Roadside digital road condition monitoring method and system based on deep learning

Abstract

The present invention discloses a roadside digital road condition monitoring method and system based on deep learning, comprising: firstly obtaining a road monitoring vibration signal collected by a vibration sensor, and using a trained road condition analysis model to determine the current road state. At the same time, combining the monitoring images collected by the camera, a road monitoring event to be processed is generated. Finally, the maintenance strategy for a specific road section is determined based on these events. With such a design, real-time monitoring of road conditions and intelligent maintenance decision-making are realized through deep learning technology, thereby improving the efficiency and accuracy of road maintenance.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a roadside digital road condition monitoring method and system based on deep learning. Background Art

[0002] With the development of intelligent transportation systems and digital cities, real-time monitoring and maintenance of road conditions have become crucial. Traditional road monitoring methods rely on manual inspections, which are inefficient and difficult to detect and deal with road problems in a timely manner. Summary of the invention

[0003] The purpose of the present invention is to provide a roadside digital road condition monitoring method and system based on deep learning.

[0004] In a first aspect, an embodiment of the present invention provides a roadside digital road condition monitoring method based on deep learning, comprising:

[0005] Acquire a target road monitoring vibration signal collected by a vibration sensor deployed on a road section to be monitored;

[0006] Inputting the target road monitoring vibration signal into a previously trained target road condition analysis model to determine the current road condition type of the road section to be monitored;

[0007] Acquire road monitoring surveillance images collected by cameras deployed on the road section to be monitored;

[0008] Generate a road monitoring event to be processed according to the road monitoring surveillance image and the current road state type;

[0009] A road maintenance strategy for the road section to be monitored is determined according to the road monitoring event to be processed.

[0010] In a second aspect, an embodiment of the present invention provides a server system, including a server, wherein the server is used to execute the method described in the first aspect.

[0011] Compared with the prior art, the beneficial effects provided by the present invention include: adopting a roadside digital road condition monitoring method and system based on deep learning disclosed by the present invention, obtaining the road monitoring vibration signal collected by the vibration sensor, and using the trained road condition analysis model to determine the current road state. At the same time, combined with the monitoring images collected by the camera, to-be-processed road monitoring events are generated. Finally, maintenance strategies for specific road sections are determined based on these events. With such a design, real-time monitoring of road conditions and intelligent maintenance decisions are achieved through deep learning technology, thereby improving the efficiency and accuracy of road maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative work.

[0013] Figure 1 A schematic diagram of the steps of a roadside digital road condition monitoring method based on deep learning provided by an embodiment of the present invention;

[0014] Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions 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. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0016] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0017] In order to solve the technical problems in the aforementioned background technology, Figure 1 A schematic flow chart of a roadside digital road condition monitoring method based on deep learning provided in an embodiment of the present disclosure is provided below. The roadside digital road condition monitoring method based on deep learning is introduced in detail.

[0018] Step S201, obtaining a target road monitoring vibration signal collected by a vibration sensor deployed on a road section to be monitored;

[0019] Step S202, inputting the target road monitoring vibration signal into a target road condition analysis model trained in advance to determine the current road condition type of the road section to be monitored;

[0020] Step S203, obtaining road monitoring images collected by cameras deployed on the road section to be monitored;

[0021] Step S204, generating a road monitoring event to be processed according to the road monitoring surveillance image and the current road state type;

[0022] Step S205: determining a road maintenance strategy for the road section to be monitored according to the road monitoring event to be processed.

[0023] In an embodiment of the present invention, illustratively, the server receives data from vibration sensors installed beside the main roads of the city. These sensors continuously capture tiny vibrations on the road surface and transmit these data to the server in real time. For example, a group of high-precision vibration sensors are deployed on the main roads of a certain city. These sensors can capture the vibration signals generated when vehicles pass by, and send these signals to the server in real time through the network for processing. After receiving the vibration signal, the server will input these data into a previously trained deep learning model. This model can identify different road conditions, such as flatness, potholes, congestion, etc. For example, the model analyzes the received vibration signal and identifies that a certain section of the road has obvious bumps, thereby judging that the section may have potholes or damage. At the same time, the server will also obtain real-time road monitoring images from cameras installed on the roadside. These cameras can capture real-time images of the road and provide the server with more intuitive road condition information. For example, at a certain intersection, a high-definition camera is set up, which can monitor the traffic conditions at the intersection in real time and transmit the image data to the server. The server combines the vibration signal analysis with the images captured by the camera to generate specific road monitoring events. For example, by analyzing the vibration signal, the server determines that there is a pothole on a certain road section, and the camera also captures the damaged image of the road section, so the server generates a monitoring event of "the road section is damaged and needs to be repaired". Based on the monitoring events generated above, the server will formulate a corresponding road maintenance strategy. In the example of the embodiment of the present invention, after the server determines that there is a pothole on a certain road section and needs to be repaired, it will automatically generate a maintenance plan, including the specific location of the repair, the estimated repair time and the required resources, etc., and send the plan to the relevant road maintenance personnel so that they can carry out the repair in time.

