Traffic overrun early warning method and system under multi-source data fusion

Abstract

The invention discloses a traffic over-limit early warning method and system under multi-source data fusion, and relates to the related field of traffic monitoring, and the method comprises the steps: carrying out the multi-source data collection of a target vehicle through the interaction of a multi-source data linkage collection system, and obtaining a multi-source data set; performing key anchor point data screening on the multi-source data set by using a cross-modal attention mechanism, and determining a key anchor point data set and an associated multi-source data set; respectively carrying out interactive semantic integration analysis on the key anchor point data set and the associated multi-source data set, and determining multi-source integration interaction features; and pre-constructing a traffic over-limit multi-dimensional label library, carrying out matching identification on the multi-source integrated interaction features and the traffic over-limit multi-dimensional label library, and generating a traffic over-limit early warning instruction according to an identification result. The technical problems of low detection accuracy and high false alarm rate of the existing traffic over-limit early warning are solved, and the technical effects of improving the detection accuracy and reducing the false alarm rate are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of traffic monitoring, in particular to a traffic over-limit early warning method and system under multi-source data fusion. BACKGROUND

[0002] The detection and early warning of traffic over-limit behavior play a crucial role in ensuring road safety, maintaining traffic order, and prolonging the service life of roads. Currently, the main method to solve the problem of traffic over-limit detection relies on the monitoring of a single data source, such as detecting vehicle load only through weighing equipment or identifying vehicle type and speed only through cameras. However, these single data source methods have obvious limitations, such as the accuracy of weighing equipment decreasing in complex weather conditions or the recognition effect of cameras being poor in night or low light conditions. In addition, a single data source cannot fully reflect the over-limit situation of vehicles, which is prone to misjudgment or omission.

[0003] In the related art, the traffic over-limit early warning has the technical problems of low detection accuracy and high false positive rate. SUMMARY

[0004] The present application provides a traffic over-limit early warning method and system under multi-source data fusion, which uses an interactive multi-source data linkage collection system to collect data from all directions for target vehicles, forms a multi-source data set, uses a cross-modal attention mechanism to filter out a core key anchor data set and its related associated multi-source data set, processes these two types of data respectively through interactive semantic analysis technology, extracts multi-source integrated interactive features with comprehensive representation ability, intelligently matches the features with a pre-built traffic over-limit multi-dimensional label library, and generates traffic over-limit early warning instructions immediately when risk features are detected, etc. Technical means solve the technical problems of low detection accuracy and high false positive rate existing in the current traffic over-limit early warning, and achieve the technical effects of improving detection accuracy and reducing false positive rate.

[0005] The present application provides a traffic over-limit early warning method under multi-source data fusion, which includes: an interactive multi-source data linkage collection system collects multi-source data for target vehicles and obtains a multi-source data set; a cross-modal attention mechanism is used to filter key anchor data from the multi-source data set, determine a key anchor data set and an associated multi-source data set; interactive semantic integration analysis is performed on the key anchor data set and the associated multi-source data set respectively to determine multi-source integrated interactive features; a traffic over-limit multi-dimensional label library is pre-built, the multi-source integrated interactive features are matched with the traffic over-limit multi-dimensional label library for identification, and traffic over-limit early warning instructions are generated according to the identification results.

[0006] In a possible implementation, the interactive multi-source data linkage collection system collects multi-source data of a target vehicle, obtains a multi-source data set, and performs the following processing: the axle group type weighing and charging system in the multi-source data linkage collection system collects the axle group weight, total weight and vehicle speed of the target vehicle to obtain weighing data; the vehicle identification system in the multi-source data linkage collection system collects the vehicle axle type, wheel number and axle spacing of the target vehicle to obtain vehicle characteristic data; the environment monitoring system in the multi-source data linkage collection system collects the road state and weather characteristics of the target vehicle at the time of measurement to obtain environment data; and the weighing data, vehicle characteristic data and environment data are summarized to obtain the multi-source data set.

[0007] In a possible implementation, the multi-source data set is screened for key anchor data by using a cross-modal attention mechanism, a key anchor data set and an associated multi-source data set are determined, and the following processing is performed: the multi-source data set is embedded and encoded to be mapped to a feature space to generate a multi-source embedded representation set; the attention weights between each multi-source embedded representation in the multi-source embedded representation set and other multi-source embedded representations are calculated by the cross-modal attention mechanism to determine a multi-source embedded representation attention coefficient set; the multi-source data corresponding to the multi-source embedded representations in the first m positions in the multi-source embedded representation attention coefficient set are added to the key anchor data set; the key anchor data set is removed from the multi-source data set to obtain the associated multi-source data set.

[0008] In a possible implementation, the attention weights between each multi-source embedded representation in the multi-source embedded representation set and other multi-source embedded representations are calculated by the cross-modal attention mechanism to determine a multi-source embedded representation attention coefficient set, and the following processing is performed: the cosine similarity function in the cross-modal attention mechanism is extracted to calculate the representation similarity of each multi-source embedded representation in the multi-source embedded representation set and other multi-source embedded representations to obtain a representation similarity set; the softmax function in the cross-modal attention mechanism is extracted to convert the attention weights of the representation similarity set to obtain an attention weight set; each multi-source embedded representation in the multi-source embedded representation set is taken as an index to search the attention weight set, determine the attention weight set participated by each multi-source embedded representation and perform mean value calculation, and the calculation result is taken as the multi-source embedded representation attention coefficient to obtain the multi-source embedded representation attention coefficient set.

[0009] In a possible implementation, the interaction semantic integration analysis is respectively performed on the key anchor point data set and the associated multi-source data set, and multi-source integration interaction features are determined, and the following processing is performed: a first key anchor point data is randomly extracted from the key anchor point data set; interaction attention weights of the first key anchor point data and the key anchor point data set are respectively calculated to determine a first interaction attention weight set; interaction semantic analysis is performed based on the first interaction attention weight set and the key anchor point data set to determine key anchor point interaction features; interaction semantic analysis is performed on the associated multi-source data set to determine associated interaction features; and the key anchor point interaction features and the associated interaction features are integrated to obtain the multi-source integration interaction features.

[0010] In a possible implementation, the interaction semantic analysis is performed based on the first interaction attention weight set and the key anchor point data set to determine key anchor point interaction features, and the following processing is performed: a semantic analyzer is pre-constructed, wherein the semantic analyzer is obtained after a framework constructed according to a convolutional neural network is trained based on sample data; and the first interaction attention weight set and the key anchor point data set are identified by using the semantic analyzer to determine the key anchor point interaction features.

