles, and obtain abnormal screening results of vehicles based on the data collected by the data acquisition module through the fare evasion vehicle screening model. The fare evasion vehicle screening model adopts an abnormal fare evasion classification network with a multi-attention mechanism, and the abnormal screening results of vehicles include abnormal types of suspicious vehicles and vehicle information;

[0063] The data application subsystem includes a suspected vehicle verification module, which is used to build a multi-branch vehicle re-identification model, and extract features of suspicious vehicles and vehicles in the search vehicle library based on the toll evasion vehicle screening model through the multi-branch vehicle re-identification model, and then classify the suspicious vehicles and vehicles in the search vehicle library based on feature distance to obtain views of the suspicious vehicles under different cameras.

[0064] Specifically, the microserver architecture divides the system into multiple services based on domains and functions. Each service adopts a monolithic architecture, and the services are isolated from each other, which greatly reduces the coupling between services and improves the maintainability and scalability of the system.

[0065] The data collection module of the data audit subsystem collects highway vehicle traffic data and transaction flow data; then the data analysis module completes the analysis and decision-making of the collected data, builds a toll-evading vehicle screening model, and realizes the automated analysis and screening of massive toll transaction data. The data application subsystem makes comprehensive applications of the data analyzed by the data audit subsystem, mainly including the suspected vehicle verification module, which constructs a multi-branch vehicle re-identification model and conducts a multi-lens verification based on the abnormal type of suspicious vehicles output by the toll-evading vehicle screening model to provide evidence for the prediction.

[0066] In one embodiment, a toll evasion vehicle screening model includes a data fusion module, a first feature extraction module and a classification module, wherein the data fusion module is used to preprocess the input data, the first feature extraction module performs feature extraction on the preprocessed data based on the self-attention ResNet backbone network, so that the model pays attention to the connection and hierarchical relationship between the three types of data to extract feature information related to toll evasion anomalies; the classification module predicts and classifies the results of feature extraction through a fully connected layer, and outputs the abnormal type of the suspicious vehicle, wherein the data collected by the data acquisition module is used as input data, including vehicle attribute information, toll station passage data and gantry data.

[0067] During the specific implementation process, the data acquisition module collects vehicle driving data obtained by the gantry and vehicle type recognition equipment, as well as the entrance and exit data of the highway toll station, to obtain the traffic record data and transaction flow data of the highway vehicles, and the data is stored in the non-relational database MongoDB.

[0068] The data audit subsystem integrates the data analysis module and establishes different screening rules for vehicles that evade fees according to different types of fee evasion. Specifically, a Multiple Attention Network for Vehicle Fee Evasion classification network model (MANet) is constructed based on deep learning. Figure 3As shown (the figure shows the screening results of three types of anomalies). Based on the abnormal evasion data that has been manually verified to be correct, a labeled data set (Abnormal Vehicle Dataset, referred to as A-V Dataset) is constructed, and the MANet network model is pre-trained using the AV Dataset. The data stored in MongoDB is input into the MANet network model for prediction, and the information of suspicious evasion vehicles (including vehicle attribute information and abnormal type) is output; the MANet abnormal evasion classification network model uses the Data Fuison module (data fusion module) to pre-process the input vehicle attribute information, toll station traffic data, and gantry data. The basic principle of DataFuison is as follows:

[0069]

[0070] in is the preprocessed feature tensor, t 1 ,t 2 is the feature tensor corresponding to two images randomly selected from the training set, and λ is a randomly generated weight coefficient.

[0071] Based on the self-attention ResNet backbone network, features are extracted from the preprocessed data, so that the model pays attention to the connection and hierarchical relationship between the three types of data, so as to better extract feature information related to the abnormal evasion of fees, and finally predict and classify the feature extraction results through the fully connected layer; different from the traditional ResNet backbone network, this embodiment designs a self-attention mechanism for the model, which can enable the model to learn more fine-grained semantic information.

[0072] In one implementation, the self-attention mechanism adopted by the first feature extraction module is:

[0073]

[0074] Among them, A(f,w) is the attention function of the toll evasion vehicle screening model, f and w are the independent variables of the attention function, and f k is the feature tensor corresponding to channel k in the last layer of the network, w k is the feature tensor f k The corresponding weight, d is the scaling factor, and C is the total number of channels.