[0024] In the embodiment of the present invention, the aforementioned step S202 may be implemented through the following example execution.

[0025] Acquire the target road monitoring vibration signal and a preset road condition description set, wherein a road condition description in the road condition description set is used to indicate a road state type in preset road state types;

[0026] The target road monitoring vibration signal and the road condition description set are loaded into a target road condition analysis model that has been trained in advance to obtain target road monitoring vibration signal features and a road condition description feature set, wherein the target road condition analysis model includes a road condition description feature extraction network, a road monitoring vibration signal feature extraction network, road condition description auxiliary guidance data, and road monitoring vibration signal auxiliary guidance data, the road condition description auxiliary guidance data and the road monitoring vibration signal auxiliary guidance data are generated by calculation from the original auxiliary guidance data, the road monitoring vibration signal auxiliary guidance data is used to be jointly loaded with the target road monitoring vibration signal into the road monitoring vibration signal feature extraction network to obtain the target road monitoring vibration signal features, and the road condition description auxiliary guidance data is used to be jointly loaded with the road condition description set into the road condition description feature extraction network to obtain the road condition description feature set;

[0027] Determine the matching degree between the target road monitoring vibration signal feature and each road condition description feature in the road condition description feature set by using the target road condition analysis model, and take the target road state type represented by the road condition description feature whose matching degree reaches a preset matching threshold as the road state type corresponding to the target road monitoring vibration signal feature;

[0028] The road state type is used as the current road state type of the road section to be monitored.

[0029] In an embodiment of the present invention, illustratively, the server first receives a monitoring vibration signal of the target road from a vibration sensor. At the same time, the server also has a preset road condition description set, which includes a variety of possible road condition descriptions, such as "flat road surface", "mild potholes", "severe potholes", "damaged road surface", etc. Each description corresponds to a specific road state type. For example, on a traffic artery in a certain city, the server receives real-time data from a vibration sensor and prepares to use these data and a preset road condition description set for further analysis. The server loads the received vibration signal and the road condition description set together into a target road condition analysis model that has been trained in advance. This model includes two feature extraction networks: a road condition description feature extraction network and a road monitoring vibration signal feature extraction network. At the same time, the model also includes two types of auxiliary guidance data generated from the original auxiliary guidance data: road condition description auxiliary guidance data and road monitoring vibration signal auxiliary guidance data. Inside the model, the road condition description auxiliary guidance data and the road condition description set are jointly loaded into the road condition description feature extraction network to obtain the road condition description feature set; while the road monitoring vibration signal auxiliary guidance data are jointly loaded into the road monitoring vibration signal feature extraction network with the target road monitoring vibration signal to obtain the target road monitoring vibration signal feature. For example, the server inputs the received vibration signal data and road condition descriptions such as "flat road surface" and "mild potholes" into the model, and the model extracts the characteristics of the vibration signal and the characteristics of each road condition through its internal feature extraction network. The model then calculates the degree of match between the extracted vibration signal features and each road condition description feature. This is usually achieved by calculating the similarity between feature vectors, such as cosine similarity, Euclidean distance, etc. For example, the model can find that the current vibration signal feature is most similar to the feature of the road condition description "mild potholes", and the degree of match reaches the preset matching threshold. Once a road condition description feature whose degree of match with the vibration signal feature reaches or exceeds the preset threshold is found, the server will determine the road state type represented by the feature as the actual state type of the current road. In the example of the embodiment of the present invention, since the vibration signal characteristics are highly matched with the description characteristics of "mild potholes", the server determines that there are mild potholes in the current road section. Finally, the server records the determined road state type (such as "mild potholes") as the current state of the road section to be monitored, and further uses it to generate road monitoring events or formulate maintenance strategies. For example, the server can automatically trigger a warning based on this state, notifying the road maintenance department that there are mild potholes in the road section and that timely processing is required.