[0011] In a possible implementation, the interaction attention weights of the first key anchor point data and the key anchor point data set are respectively calculated to determine a first interaction attention weight set, and the following processing is performed: the similarity of the first key anchor point data and key anchor point data in the key anchor point data set is respectively calculated to obtain a first similarity set; each first similarity in the first similarity set is divided by the sum of the first similarity set to obtain the first interaction attention weight set.

[0012] In a possible implementation, a traffic over-limit multi-dimensional label library is pre-constructed, and the following processing is performed: a historical traffic over-limit data set is obtained; the historical traffic over-limit data set is extracted according to over-limit types, over-limit levels and over-limit behavior features to construct an initial over-limit label set; the initial over-limit label set is processed by set union, and the processed labels are added to an initially empty database to obtain the traffic over-limit multi-dimensional label library.

[0013] In a possible implementation, the following processing is performed: according to the traffic over-limit early warning instruction, a preset feedback window is obtained; static weight reweighing is performed on the target vehicle in the preset feedback window, and if the reweighing result meets the requirements, a passing instruction is obtained.

[0014] The application also provides a traffic overrun early warning system under multi-source data fusion, comprising: a multi-source data acquisition module, used for an interactive multi-source data linkage acquisition system, for multi-source data acquisition on a target vehicle, to obtain a multi-source data set; a key anchor point data screening module, used for key anchor point data screening on the multi-source data set by using a cross-modal attention mechanism, to determine a key anchor point data set and an associated multi-source data set; an interactive semantic analysis module, used for interactive semantic integration analysis on the key anchor point data set and the associated multi-source data set respectively, to determine a multi-source integrated interactive feature; and a traffic overrun early warning instruction generation module, used for pre-building a traffic overrun multi-dimensional label library, matching and identifying the multi-source integrated interactive feature with the traffic overrun multi-dimensional label library, and generating a traffic overrun early warning instruction according to an identification result.

[0015] The traffic overrun early warning method and system under multi-source data fusion provided by the application first acquire multi-source data of a target vehicle by using an interactive multi-source data linkage acquisition system, to obtain a multi-source data set, then screen key anchor point data from the multi-source data set by using a cross-modal attention mechanism, to determine a key anchor point data set and an associated multi-source data set, then perform interactive semantic integration analysis on the key anchor point data set and the associated multi-source data set respectively, to determine a multi-source integrated interactive feature, and finally pre-build a traffic overrun multi-dimensional label library, match and identify the multi-source integrated interactive feature with the traffic overrun multi-dimensional label library, and generate a traffic overrun early warning instruction according to an identification result. The technical effect of improving detection accuracy and reducing false positive rate is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings of the embodiments of the application will be briefly introduced below. In the present application, a flowchart is used to illustrate the operations performed by the system according to the embodiments of the application. It should be understood that the foregoing or the following operations are not necessarily performed in sequence. On the contrary, various steps can be processed in reverse order or simultaneously as needed. Meanwhile, other operations can be added to these processes, or a step or several steps can be removed from these processes.

[0017] Figure 1 The flowchart of the traffic overrun early warning method under multi-source data fusion provided by the embodiments of the application.

[0018] Figure 2 The structural diagram of the traffic overrun early warning system under multi-source data fusion provided by the embodiments of the application.

[0019] Explanation of reference signs: multi-source data acquisition module 10, key anchor point data screening module 20, interactive semantic analysis module 30, traffic overrun early warning instruction generation module 40. DETAILED DESCRIPTION

[0020] The above description is only a summary of the technical solutions of the present application. In order to make the technical means of the present application more clearly understood and implemented according to the content of the description, and in order to make the above and other purposes, characteristics and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described.

[0021] In order to make the purposes, technical solutions and advantages of the present application more clear, the present application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by those skilled in the art without making creative labor fall within the scope of protection of the present application.

[0022] In the following description, "some embodiments" are related to a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subset of all possible embodiments, and can be combined with each other without conflict. The term "first\second" is only to distinguish similar objects, and does not represent the specific order of the object. The terms "include" and "have" and any variations, are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or modules that are not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.

[0023] The embodiments of the present application provide a traffic over-limit early warning method under multi-source data fusion, as shown in Figure 1 The method comprises the following steps: Step S100, the interactive multi-source data linkage acquisition system acquires multi-source data of the target vehicle to obtain a multi-source data set.

[0024] Specifically, the multi-source data linkage collection system is a system that can simultaneously collect data from different sources (such as cameras, radars, sensors, etc.) and ensure consistency in time and space through linkage mechanism. Deploy multiple sensors on key traffic sections (such as highways, bridge entrances, etc.), including but not limited to high-definition cameras, laser radars, ground coils, millimeter wave radars, etc. Each sensor collects data on target vehicles according to the preset sampling frequency (such as 30 frames of image per second for high-definition cameras, 10 times of speed information per second for millimeter wave radars, etc.). The data collection module performs preliminary formatting processing on the collected data, such as converting image data to a unified pixel format and converting speed information to a unified unit. The formatted data is transmitted to the data processing center through a data transmission network (such as an optical fiber network, a wireless communication network, etc.), and time stamps and other synchronization information are attached in the data packet.

[0025] In one possible implementation, the interactive multi-source data linkage collection system collects multi-source data on target vehicles to obtain a multi-source data set, and step S100 further includes step S110, using the axle group type weighing toll collection system in the multi-source data linkage collection system to collect the vehicle axle group weight, total weight and speed of the target vehicle to obtain weighing data. Specifically, the axle group type weighing toll collection system installs pressure sensors or weighing sensor arrays on the road surface, and when a vehicle passes, the sensors can accurately measure the weight of each axle and the total weight of the vehicle. At the same time, combined with the time difference of the vehicle passing through the sensor, the speed of the vehicle is calculated. Specifically, the vehicle enters the detection area of the axle group type weighing toll collection system, and the pressure sensors or weighing sensor arrays measure the weight of each axle in real time and accumulate to obtain the total weight of the vehicle. The speed is calculated according to the time difference of the vehicle passing through the sensor. The weight and speed data collected by the sensor are preliminarily processed (such as filtering, calibration, etc.) by the data collection module, and then transmitted to the data processing center through wired or wireless communication. For example, the axle group type weighing toll collection system measures the weight of each axle of a truck as 10 tons, 12 tons and 8 tons, the total weight as 30 tons, and the speed as 80 km / h. These weighing data are transmitted to the data processing center through the data collection module.