[0075] In one embodiment, the abnormal types of suspicious vehicles include: large vehicle with small logo model abnormality, entry without exit path abnormality screening, U / J driving path abnormality, blocked passage medium abnormality and fake green pass abnormality. MANet can realize the screening of the above centralized abnormal types; the data audit subsystem should optimize data analysis performance and memory; the performance and memory optimization introduces a custom thread pool to store high-frequency access data in redis, thereby reducing audit processing time and memory usage.

[0076] U / J driving path abnormality refers to U-shaped driving path or J-shaped driving path. U-shaped driving means that the vehicle enters the highway from the same highway toll station, turns back halfway and finally returns to the same toll station to exit the highway. J-shaped driving means that the vehicle enters the highway from the A toll station, turns around at the highway gate after driving a certain distance and continues to drive in the reverse lane adjacent to the initial lane, and finally exits the highway at other stations other than the A toll station.

[0077] Furthermore, the data application subsystem integrates a suspected vehicle information display function, which can display big data of the suspected vehicle information (license plate, vehicle model, license plate color, passId, transaction serial number, evasion type, upper station name, lower station name, upper station time, lower station time, driving route, and whether evasion has been verified and confirmed) screened out by the data audit subsystem. The highway operating unit can conduct subsequent verification work on whether the suspected vehicle has evaded fees based on the information.

[0078] In one embodiment, a multi-branch vehicle re-identification model includes an image processing module, a second feature extraction module, a similarity measurement and reordering module, wherein the image processing module is used to perform dehazing processing on the image, the second feature extraction module is used to use a self-attention-based ResNet backbone network to extract features, and then use a linear fusion strategy to fuse the extracted global features and local features, the similarity measurement and reordering module is used to use a distance function to measure the similarity between the features of suspicious vehicle samples and the features of vehicle samples in the retrieved vehicle library, and reorder according to the results of the similarity measurement, and output views of the suspicious vehicle under different cameras.

[0079] Specifically, the data application subsystem integrates the suspicious vehicle verification function, which is implemented through the Ve-RI model to verify whether the suspected vehicle has evaded the fee. For suspected evading vehicles, a multi-branch vehicle re-identification model (Multi-Branch Model for Vehicle Re-Identification, referred to as Ve-RI model) is constructed to extract features from suspicious vehicles and search vehicle databases, and classify sample vehicles based on feature distances, thereby realizing cross-border vehicle tracking.

[0080] in, Figure 4 is a structural diagram of the Ve-RI model in the embodiment of the present invention. This figure shows the data processing flow of the model and the intermediate output results obtained, where X 1 ~X 4 Refers to the feature maps obtained from the first to fourth layers of the model. These feature maps need to be separately subjected to the subsequent pooling operation. 1 -P k Refers to the basic output of the entire model, which needs to be processed by maximum pooling. GMP and GAP represent maximum pooling and average pooling respectively. Represents the feature tensor of the feature map of the third layer above after maximum pooling, It represents the feature tensor of the feature map of the third layer after average pooling, and the others are similar. g refers to two pooling operations (maximum pooling and average pooling), F 0 Refers to the feature tensor of the upper left image (target vehicle) after being processed by the model, F 1-n Refers to the feature tensor of the lower left image (multiple filtered vehicle images) after being processed by the model, and the f below 1 Generally refers to the feature tensor output by the model.

[0081] In one implementation, the image processing module performs defogging on the image in the following manner:

[0082] MSRCR(x,y)=γ(x,y)×RMSR(x,y)+β (3)

[0083] Among them, MSRCR(x,y) represents the image after color restoration, x and y represent the horizontal and vertical coordinates of the input image, RMSR(x,y) represents the estimated value of the reflection component after multi-scale Retinex processing, γ is the scaling factor used to adjust the scale of the data, β is the bias term used to restore the original distribution characteristics of the data, and the formula for multi-scale Retinex processing is:

[0084] RMSR(x,y)=R(x,y)×L(x,y) (4)

[0085] Where R(x,y) represents the reflection component of the object, and L(x,y) represents the illumination component of the ambient light;

[0086] The linear fusion strategy adopted is:

[0087]

[0088] in represents the fused features, λ 1 ,λ 2 ∈[0,1] is a hyperparameter; f G is the global feature, f iis the local feature at position i, w i is the local feature f i The weight coefficient of , L represents the total number of local features.

[0089] Specifically, the reflection component of an object refers to the part that carries image detail information.