[0030] In order to more clearly describe the solution provided by the embodiment of the present invention, a more detailed description is given below.

[0031] In the embodiment of the present invention, the target road monitoring vibration signal refers to a data signal collected in real time by a vibration sensor deployed on the road section to be monitored. These data signals reflect the vibration of the road surface, which is usually caused by factors such as passing vehicles, pedestrians and the condition of the road itself. By analyzing these vibration signals, the use and status of the road can be inferred. Assume that a vibration sensor is installed on the main road of a city. Whenever a vehicle passes by, the sensor will capture the vibration generated and convert these vibrations into electrical signals. These electrical signals are the so-called "target road monitoring vibration signals", which will be transmitted to the server for further analysis and processing. The preset road condition description set is a predefined set that contains a series of possible road condition descriptions. Each description corresponds to a specific road condition type, which is used to classify and identify the actual condition of the road. These descriptions are usually based on a summary of historical data and expert knowledge, which can help analyze and identify the signals collected from the vibration sensor. The preset road condition description set can include multiple descriptions such as "flat road surface", "mild potholes", "severe potholes", "pavement damage", and "road icing". These descriptions are designed to cover all possible road conditions so that the specific condition of the current road can be determined by matching the vibration signal features. A road condition description refers to a specific description in a preset road condition description set. Each description describes a specific road condition in detail, including information such as the flatness of the road surface, whether it is damaged, and whether there are potholes. These descriptions are the basis for subsequent road condition identification and classification. "Mild potholes" is a specific road condition description. It indicates that there are some small depressions or protrusions on the road surface. Although they will not seriously affect driving safety, they can affect driving comfort. When the signal collected by the vibration sensor matches this description, it can be determined that there are mild potholes on the current road. The preset road condition type is a set of defined type labels used to describe different road conditions. Each type corresponds to one or several specific road condition descriptions, which is an abstract and generalization of the road condition. "Flat road surface", "pothole road surface", "damaged road surface", etc. can all be regarded as preset road condition types. These types are defined based on the actual use and maintenance needs of the road, which helps to quickly understand and judge the overall condition of the road. For example, the type of "pothole road" corresponds to specific road condition descriptions such as "mild potholes" and "severe potholes".

[0032] In addition, the target road condition analysis model is a trained machine learning model used to analyze vibration signals and road condition descriptions to identify the specific conditions of the current road. The road condition description feature extraction network and the road monitoring vibration signal feature extraction network are part of the target road condition analysis model, responsible for extracting key features from the road condition description and vibration signals, respectively. The road condition description auxiliary guidance data and the road monitoring vibration signal auxiliary guidance data are additional information that assists the model in feature extraction and recognition, which can improve the accuracy and efficiency of the model. The original auxiliary guidance data is unprocessed data directly derived from actual observations or measurements, and is used to generate road condition description auxiliary guidance data and road monitoring vibration signal auxiliary guidance data. The target road monitoring vibration signal features are key features extracted from the vibration signal of the target road. These features may include the frequency, amplitude, duration, etc. of the vibration, which can reflect the specific usage status of the road. The road condition description feature set is a feature set containing multiple road condition descriptions. Each description feature corresponds to a specific road state, such as "flat road surface", "pothole road surface", etc. The matching degree refers to the similarity or consistency between the vibration signal features and the road condition description features. The higher the matching degree, the more consistent the road state reflected by the vibration signal is with a certain descriptive feature. The preset matching threshold is a pre-set standard value used to determine whether the matching degree between the vibration signal feature and the road condition description feature is high enough. Only when the matching degree reaches or exceeds this threshold, the two are considered to be matched. The target road state type refers to the specific road state type determined by the matching process, such as "flat road surface", "pothole road surface", etc. It reflects the actual usage status of the current road.

[0033] In addition, the road state type refers to the specific road condition classification determined after analysis and matching, such as "flat road surface", "pothole road surface", "damaged road surface", etc. This type is derived from the matching results of the vibration signal characteristics and the road condition description characteristics. Assuming that after model analysis, it is determined that the vibration signal characteristics of a certain section of road are highly matched with the description characteristics of "pothole road surface", then "pothole road surface" is the road state type of this section of road. The section to be monitored refers to a specific section of road that needs to be monitored for road condition. This section of road can be any road section that needs to pay attention to its usage status and safety, such as the main road, highway, bridge, tunnel, etc. in a city. For example, an important bridge in a city is selected as a "section to be monitored" because of its large traffic volume and long service life, so as to timely discover and deal with potential safety problems. The current road state type refers to the actual road state of the section to be monitored at a specific time point or time period. This state is obtained through the real-time collected vibration signal and the road condition analysis model, reflecting the immediate usage status of the section of road. At a certain point in time, the vibration signal of the road section to be monitored is collected by the vibration sensor, and after model analysis, it is determined that the "current road state type" of the road section is "mild potholes".