[0026] Step S120, the vehicle identification system in the multi-source data linkage collection system collects the vehicle axle type, wheel number, and axle spacing of the target vehicle to obtain vehicle characteristic data. Specifically, the vehicle identification system includes a high-definition camera and an image recognition algorithm. The high-definition camera captures images of the vehicle, and the image recognition algorithm (such as a target detection model based on deep learning) analyzes the images to identify characteristics such as the vehicle's axle type, wheel number, and axle spacing. The identified vehicle characteristic data is formatted by the data collection module and transmitted to the data processing center. For example, the vehicle identification system captures an image of a truck and identifies through the image recognition algorithm that the vehicle is a three-axle vehicle, has 12 wheels, and the axle spacings are 3 meters and 4 meters respectively. The vehicle characteristic data is also transmitted to the data processing center.

[0027] Step S130, the environment monitoring system in the multi-source data linkage collection system collects the road state and weather characteristics of the target vehicle at the time of measurement to obtain environmental data. Specifically, the environment monitoring system includes road state sensors (such as road surface humidity sensors, temperature sensors, etc.) and weather monitoring equipment (such as weather stations, rain gauges, etc.). These devices can monitor the humidity, temperature, icing conditions of the road, and weather characteristics (such as rainfall, wind speed, etc.) in real time. The road state sensors and weather monitoring equipment collect real-time road state and weather characteristic data, and the data collection module formats the collected environmental data and transmits it to the data processing center. For example, the environment monitoring system detects that the current road state is dry, the temperature is 25°C, and the weather is sunny. The environmental data is also transmitted to the data processing center.

[0028] Step S140, the weighing data, vehicle characteristic data, and environmental data are summarized to obtain the multi-source data set. Specifically, in the data processing center, the data summary module receives data from the axle group type weighing toll collection system, the vehicle identification system, and the environment monitoring system. According to the time stamp or other synchronization information, different data are associated with the same target vehicle. The associated data are integrated into a multi-source data set for subsequent processing.

[0029] This implementation integrates weighing data, vehicle characteristic data, and environmental data, which can comprehensively reflect the running state of the target vehicle and the environment it is in. This not only includes the basic information of the vehicle such as weight and speed, but also includes the characteristics of the vehicle such as axle type and wheel number, as well as external factors such as road state and weather conditions. This comprehensiveness makes the traffic over-limit warning more accurate and reliable, avoiding misjudgment caused by a single data source. For example, in bad weather conditions (such as rainy and snowy weather), the road friction coefficient decreases and the vehicle's braking distance increases. By collecting environmental data, the system can consider the impact of these factors on vehicle operation, thus more accurately determining whether the vehicle has over-limit behavior.

[0030] Step S200, key anchor data screening is performed on the multi-source data set by using a cross-modal attention mechanism to determine a key anchor data set and an associated multi-source data set.

[0031] Specifically, the cross-modal attention mechanism is a deep learning-based mechanism that can simultaneously process data from different modalities (such as images, text, audio, etc.) and automatically learn the most critical information in different modalities for a specific task (such as traffic over-limit warning). The key information is filtered through attention weight distribution. The key anchor data set is the data set that is most critical to traffic over-limit warning in the multi-source data set, which is filtered by the cross-modal attention mechanism. These data contain the most important information for determining whether a vehicle is over-limit, such as vehicle speed, axle group weight, road condition, and axle spacing. The associated multi-source data set is the data set in the multi-source data set other than the key anchor data set. Although these data are not the most critical, they can provide auxiliary information in the interactive semantic integration analysis process, such as vehicle axle type and road icing conditions.

[0032] Specifically, a cross-modal attention model is constructed based on a deep learning framework (such as TensorFlow, PyTorch, etc.). The collected multi-source data is input into the cross-modal attention model, which extracts features from the multi-source data, such as using a convolutional neural network (CNN) to extract visual features from image data and using a recurrent neural network (RNN) to extract time series features from radar signals. The model calculates the attention weight of each modality data through the attention mechanism, and filters out the key anchor data according to the weight size. For example, if the license plate area in the vehicle image at a certain time occupies a high proportion in the attention weight, then the image data is determined as the key anchor data. The filtered key anchor data is stored separately as the key anchor data set, and the remaining data is stored as the associated multi-source data set.

[0033] In one possible implementation, the multi-source data set is screened for key anchor data using a cross-modal attention mechanism to determine a key anchor data set and an associated multi-source data set, and step S200 further includes step S210 of embedding coding the multi-source data set to map to a feature space to generate a multi-source embedding representation set. Specifically, a dedicated feature extraction model is designed for data of different modalities (such as weighing data, vehicle feature data, environmental data, etc.). For example, numerical feature extraction methods (such as standardization, normalization, etc.) are used to extract key features from weighing data (such as vehicle axle group weight, total weight, vehicle speed, etc.). An image recognition model (such as a convolutional neural network) is used to extract image features from vehicle feature data (such as vehicle axle type, number of wheels, axle spacing, etc.). Feature engineering methods are used to extract key features from environmental data (such as road conditions, weather features, etc.). The extracted features are mapped to a unified feature space through embedding coding to generate a multi-source embedding representation set, so that data of different modalities can be compared and fused in the same space.

[0034] Step S220, the attention weights between each multi-source embedding representation in the multi-source embedding representation set and other multi-source embedding representations are calculated respectively by a cross-modal attention mechanism to determine a multi-source embedding representation attention coefficient set. Specifically, a cross-modal attention model is constructed, which can use a self-attention mechanism (Self-Attention) or its variants (such as multi-head attention mechanism in Transformer architecture) to calculate the correlation between different modal embedding representations. The multi-source embedding representation set is input into the cross-modal attention model, and the model calculates the similarity (such as dot product similarity, cosine similarity, etc.) between each multi-source embedding representation and other multi-source embedding representations to generate an attention weight matrix. The attention weight reflects the importance of each multi-source embedding representation in the multi-source data. According to the attention weight matrix, the attention coefficient (i.e. the importance indicator of each multi-source embedding representation) of each multi-source embedding representation is determined. All attention coefficients are summarized to form a multi-source embedding representation attention coefficient set. For example, the model may calculate that the attention weight of the total weight is the highest, the axle spacing is the second, the vehicle speed is the third, and the weight of the environmental data is the lowest.