[0090] In one embodiment, a distance function is used to measure the similarity of features of different samples, and the samples are sorted from large to small based on the measurement results. The distance function is expressed as follows:

[0091]

[0092] Among them, D(f 0 ,f 1 ) represents the characteristic distance between different samples (suspicious vehicle samples (target samples) and retrieved vehicle database vehicle samples (samples to be compared)), f 0 and f 1 They represent the feature tensors of the target sample and the sample to be compared respectively, and δ∈[0,1] is a balancing hyperparameter.

[0093] Finally, the view of suspicious vehicles under non-overlapping cameras is output. The MSRCR defogging module (image processing module) improves the image quality by color balancing the color image to reduce the impact of bad weather on the feature extraction of vehicle images at high speeds; the feature fusion strategy performs weighted fusion of the vehicle's global feature information and the vehicle's local feature information (vehicle color, front, and logo) to better capture vehicle detail information and achieve high-precision feature extraction; the similarity measurement and reordering perform feature similarity measurement on suspicious vehicles and vehicles in the vehicle retrieval library to obtain views of suspicious vehicles under different cameras, extract vehicle driving trajectories based on the position of the camera, and improve the evidence of vehicle evasion. Based on the evidence of evasion, it is confirmed whether the vehicle has evaded fees, and then subsequent evasion recovery work is carried out.

[0094] In one embodiment, the data application subsystem also includes an automated fee evasion work order processing module for creating and initiating fee evasion work orders, thereby solving the cumbersome process of manually processing fee evasion work orders on the department audit platform and realizing an efficient and automated fee evasion recovery process.

[0095] In one embodiment, the data application subsystem also includes a vehicle credit system construction module and a toll evasion type frequency statistics module, wherein the credit system construction module is used to establish a credit database for highway vehicles and vehicle owners, and manage vehicles in blacklists and whitelists according to the number of toll evasions and toll evasion types; it plays a role in regulating the driving and driving norms of vehicle owners, thereby curbing toll evasion behavior. The toll evasion type frequency statistics module is used to count the frequency of various toll evasion types of vehicles, thereby effectively helping highway operators to lock down vehicles that frequently evade tolls, classify common toll evasion methods, and take corresponding containment measures to reduce toll losses.

[0096] In one embodiment, the data application subsystem includes a system management module for performing system authority management and verification, and authorizing and authenticating system user authority.

[0097] Specifically, system permission management includes user management, role management, menu management, user-role association, role-menu association, and providing a permission management interface to the outside world; system permission verification is performed by the system permission interception module on the token carried by the user when requesting the system function interface. If the permission information carried in the token is inconsistent with the permission required by the interface, the user access is intercepted to maintain the security of the system function. The system management module is shown in the attached figure. Figure 5 shown.

[0098] In one embodiment, the system also includes a security defense subsystem and a visualization display layer, wherein the security defense subsystem is used to maintain system security and resist external network attacks, and the visualization display layer provides users with data visualization and system operation visualization functions.

[0099] Specifically, the security defense subsystem is used to defend against common network attacks, such as CSRF attacks (cross-site request forgery attacks), DoS (denial of service attacks) and DDoS (distributed denial of service attacks), SQL injection attacks, and brute force password attacks; security defense specifically includes identifying and intercepting CSRF attacks, using gateway truncation and current limiting methods to defend against DoS and DDoS attacks, using parameterized queries to defend against SQL injection attacks, and using verification codes to defend against brute force password attacks. The system security defense subsystem is shown in the attached Figure 6 shown.

[0100] The visualization layer is used to provide visualization data display and visualization system operation functions, making it convenient for highway operation users to view audit information and perform business operations.

[0101] In addition, the highway toll audit system based on microservice architecture is deployed using Docker containerization technology, which can adapt to different operating system platforms and has good portability.

[0102] In one embodiment, the highway toll audit system can be composed of two single-structure systems, a data audit subsystem and a data application subsystem, which are separated into a toll audit service and a toll evasion recovery service. The toll audit service realizes the screening of suspicious vehicles, and the toll evasion recovery service realizes the functions of verifying suspicious vehicles, recovering fees, and processing work orders.

[0103] In other implementations, a security defense subsystem and a visual display layer can be added on the basis of the previous data audit subsystem and data application subsystem. The overall system structure is as follows: Figure 2 shown.