[0034] In an embodiment of the present invention, the target road monitoring vibration signal and the road condition description set are loaded into a pre-trained target road condition analysis model to obtain target road monitoring vibration signal features and a road condition description feature set, which can be implemented through the following example.

[0035] Performing a linear adjustment operation on the original auxiliary guidance data and the vibration signal adjustment vector to determine the road monitoring vibration signal auxiliary guidance data, wherein the linear adjustment operation is used to use the component with the first sequence number in the vector corresponding to the road monitoring vibration signal auxiliary guidance data as the multiplication result of the component with the first sequence number in the original auxiliary guidance data and the component with the first sequence number in the vibration signal adjustment vector;

[0036] Combining the target road monitoring vibration signal and the road monitoring vibration signal auxiliary guidance data to determine a vibration signal merging feature;

[0037] The vibration signal combined features are converted into a target feature domain to determine the target road monitoring vibration signal features.

[0038] In an embodiment of the present invention, illustratively, the server receives the original auxiliary guidance data and the vibration signal adjustment vector. The original auxiliary guidance data may include a series of numerical values ​​related to road conditions, such as the measured values ​​of environmental parameters such as road surface hardness, temperature, and humidity. The vibration signal adjustment vector includes an adjustment coefficient set according to historical data and experience. The server starts to perform a linear adjustment operation. For example, suppose that there is a component in the original auxiliary guidance data representing the road surface hardness, and its value is 0.8 (range 0-1, indicating a relatively hard road surface), and the component corresponding to the road surface hardness in the vibration signal adjustment vector is 1.2 (indicating that the vibration signal needs to be amplified to a certain extent according to the road surface hardness). The server multiplies these two components to obtain a new component value of 0.96 (0.8*1.2), and this new value will be used as the value of the road surface hardness component in the road monitoring vibration signal auxiliary guidance data. The server performs similar multiplication operations on all corresponding components, and finally obtains complete road monitoring vibration signal auxiliary guidance data. After obtaining the road monitoring vibration signal auxiliary guidance data, the server merges it with the target road monitoring vibration signal for processing. The target road monitoring vibration signal is data captured in real time from vibration sensors deployed on the road, reflecting the real-time vibration of the road. The merging process can be to fuse two sets of data together through some algorithm (such as weighted average, feature splicing, etc.). For example, the server can perform weighted average of the amplitude value of the vibration signal at each time point and the road surface hardness value at the corresponding time point in the auxiliary guidance data to obtain a new numerical sequence, which contains both the information of the vibration signal and the influence of the road surface hardness. Through such merging processing, the server obtains a more comprehensive and accurate vibration signal merging feature. After obtaining the vibration signal merging feature, the server needs to further convert it into the target feature domain. The target feature domain can be a more abstract and discriminative feature space, which helps the model better identify and classify different road conditions. The conversion process designs a series of complex mathematical transformations, such as Fourier transform, wavelet transform, or feature mapping in a deep learning network. For example, the server can use a pre-trained deep learning model, take the vibration signal merging feature as input, perform nonlinear transformation through multiple hidden layers of the model, and finally output it to the representation in the target feature domain. In the target feature domain, the server obtains a more refined and representative target road monitoring vibration signal feature. These characteristics will serve as an important basis for subsequent road condition analysis and classification.

[0039] In the embodiment of the present invention, the target road monitoring vibration signal and the road monitoring vibration signal auxiliary guidance data are merged and processed to determine the vibration signal merging feature, which can be implemented through the following examples.

[0040] dividing the target road monitoring vibration signal into a plurality of vibration signal segments;

[0041] Mapping the multiple vibration signal segments to a vibration signal feature domain to obtain multiple local vibration signal features;

[0042] The multiple local vibration signal features and the road monitoring vibration signal auxiliary guidance data are combined and processed to determine the vibration signal combination feature.