[0035] Step S230, add the multi-source data corresponding to the multi-source embedding representations with the top m attention coefficients in the multi-source embedding representation attention coefficient set into the key anchor data set. Specifically, sort the multi-source embedding representations according to the multi-source embedding representation attention coefficient set in descending order of attention coefficients. Select the top m multi-source embedding representations with the highest attention coefficients, map the selected multi-source embedding representations back to the original multi-source data, find the corresponding multi-source data, and add these original multi-source data to the key anchor data set. For example, m = 2, that is, select the embedding representations corresponding to the total weight and the axle spacing, and add their corresponding original data (such as total weight 30 tons, axle spacing 3 meters and 4 meters) to the key anchor data set.

[0036] Step S240, remove the key anchor data set from the multi-source data set to obtain the associated multi-source data set. Specifically, determine the data in the key anchor data set, remove these data from the multi-source data set, and the remaining data is the associated multi-source data set. For example, the vehicle speed (such as 80 km / h) and the environmental data (such as dry road, temperature 25℃, sunny day) form the associated multi-source data set.

[0037] This implementation can automatically identify the most critical information in multi-source data for traffic overload warning through the cross-modal attention mechanism. For example, if the total weight and the axle spacing of the vehicle are the most important factors in determining the overload behavior, the attention mechanism will give these data higher attention coefficients, so that they are selected as key anchor data. This precise selection improves the efficiency and accuracy of subsequent analysis. Embedding coding maps data of different modalities into the same feature space, enabling direct comparison of cross-modal data, further improving the performance of the traffic overload warning system. The separation of the key anchor data set and the associated multi-source data set enables the system to more clearly identify and process core data and auxiliary data. Key anchor data can provide direct decision basis, and associated multi-source data can be used as auxiliary information to further verify the warning result, thereby improving the reliability of the warning.

[0038] In one possible implementation, the cross-modal attention mechanism is used to calculate the attention weight between each multi-source embedding representation in the multi-source embedding representation set and other multi-source embedding representations to determine the multi-source embedding representation attention coefficient set, and step S220 further includes step S221, extracting the cosine similarity function in the cross-modal attention mechanism to calculate the representation similarity between each multi-source embedding representation in the multi-source embedding representation set and other multi-source embedding representations to obtain a representation similarity set. Specifically, the cosine similarity function is used to calculate the similarity between each multi-source embedding representation. Cosine similarity measures the similarity between two vectors by calculating the cosine of the angle between them. Suppose the multi-source embedding representation set is {E1, E2, …, En}, where En represents the embedding representation of the nth multi-source data, and the cosine similarity between two embedding representations Ei and Ej is calculated as follows: n}i is the i-th multi-source embedding representation. For each multi-source embedding representation E i , the cosine similarity between it and other multi-source embedding representations E j (j≠i) is calculated. All calculated cosine similarity values are stored in a representation similarity set. For example, assuming that the multi-source embedding representation set has 4 multi-source embedding representations {E1, E2, E3, E4}, the calculation results are shown in Table 1.

[0039] Table 1: Example of representation similarity set In step S222, the softmax function in the cross-modal attention mechanism is used to convert the attention weights of the representation similarity set, and an attention weight set is obtained. Specifically, the softmax function is used to convert the representation similarity into attention weights. The softmax function can convert a set of numerical values into a probability distribution, and the softmax function is applied to each similarity value in the representation similarity set. The converted values are stored in the attention weight set, and an example of the data is shown in Table 2.

[0040] Table 2: Example of attention weight set In step S223, each multi-source embedding representation in the multi-source embedding representation set is used as an index to retrieve the attention weight set, determine the attention weight set participated by each multi-source embedding representation, and calculate the mean value, and the calculation result is used as the multi-source embedding representation attention coefficient, and a multi-source embedding representation attention coefficient set is obtained. Specifically, for each multi-source embedding representation E i , all weight values participating in the attention weight set are retrieved, and the mean value of these weight values is calculated, and the mean value is used as the attention coefficient of the multi-source embedding representation. The attention coefficients of all multi-source embedding representations are stored in the attention coefficient set, and an example of the data is shown in Table 3.

[0041] Table 3: Example of multi-source embedding representation attention coefficient set This implementation can accurately measure the correlation between each multi-source embedding representation and other multi-source embedding representations through the combination of cosine similarity and softmax function, and convert it into attention weights. This enables the system to more accurately identify the data that is most critical for traffic over-limit warning. The softmax function converts the similarity values into a probability distribution, ensuring the normalization of the attention weights and avoiding the deviation of the weights caused by the large difference in similarity values. This processing method enhances the reliability and stability of data fusion.

[0042] Step S300, respectively, on the key anchor point data set and the associated multi-source data set are analyzed by interactive semantic integration, determine multi-source integration interaction features.

[0043] Specifically, interactive semantic integration analysis refers to the analysis and fusion of data from different sources at the semantic level, and extracts interactive features that can reflect the state and behavior of the vehicle. Multi-source integrated interaction features are features obtained through interactive semantic integration analysis, which integrate semantic information from different modal data and can fully reflect the state and behavior of the vehicle, providing a basis for subsequent matching recognition.

[0044] Specifically, a semantic analysis model is constructed for the key anchor point data set and the associated multi-source data set. For example, for image data, a target detection model (such as YOLO, Faster R-CNN, etc.) can be used to identify the type of vehicle, license plate information, and other semantic information; for radar signal data, signal processing algorithms can be used to extract the speed, acceleration, and other semantic information of the vehicle. The semantic analysis results of the key anchor point data set and the associated multi-source data set are integrated interactively to extract multi-source integrated interaction features. The integration method can be simple concatenation or more complex feature fusion methods, such as weighted summation or feature embedding. In order to improve the efficiency and accuracy of subsequent matching recognition, feature optimization algorithms (such as principal component analysis PCA, etc.) can be used to optimize and reduce the dimension of the extracted multi-source integrated interaction features, extracting the most important feature dimensions.