[0104] In general, the present invention can achieve the following beneficial technical effects:

[0105] 1. Based on the toll evasion vehicle screening model (MANet) of the data audit subsystem, a large number of toll evasion vehicles (≥10 9 The time cost is reduced by more than 70%, which greatly overcomes the difficulty of manual screening.

[0106] 2. The toll-evading vehicle screening model based on the data audit subsystem can greatly improve the accuracy of suspected vehicle screening. Compared with the manual audit method (accuracy of about 20%), the accuracy is increased to more than 60%.

[0107] 3. The toll evasion vehicle screening model of the data audit subsystem can cover the screening of five types of toll evasion and can screen 80% of the abnormal toll evasion data.

[0108] 4. Based on the data display and data application functions of the data application subsystem, the massive mixed evasion big data can be fully utilized, and the data availability can be improved by more than 70%.

[0109] 5. Based on the automated fare evasion work order processing of the data application subsystem, the work order for fee recovery of fare evasion vehicles can be initiated and processed quickly and efficiently, and the evidence chain of vehicle fare evasion can be provided to achieve efficient connection with the department's audit platform. Compared with manual work order processing (efficiency of about 50%), the efficiency of fee evasion recovery and audit work order processing is increased to 80%.

[0110] Embodiment 2

[0111] Based on the same inventive concept, this embodiment discloses a highway toll audit method based on a microservice architecture, which is implemented based on the system described in the first embodiment. The method includes:

[0112] The data collection module collects vehicle attribute information, toll station traffic data and gantry data;

[0113] A screening model for toll evasion vehicles is constructed through the data analysis module, and an abnormal screening result of the vehicle is obtained through the screening model for toll evasion vehicles based on the data collected by the data collection module, wherein the screening model for toll evasion vehicles adopts an abnormal toll evasion classification network with a multi-attention mechanism, and the abnormal screening result of the vehicle includes the abnormal type of the suspicious vehicle and the vehicle information;

[0114] A multi-branch vehicle re-identification model is constructed through the suspected vehicle verification module, and the multi-branch vehicle re-identification model is used to extract features of suspicious vehicles and vehicles in the retrieval vehicle library based on the toll evasion vehicle screening model. The suspicious vehicles and vehicles in the retrieval vehicle library are then classified based on feature distance to obtain views of the suspicious vehicles under different cameras.

[0115] Since the method introduced in the second embodiment of the present invention is based on the method adopted by the system in the first embodiment of the present invention, the specific implementation of the method can be understood by those skilled in the art based on the system introduced in the first embodiment of the present invention, so it will not be described in detail here. All audit methods implemented by the system in the first embodiment of the present invention belong to the scope of protection of the present invention.

[0116] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0117] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0118] Although preferred embodiments of the present invention have been described, additional changes and modifications may be made to these embodiments by those skilled in the art once the basic creative concepts are known. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the embodiments of the present invention without departing from the spirit and scope of the embodiments of the present invention. Thus, if these modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A highway toll audit system based on microservice architecture, characterized in that: include: Data audit subsystem and data application subsystem. The data audit subsystem and data application subsystem are different services. Each service adopts a monolithic architecture and the services are isolated from each other. The data audit subsystem includes a data acquisition module and a data analysis module. The data acquisition module is used to collect data. The data analysis module is used to build a screening model for fare evasion vehicles, and obtain abnormal screening results of vehicles based on the data collected by the data acquisition module through the fare evasion vehicle screening model. The fare evasion vehicle screening model adopts an abnormal fare evasion classification network with a multi-attention mechanism, and the abnormal screening results of vehicles include abnormal types of suspicious vehicles and vehicle information; The data application subsystem includes a suspected vehicle verification module, which is used to build a multi-branch vehicle re-identification model, and extract features of suspicious vehicles and vehicles in the search vehicle library based on the toll evasion vehicle screening model through the multi-branch vehicle re-identification model, and then classify the suspicious vehicles and vehicles in the search vehicle library based on feature distance to obtain views of the suspicious vehicles under different cameras.

2. The highway toll audit system based on microservice architecture as claimed in claim 1, characterized in that: The toll evasion vehicle screening model includes a data fusion module, a first feature extraction module and a classification module, wherein the data fusion module is used to preprocess the input data, and the first feature extraction module performs feature extraction on the preprocessed data based on the self-attention ResNet backbone network, so that the model pays attention to the connection and hierarchical relationship between the three types of data to extract feature information related to toll evasion anomalies; the classification module predicts and classifies the results of feature extraction through a fully connected layer, and outputs the abnormal type of the suspicious vehicle, wherein the data collected by the data acquisition module is used as input data, including vehicle attribute information, toll station passage data and gantry data.