[0043] In an embodiment of the present invention, illustratively, the server receives a continuous target road monitoring vibration signal, which contains all vibration data within a certain period of time on the road. In order to analyze these data more carefully, the server first needs to divide this long signal into multiple shorter vibration signal segments. For example, the server can cut the signal according to a fixed time window (such as per second). If the original signal is 10 seconds of data, it can be cut into 10 1-second vibration signal segments. Each signal segment contains all vibration information within the time period. After dividing the vibration signal segments, the server then needs to map each signal segment to the vibration signal feature domain to extract the local features of each signal segment. This process is similar to feature extraction or feature conversion of data. Specifically, the server can use Fourier transform, wavelet transform and other methods to convert the vibration signal of each time period into a representation in the frequency domain or other feature domains. For example, through Fourier transform, the server can obtain the intensity of different frequency components in each signal segment, and these frequency components and intensities constitute the local vibration signal characteristics of the signal segment. After extracting the local features of each vibration signal segment, the server needs to merge these local features with the auxiliary guidance data of the road monitoring vibration signal. The auxiliary guidance data can include external factors such as road surface temperature, humidity, and vehicle load, which will affect the vibration characteristics of the road. During the merging process, the server can use a variety of methods, such as feature splicing, feature fusion, etc. Taking feature splicing as an example, the server can directly splice the local vibration signal features of each signal segment with the corresponding auxiliary guidance data to form a longer feature vector. This feature vector contains both the intrinsic characteristics of the vibration signal and the influence of external factors, so it is more representative. Through such a merging process, the server finally obtains a vibration signal merge feature that contains multiple local vibration signal features and auxiliary guidance data. This merged feature will serve as an important basis for subsequent road condition analysis and classification.

[0044] In the embodiment of the present invention, the road condition description set is loaded into a pre-trained target road condition analysis model to obtain a road condition description feature set, which can be implemented through the following example.

[0045] Performing a linear adjustment operation on the original auxiliary guidance data and the road condition description adjustment vector to determine the road condition description auxiliary guidance data, wherein the linear adjustment operation is used to use the component with the second sequence number in the vector corresponding to the road condition description auxiliary guidance data as the multiplication result of the component with the second sequence number in the original auxiliary guidance data and the component with the second sequence number in the road condition description adjustment vector;

[0046] The road condition description set and the road condition description auxiliary guidance data are respectively merged to determine a road condition description merge feature set;

[0047] The road condition description combined feature set is converted into a target feature domain respectively to determine the road condition description feature set.

[0048] In the embodiment of the present invention, illustratively, the server receives a set of original auxiliary guidance data, which may include weather conditions, traffic flow, etc., and has an important auxiliary role in road condition analysis. At the same time, the server also has a road condition description adjustment vector, which is set based on historical data and expert experience, and is used to adjust the original auxiliary guidance data to make it more consistent with the actual situation of the current road condition. For example, assume that the original auxiliary guidance data contains a component representing the current traffic flow, and its value is 80 (indicating that the traffic is relatively congested). The component corresponding to the traffic flow in the road condition description adjustment vector is 0.9 (indicating that according to the actual situation, the impact of the traffic flow needs to be slightly adjusted down). The server performs a linear adjustment operation, multiplies the two components, and obtains a new component value of 72 (80*0.9). This new value will be used as the component value of the traffic flow in the road condition description auxiliary guidance data. After obtaining the road condition description auxiliary guidance data, the server needs to merge it with the road condition description set. The road condition description set can contain multiple text descriptions of the road condition, such as "the road surface is seriously damaged", "there is a small amount of water accumulation", etc. During the merging process, the server can first convert each road condition description into a numerical feature vector, for example, by converting the text description into a vector in a high-dimensional space through word embedding technology. Then, the server merges these numerical feature vectors with the auxiliary guidance data of the road condition description, which can be done by concatenation, weighted averaging, etc. The merged data contains both the text description information of the road condition and the influence of the auxiliary guidance data. After obtaining the merged feature set of the road condition description, the server needs to further convert these features into the target feature domain. The target feature domain is a more abstract and discriminative feature space that helps the model better identify and classify different road conditions. For example, the server can use a deep learning model, such as a convolutional neural network (CNN) or a recurrent neural network (RNN), and take the merged feature set of the road condition description as input. Through nonlinear transformation and learning of multiple hidden layers of the model, the server finally maps the features into the target feature domain. In this feature domain, different road conditions can have more obvious discrimination, which helps the server to perform more accurate road condition analysis and classification.

[0049] In the embodiment of the present invention, the road condition description set and the road condition description auxiliary guidance data are merged separately to determine the road condition description merge feature set, which can be implemented through the following examples.