[0045] In one possible implementation, respectively, on the key anchor point data set and the associated multi-source data set are analyzed by interactive semantic integration, determine multi-source integration interaction features, step S300 further includes step S310, randomly extract a first key anchor point data from the key anchor point data set. Specifically, a random sampling algorithm is used to randomly select a data from the key anchor point data set as the "first key anchor point data". This randomness can increase the robustness of the system and prevent bias caused by fixed selection of a certain data. Assuming that the key anchor point data set is {E1, E2}, one of the data is randomly selected, for example, E1, as the first key anchor point data.

[0046] Step S320, respectively, the first key anchor point data and the key anchor point data set are calculated by interactive attention weight, determine the first interactive attention weight set. Specifically, the attention mechanism is used to calculate the interactive attention weight between the first key anchor point data and other key anchor point data. For example, the similarity (such as cosine similarity) between them can be calculated and the softmax function is applied to achieve this.

[0047] Step S330, based on the first set of interaction attention weights and the set of key anchor data, interaction semantic analysis is performed to determine the key anchor interaction feature. Specifically, through the interaction relationship between the first key anchor data and other data in the set of key anchor data, the fusion feature reflecting the core state of the vehicle (such as the risk of over-limit) is extracted. The interaction attention weight quantifies the semantic association strength between different data, and the semantic analysis combines these weights with the original data to generate more discriminative features. Specifically, according to the first set of interaction attention weights, the data in the set of key anchor data is weighted and summed. For example, assuming that the set of key anchor data is {E1, E2}, the first key anchor data is E1, and the interaction attention weight between E1 and E2 is α 12 . The key anchor interaction feature F key may be represented as: F key = α 11 ·E1 + α 12 ·E2. Wherein, α 11 is the weight of E1 with itself (the remaining weight after normalization). The weighted result is input into a multi-layer perception (MLP) or neural network to further extract high-order interaction features through a nonlinear activation function (such as ReLU).

[0048] Step S340, the set of associated multi-source data is subjected to interaction semantic analysis to determine the associated interaction feature. Specifically, the same analysis method as the set of key anchor data is used to perform interaction semantic analysis on the set of associated multi-source data to generate the associated interaction feature reflecting the environment, vehicle speed and other auxiliary information.

[0049] Step S350, the key anchor interaction feature and the associated interaction feature are integrated to obtain the multi-source integrated interaction feature. Specifically, the key anchor interaction feature and the associated interaction feature are spliced to generate the multi-source integrated interaction feature.

[0050] This implementation can dynamically emphasize the relevance between key anchor data by randomly selecting the first key anchor data and calculating the interaction attention weight, thereby enhancing the representativeness of the feature. Splicing the key anchor interaction feature and the associated interaction feature not only retains all important information, but also simplifies the feature representation, facilitating subsequent processing and analysis.

[0051] In a possible implementation, the interaction attention weight of the first key anchor data and the set of key anchor data is calculated respectively, and a first set of interaction attention weights is determined, and step S320 further includes step S321: the similarity between the first key anchor data and the key anchor data in the set of key anchor data is calculated respectively to obtain a first set of similarities. Specifically, the first key anchor data and the set of key anchor data are determined, and for other key anchor data in the set of key anchor data, the cosine similarity between the first key anchor data and the other key anchor data is calculated, and all the calculated similarity values are stored in the first set of similarities.

[0052] Step S322: each first similarity in the first set of similarities is divided by the sum of the first set of similarities to obtain the first set of interaction attention weights. Specifically, the sum of all similarity values in the first set of similarities is calculated, each similarity value is divided by the sum to obtain a normalized attention weight. All normalized attention weights are stored in the first set of interaction attention weights. This implementation reflects the relative importance between the first key anchor data and other key anchor data by normalizing the similarity values into attention weights. In this way, the system can dynamically adjust the attention degree to different data, so as to more effectively capture key information.

[0053] In a possible implementation, the first set of interaction attention weights and the set of key anchor data are used for interaction semantic analysis to determine key anchor interaction features, and step S330 further includes step S331: a pre-constructed semantic analyzer is used, wherein the semantic analyzer is obtained by training a framework constructed according to a convolutional neural network based on sample data. Specifically, a semantic analyzer is constructed using a convolutional neural network (CNN) framework. CNN can automatically extract features in data and learn semantic relationships between data through structures such as convolutional layers, pooling layers and fully connected layers. The semantic analyzer is trained using labeled sample data. The sample data can include vehicle weighing data, vehicle feature data, environmental data, etc., and the label information can be labels such as whether the vehicle is overloaded or speeding. Through training, the semantic analyzer can learn the semantic relationships between these data, for example: the relationship between the total weight of the vehicle and the overload state, the relationship between the type of the vehicle and the carrying capacity, the influence of environmental conditions on vehicle driving, etc.

[0054] Step S332, the semantic analyzer is used to identify the first interaction attention weight set and the key anchor data set, and determine the key anchor interaction feature. Specifically, the first interaction attention weight set and the key anchor data set are input into the pre-constructed semantic analyzer as input features. The semantic analyzer processes the input data through convolution layers, pooling layers, and fully connected layers, etc., and extracts interaction features that can reflect the semantic relationship between the data. The key anchor interaction feature output by the semantic analyzer can be a vector, representing the semantic association between the data. In this implementation, the attention weight enables the semantic analyzer to dynamically adjust the attention degree to different data, thereby more effectively capturing key information.

[0055] Step S400, pre-constructing a traffic over-limit multi-dimensional label library, matching and identifying the multi-source integrated interaction features with the traffic over-limit multi-dimensional label library, and generating a traffic over-limit early warning instruction according to the identification result.

[0056] Specifically, the traffic over-limit multi-dimensional label library is a pre-constructed database that contains feature labels of various traffic over-limit situations. These labels describe the standards and characteristics of over-limit behavior from multiple dimensions (such as load, speed, vehicle type, etc.), which are used for matching and identifying with the actual detected vehicle features. Matching and identification is a process of judging whether the current vehicle conforms to the over-limit behavior described in the traffic over-limit multi-dimensional label library by calculating the similarity or distance between features and labels. The traffic over-limit early warning instruction is generated according to the matching and identification result, which is used to notify the traffic management department or the driver that the vehicle has over-limit behavior. The instruction contains the license plate number, over-limit type, over-limit degree, and other information of the vehicle.