3. The highway toll audit system based on microservice architecture as claimed in claim 1, characterized in that: The self-attention mechanism adopted by the first feature extraction module is: Among them, A(f,w) is the attention function of the toll evasion vehicle screening model, f and w are the independent variables of the attention function, and f k is the feature tensor corresponding to channel k in the last layer of the network, w k is the feature tensor f k The corresponding weight, d is the scaling factor, and C is the total number of channels.

4. The highway toll audit system based on microservice architecture as claimed in claim 1, characterized in that: The multi-branch vehicle re-identification model includes an image processing module, a second feature extraction module, a similarity measurement and reordering module, wherein the image processing module is used to perform dehazing processing on the image, the second feature extraction module is used to use a ResNet backbone network based on self-attention to extract features, and then use a linear fusion strategy to fuse the extracted global features and local features, the similarity measurement and reordering module is used to use a distance function to measure the similarity between the features of the suspicious vehicle samples and the features of the vehicle samples in the retrieved vehicle library, and reorder according to the results of the similarity measurement, and output the views of the suspicious vehicle under different cameras.

5. The highway toll audit system based on microservice architecture as claimed in claim 1, characterized in that: The image processing module performs defogging on the image in the following manner: MSRCR(x,y)=γ(x,y)×RMSR(x,y)+β Among them, MSRCR(x,y) represents the image after color restoration, x and y represent the horizontal and vertical coordinates of the input image, RMSR(x,y) represents the estimated value of the reflection component after multi-scale Retinex processing, γ is the scaling factor used to adjust the scale of the data, β is the bias term used to restore the original distribution characteristics of the data, and the formula for multi-scale Retinex processing is: RMSR(x,y)=R(x,y)×L(x,y) Where R(x,y) represents the reflection component of the object, and L(x,y) represents the illumination component of the ambient light; The linear fusion strategy adopted is: in represents the fused features, λ1,λ2∈[0,1] are hyperparameters; f G is the global feature, f i is the local feature at position i, w i is the local feature f i The weight coefficient of , L represents the total number of local features.

6. The highway toll audit system based on microservice architecture as claimed in claim 1, characterized in that: The data application subsystem also includes an automated fee evasion work order processing module for creating and initiating fee evasion work orders.

7. The highway toll audit system based on microservice architecture as claimed in claim 1, characterized in that: The data application subsystem also includes a vehicle credit system construction module and a toll evasion type frequency statistics module, wherein the credit system construction module is used to establish a credit database for highway vehicles and owners, and to manage vehicles in blacklists and whitelists according to the number of toll evasions and toll evasion types; the toll evasion type frequency statistics module is used to count the frequency of various toll evasion types of vehicles.

8. The highway toll audit system based on microservice architecture as claimed in claim 1, characterized in that: The data application subsystem includes a system management module, which is used to manage and verify system authority and to authorize and authenticate system user authority.

9. The highway toll audit system based on microservice architecture as claimed in claim 1, characterized in that: It also includes a security defense subsystem and a visualization display layer, wherein the security defense subsystem is used to maintain system security and resist external network attacks, and the visualization display layer provides users with data visualization and system operation visualization functions.

10. A highway toll audit method based on microservice architecture, characterized in that: Based on the system implementation according to any one of claims 1 to 9, the method comprises: The data collection module collects vehicle attribute information, toll station traffic data and gantry data; A screening model for toll evasion vehicles is constructed through the data analysis module, and an abnormal screening result of the vehicle is obtained through the screening model for toll evasion vehicles based on the data collected by the data collection module, wherein the screening model for toll evasion vehicles adopts an abnormal toll evasion classification network with a multi-attention mechanism, and the abnormal screening result of the vehicle includes the abnormal type of the suspicious vehicle and the vehicle information; A multi-branch vehicle re-identification model is constructed through the suspected vehicle verification module, and the multi-branch vehicle re-identification model is used to extract features of suspicious vehicles and vehicles in the retrieval vehicle library based on the toll evasion vehicle screening model. The suspicious vehicles and vehicles in the retrieval vehicle library are then classified based on feature distance to obtain views of the suspicious vehicles under different cameras.

Images