[0057] Specifically, according to the relevant regulations and standards of traffic over-limit, the traffic over-limit multi-dimensional label library is constructed. The label library contains feature labels of various over-limit situations, such as overloading weight range, overspeed speed range, vehicle type restriction, and other multi-dimensional information. Machine learning or deep learning algorithms (such as support vector machine SVM, neural network, etc.) are used to match and identify the multi-source integrated interaction features with the traffic over-limit multi-dimensional label library, and to judge whether the current vehicle is in an over-limit state by calculating the similarity or distance between features and labels. According to the matching and identification result, if it is judged that the vehicle has over-limit behavior, the corresponding traffic over-limit early warning instruction is generated. The early warning instruction can include the license plate number, over-limit type, over-limit degree, and other information of the vehicle, and is sent to the traffic management department or related equipment (such as electronic display screen, etc.) through the communication network.

[0058] For example, if the multi-source integrated interaction features of a vehicle are "license plate number XXX, load weight 50 tons, speed 130 km / h", and the traffic overload multi-dimensional label library stipulates that the upper limit of overload weight on this section is 49 tons and the upper limit of speed is 120 km / h, then the matching degree of the features of the vehicle with the overload and overspeed labels is calculated, and it is judged that the vehicle has both overload and overspeed behaviors. According to the matching result, a traffic overload warning instruction is generated, such as "the vehicle with license plate number XXX is overloaded by 1 ton and overspeeds by 10 km / h”, and the instruction is sent to the monitoring system of the traffic management department or the electronic display screen on the roadside through the communication network, reminding the law enforcement personnel and the driver.

[0059] In a possible implementation, the traffic overload multi-dimensional label library is pre-constructed, and step S400 further includes step S410 of acquiring a historical traffic overload data set. Specifically, historical traffic overload records are acquired from the traffic management department, including information such as license plate number, load weight, vehicle speed, and overload type of the vehicle. Image data and video data of the vehicle are acquired from the monitoring system for extracting vehicle features. Weighing data and vehicle speed data of the vehicle are acquired from the toll station. The collected data is sorted to remove duplicate or incorrect data, and a complete historical traffic overload data set is formed.

[0060] Step S420, data extraction is performed on the historical traffic overload data set according to the overload type, overload level, and overload behavior features, and an initial overload label set is constructed. Specifically, the historical traffic overload data is classified according to the overload type, for example, the overload data and the overspeed data are separated. Under each overload type, further classification is performed according to the overload level, for example, the overload data is classified into first-level overload and second-level overload. Under each overload level, further classification is performed according to the overload behavior features, for example, the first-level overload data is classified into overload by 5 tons and overload by 7 tons. The label information in each classification is extracted to form an initial overload label set.

[0061] Step S430, the initial overload label set is processed by set union, and the processed labels are added to an initially empty database to obtain a traffic overload multi-dimensional label library. Specifically, all labels in the initial overload label set are processed by set union to remove duplicate labels and ensure the uniqueness of the labels. An initially empty database is created for storing the traffic overload multi-dimensional label library. The processed labels are added to the database one by one to form a complete traffic overload multi-dimensional label library.

[0062] This approach collects and organizes historical traffic overload data to construct a multidimensional traffic overload tag library that comprehensively covers various overload types, levels, and behavioral characteristics, providing rich reference information for subsequent traffic overload warnings. By removing duplicate tags through union processing, the uniqueness and accuracy of the tags in the multidimensional traffic overload tag library are ensured. This enables the system to more accurately determine whether a vehicle is overloaded during matching and identification.

[0063] In a possible implementation, the method further includes: obtaining a preset feedback window according to the traffic overload warning instruction; reweighing the static weight of the target vehicle within the preset feedback window, and obtaining a pass instruction if the reweighing result meets the requirements.

[0064] Specifically, a time window is set based on actual needs, within which the static weight reweighing of the target vehicle is completed. This preset feedback window can be dynamically adjusted based on factors such as traffic flow and road conditions. After the system generates a traffic overload warning instruction, it immediately notifies traffic management personnel or automatically directs the vehicle to a static weighing area. Static weighing equipment can provide highly accurate weight measurement results. Within the preset feedback window (e.g., 10 minutes), the static weight reweighing of the target vehicle is completed. The reweighing results, including the vehicle's gross weight and overload status, are recorded. The reweighed results are compared with preset load standards to determine whether they meet the requirements. If the reweighed results meet the requirements, the system generates a pass order, allowing the vehicle to continue traveling. If not, further processing instructions are generated (e.g., requiring the vehicle to unload any overloaded parts). This implementation verifies the accuracy of the system-generated traffic overload warning instructions through static weight reweighing, reducing the possibility of misjudgment.

[0065] The embodiment of the present application adopts an interactive multi-source data linkage collection system to collect all-round data of the target vehicle to form a multi-source data set, adopts a cross-modal attention mechanism to screen out the core key anchor point data set and its related associated multi-source data set, and processes these two types of data separately through interactive semantic analysis technology to extract multi-source integrated interactive features with comprehensive characterization capabilities, and intelligently match the features with a pre-built traffic overload multi-dimensional label library. When risk features are detected, a traffic overload warning instruction is immediately generated. Such technical means solve the technical problems of low detection accuracy and high false alarm rate in existing traffic overload warnings, and achieve the technical effect of improving detection accuracy and reducing false alarm rate.

[0066] In the above, refer to Figure 1 The traffic overload warning method based on multi-source data fusion according to an embodiment of the present invention is described in detail. Figure 2 A traffic overload warning system based on multi-source data fusion according to an embodiment of the present invention is described.

[0067] The traffic over-limit early warning system under multi-source data fusion according to the embodiment of the application is used to solve the technical problem of low detection accuracy and high false alarm rate existing in the current traffic over-limit early warning, and achieves the technical effects of improving detection accuracy and reducing false alarm rate. The traffic over-limit early warning system under multi-source data fusion comprises a multi-source data acquisition module 10, a key anchor point data screening module 20, an interactive semantic analysis module 30, and a traffic over-limit early warning instruction generation module 40.

[0068] The multi-source data acquisition module 10 is used to acquire multi-source data of a target vehicle by using an interactive multi-source data linkage acquisition system, and obtain a multi-source data set. The key anchor point data screening module 20 is used to screen key anchor point data from the multi-source data set by using a cross-modal attention mechanism, determine a key anchor point data set and an associated multi-source data set. The interactive semantic analysis module 30 is used to respectively perform interactive semantic integration analysis on the key anchor point data set and the associated multi-source data set, and determine a multi-source integrated interactive feature. The traffic over-limit early warning instruction generation module 40 is used to pre-construct a traffic over-limit multi-dimensional label library, match and identify the multi-source integrated interactive feature with the traffic over-limit multi-dimensional label library, and generate a traffic over-limit early warning instruction according to the identification result.

[0069] In the following, the specific configuration of the multi-source data acquisition module 10 will be described in detail. As described above, the interactive multi-source data linkage acquisition system acquires multi-source data of a target vehicle and obtains a multi-source data set. The multi-source data acquisition module 10 can further comprise a weighing data acquisition unit for acquiring vehicle axle group weight, total weight and vehicle speed of the target vehicle by using an axle group type weighing and charging system in the multi-source data linkage acquisition system, and obtaining weighing data; a vehicle feature data acquisition unit for acquiring vehicle axle type, wheel number and axle spacing of the target vehicle by using a vehicle recognition system in the multi-source data linkage acquisition system, and obtaining vehicle feature data; an environmental data acquisition unit for acquiring road state and weather characteristics of the target vehicle at the time of measurement by using an environmental monitoring system in the multi-source data linkage acquisition system, and obtaining environmental data; and a data summarizing unit for summarizing the weighing data, vehicle feature data and environmental data, and obtaining the multi-source data set.

[0070] In the following, the specific configuration of the key anchor data screening module 20 will be described in detail. As described above, the key anchor data screening is performed on the multi-source data set by using the cross-modal attention mechanism to determine the key anchor data set and the associated multi-source data set, and the key anchor data screening module 20 can further include: an embedded coding unit for embedding and coding the multi-source data set to map to a feature space to generate a multi-source embedded representation set; an attention weight calculation unit for calculating the attention weight between each multi-source embedded representation in the multi-source embedded representation set and other multi-source embedded representations by using the cross-modal attention mechanism to determine a multi-source embedded representation attention coefficient set; a key anchor data acquisition unit for adding the multi-source data corresponding to the multi-source embedded representation located in the top m positions in the multi-source embedded representation attention coefficient set into the key anchor data set; and an associated multi-source data acquisition unit for removing the key anchor data set from the multi-source data set to obtain an associated multi-source data set.

[0071] In the above, the attention weight between each multi-source embedded representation in the multi-source embedded representation set and other multi-source embedded representations is calculated by using the cross-modal attention mechanism to determine the multi-source embedded representation attention coefficient set, and the attention weight calculation unit can further include: a representation similarity calculation subunit for extracting the cosine similarity function in the cross-modal attention mechanism to calculate the representation similarity between each multi-source embedded representation in the multi-source embedded representation set and other multi-source embedded representations to obtain a representation similarity set; an attention weight conversion subunit for extracting the softmax function in the cross-modal attention mechanism to convert the attention weight of the representation similarity set to obtain an attention weight set; and a mean value calculation subunit for taking each multi-source embedded representation in the multi-source embedded representation set as an index to retrieve the attention weight set to determine the attention weight set participated by each multi-source embedded representation and perform mean value calculation, and taking the calculation result as the multi-source embedded representation attention coefficient to obtain the multi-source embedded representation attention coefficient set.

[0072] Hereinafter, the specific configuration of the interactive semantic analysis module 30 will be described in detail. As described above, the interactive semantic integration analysis is performed on the set of key anchor data and the set of associated multi-source data respectively to determine the multi-source integrated interaction feature, and the interactive semantic analysis module 30 can further include: a first key anchor data extraction unit for randomly extracting first key anchor data from the set of key anchor data; an interactive attention weight calculation unit for calculating the interactive attention weight of the first key anchor data and the set of key anchor data respectively to determine a first set of interactive attention weights; a first interactive semantic analysis unit for performing interactive semantic analysis based on the first set of interactive attention weights and the set of key anchor data to determine key anchor interaction features; a second interactive semantic analysis unit for performing interactive semantic analysis on the set of associated multi-source data to determine associated interaction features; and an interactive feature integration unit for integrating the key anchor interaction features and the associated interaction features to obtain the multi-source integrated interaction feature.

[0073] In the interactive semantic analysis based on the first set of interactive attention weights and the set of key anchor data to determine the key anchor interaction features, the first interactive semantic analysis unit can further include: a semantic analyzer construction subunit for pre-constructing a semantic analyzer, wherein the semantic analyzer is obtained by training a framework constructed according to a convolutional neural network based on sample data; and a semantic analysis and identification subunit for identifying the first set of interactive attention weights and the set of key anchor data using the semantic analyzer to determine the key anchor interaction features.

[0074] In the calculation of the interactive attention weight of the first key anchor data and the set of key anchor data respectively to determine the first set of interactive attention weights, the interactive attention weight calculation unit can further include: a similarity calculation subunit for calculating the similarity of the first key anchor data and the key anchor data in the set of key anchor data respectively to obtain a first set of similarities; and a first set of interactive attention weight acquisition subunit for dividing each first similarity in the first set of similarities by the sum of the first set of similarities to obtain the first set of interactive attention weights.

[0075] In the following, the specific configuration of the traffic over-limit early warning instruction generation module 40 will be described in detail. As described above, the pre-constructed traffic over-limit multi-dimensional label library, the traffic over-limit early warning instruction generation module 40 can further include: a historical traffic over-limit data set acquisition unit for acquiring a historical traffic over-limit data set; a data extraction unit for data extraction on the historical traffic over-limit data set according to the over-limit type, the over-limit level and the over-limit behavior characteristics, and constructing an initial over-limit label set; and a traffic over-limit multi-dimensional label library construction unit for set processing on the initial over-limit label set, adding the processed label into the initially empty database, and obtaining the traffic over-limit multi-dimensional label library.

[0076] The system can further include: a preset feedback window acquisition module for acquiring a preset feedback window according to the traffic over-limit early warning instruction; and a static weight re-measuring module for static weight re-measuring of the target vehicle within the preset feedback window, and obtaining a passing instruction if the re-measuring result meets the requirements.

[0077] The traffic over-limit early warning system under multi-source data fusion provided by the embodiments of the present application can execute the traffic over-limit early warning method under multi-source data fusion provided by any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0078] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server, and the various units and modules are only divided according to the functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy mutual differentiation, and do not limit the protection scope of the present application.

[0079] The above specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, 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 the present application shall be included in the protection scope of the present application. In some cases, the actions or steps described in the present application can be executed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

Claims

1. Traffic overload warning method based on multi-source data fusion, characterized by: The method comprises: Interactive multi-source data linkage acquisition system collects multi-source data from target vehicles and obtains multi-source data sets; Using a cross-modal attention mechanism to screen key anchor data for the multi-source data set, and determine a key anchor data set and an associated multi-source data set; Performing interactive semantic integration analysis on the key anchor point data set and the associated multi-source data set respectively to determine multi-source integration interaction features; A traffic overload multi-dimensional tag library is pre-built, the multi-source integrated interactive features are matched and identified with the traffic overload multi-dimensional tag library, and a traffic overload warning instruction is generated according to the identification result.

2. The traffic overload warning method based on multi-source data fusion according to claim 1 is characterized in that: The interactive multi-source data linkage acquisition system collects multi-source data from the target vehicle and obtains a multi-source data set, including: The axle group weight charging system in the multi-source data linkage acquisition system is used to collect the axle group weight, total weight and vehicle speed of the target vehicle to obtain weighing data; Using the vehicle identification system in the multi-source data linkage acquisition system to collect the vehicle axle type, number of wheels, and axle spacing of the target vehicle to obtain vehicle feature data; Using the environmental monitoring system in the multi-source data linkage acquisition system to collect the road conditions and weather characteristics of the target vehicle during measurement to obtain environmental data; The weighing data, vehicle characteristic data and environmental data are aggregated to obtain the multi-source data set.

3. The traffic overload warning method based on multi-source data fusion according to claim 1 is characterized in that: The multi-source data set is screened for key anchor data using a cross-modal attention mechanism to determine a key anchor data set and an associated multi-source data set, including: Embedding the multi-source data set, mapping it to the feature space, and generating a multi-source embedding representation set; Calculating the attention weights between each multi-source embedding representation and other multi-source embedding representations in the multi-source embedding representation set by a cross-modal attention mechanism, and determining a multi-source embedding representation attention coefficient set; Add the multi-source data corresponding to the multi-source embedding representation located at the first m positions in the multi-source embedding representation attention coefficient set to the key anchor point data set; The key anchor point data set is removed from the multi-source data set to obtain a related multi-source data set.

4. The traffic overload warning method based on multi-source data fusion as claimed in claim 3 is characterized in that: The attention weights between each multi-source embedding representation and other multi-source embedding representations in the multi-source embedding representation set are calculated respectively by a cross-modal attention mechanism to determine a multi-source embedding representation attention coefficient set, including: Extracting the cosine similarity function in the cross-modal attention mechanism and respectively calculating the representation similarity between each multi-source embedding representation and other multi-source embedding representations in the multi-source embedding representation set to obtain a representation similarity set; Extracting the softmax function in the cross-modal attention mechanism to perform attention weight conversion on the representation similarity set to obtain an attention weight set; Taking each multi-source embedding representation in the multi-source embedding representation set as an index, the attention weight set is retrieved, the attention weight set in which each multi-source embedding representation participates is determined and the mean is calculated, and the calculation result is used as the multi-source embedding representation attention coefficient to obtain the multi-source embedding representation attention coefficient set.

5. The traffic overload warning method based on multi-source data fusion according to claim 1 is characterized in that: Performing interactive semantic integration analysis on the key anchor point data set and the associated multi-source data set respectively to determine multi-source integration interaction features, including: Randomly extracting first key anchor point data from the key anchor point data set; Calculating the interaction attention weights of the first key anchor point data and the key anchor point data set respectively, and determining a first interaction attention weight set; Performing interaction semantic analysis based on the first interaction attention weight set and the key anchor point data set to determine key anchor point interaction features; Performing interactive semantic analysis on the associated multi-source data set to determine associated interactive features; The key anchor point interaction feature and the associated interaction feature are integrated to obtain the multi-source integrated interaction feature.

6. The traffic overload warning method based on multi-source data fusion according to claim 5 is characterized in that: Performing interaction semantic analysis based on the first interaction attention weight set and the key anchor point data set to determine key anchor point interaction features includes: A pre-built semantic analyzer, wherein the semantic analyzer is obtained by training a framework constructed according to a convolutional neural network based on sample data; The semantic analyzer is used to identify the first interaction attention weight set and the key anchor point data set to determine the key anchor point interaction features.

7. The traffic overload warning method based on multi-source data fusion according to claim 6 is characterized in that: Calculating the interaction attention weights of the first key anchor point data and the key anchor point data set respectively, and determining a first interaction attention weight set, including: Calculating similarities between the first key anchor point data and key anchor point data in the key anchor point data set respectively to obtain a first similarity set; Each first similarity in the first similarity set is compared with the sum of the first similarity sets to obtain a first interactive attention weight set.

8. The traffic overload warning method based on multi-source data fusion according to claim 1 is characterized in that: Pre-built traffic overload multi-dimensional label library, including: Obtain historical traffic overload data collection; Extracting data from the historical traffic overload data set according to overload type, overload level, and overload behavior characteristics to construct an initial overload label set; The initial overload label set is subjected to a union process, and the processed labels are added into an initially empty database to obtain a traffic overload multi-dimensional label library.

9. The traffic overload warning method based on multi-source data fusion according to claim 1 is characterized in that: include: Obtaining a preset feedback window according to the traffic overload warning instruction; The target vehicle is statically reweighed within the preset feedback window, and a pass instruction is obtained if the reweighing result meets the requirements.

10. Traffic overload warning system based on multi-source data fusion is characterized by: The system is used to implement the traffic overload warning method based on multi-source data fusion according to any one of claims 1 to 9, and the system includes: Multi-source data acquisition module, used for interactive multi-source data linkage acquisition system, to collect multi-source data of target vehicles and obtain multi-source data sets; A key anchor data screening module is used to screen the multi-source data set for key anchor data using a cross-modal attention mechanism to determine the key anchor data set and the associated multi-source data set; An interactive semantic analysis module, configured to perform interactive semantic integration analysis on the key anchor point data set and the associated multi-source data set, respectively, to determine multi-source integration interactive features; The traffic overload warning instruction generation module is used to pre-build a traffic overload multi-dimensional label library, match and identify the multi-source integrated interactive features with the traffic overload multi-dimensional label library, and generate a traffic overload warning instruction according to the identification result.

Images