[0050] The road condition description set and the road condition description auxiliary guidance data are merged and processed respectively by the following process to determine a road condition description merge feature set, wherein each merged road condition description is regarded as the current road condition description, and each merged road condition description merge feature is regarded as the current road condition description merge feature:

[0051] Performing text segmentation parsing on the current road condition description to obtain a road condition parsing result set;

[0052] The road condition analysis result set is mapped to a road condition description feature domain to obtain the current road condition description combined feature.

[0053] In an embodiment of the present invention, exemplarily, a server receives a description of the current road condition: "There are many potholes in the XX road section, and there is water accumulation. Vehicles need to slow down when passing through." In order to extract specific road condition information from this description, the server needs to parse this text segment by segment. The server first uses natural language processing technology, such as word segmentation, part-of-speech tagging, etc. to pre-process the text. Then, according to the preset rules or models, the text is divided into different paragraphs or sentences, and further analysis is performed on each paragraph or sentence. In this example, the server can divide the description into three parts: "There are many potholes in the XX road section", "and there is water accumulation", and "Vehicles need to slow down when passing through". After parsing each segment, the server will obtain a road condition analysis result set. This set may contain the following elements: "potholes", "water accumulation", "slow down", etc. These elements are key road condition information extracted from the text, which is crucial for subsequent road condition evaluation and classification. The server then needs to map the road condition analysis result set to the road condition description feature domain. This process is to convert the text description into numerical features that can be understood and processed by the computer. For example, the server can predefine a feature mapping table to map "potholes" to a specific value or vector, "water accumulation" to another value or vector, and so on. In this way, each road condition analysis result will be converted into a numerical feature. After completing the feature mapping, the server will obtain a set of numerical features that together describe the current road condition. In order to obtain a unified representation, the server may need to further merge these features. The merging process can be a simple splicing, weighted averaging, or feature fusion through a more complex model (such as a neural network). In the end, the server will obtain a merged feature describing the current road condition, which integrates all the key information in the text description and can be used for subsequent road condition classification, evaluation or prediction tasks. Through the above steps, the server can effectively extract the key information of the road condition from the text description and convert it into numerical features that can be processed by the computer, providing strong data support for subsequent road management and maintenance work.

[0054] In the embodiments of the present invention, the following implementation modes are also provided.

[0055] Acquire a sample road monitoring vibration signal and a sample road condition description set, wherein the sample road monitoring vibration signal is pre-labeled to correspond to a target sample road condition description, and the sample road condition description set includes the target sample road condition description;

[0056] The sample road monitoring vibration signal and the sample road condition description set are loaded into an untrained original road condition analysis model to obtain sample road monitoring vibration signal features and sample road condition description set features, wherein the original road condition analysis model includes the road condition description feature extraction network, the road monitoring vibration signal feature extraction network, initial road condition description auxiliary guidance data and original road monitoring vibration signal auxiliary guidance data, and the initial road condition description auxiliary guidance data and the original road monitoring vibration signal auxiliary guidance data are generated by computing the initial original auxiliary guidance data;

[0057] Determining the sample matching degree between the sample road monitoring vibration signal feature and each sample road condition description feature in the sample road condition description set feature by using the original road condition analysis model;

[0058] A cost parameter is calculated based on the sample matching degree, and the initial original auxiliary guidance data and the initial adjustment vector are optimized based on the cost parameter until the target road condition analysis model trained based on the original road condition analysis model is obtained.

[0059] In an embodiment of the present invention, exemplarily, the server obtains a set of sample data from a database, including a sample road monitoring vibration signal and a corresponding sample road condition description set. For example, a sample road monitoring vibration signal may record the vibration data of a section of a city trunk road, and the corresponding sample road condition description set includes descriptions such as "smooth road surface" and "large traffic volume". These sample data have been preprocessed, wherein the sample road monitoring vibration signal has been labeled with the target sample road condition description. The server loads the obtained sample data into an untrained original road condition analysis model. This model includes a road condition description feature extraction network, a road monitoring vibration signal feature extraction network, and initial auxiliary guidance data and adjustment vectors. During the loading process, the sample road monitoring vibration signal is converted into a vibration signal feature through the vibration signal feature extraction network, and the sample road condition description set is converted into a description feature through the road condition description feature extraction network. The server uses the original road condition analysis model to calculate the degree of match between each sample road monitoring vibration signal feature and the corresponding sample road condition description feature. For example, the degree of match is quantified by calculating the similarity or correlation between the two. If the vibration signal features of a sample are highly similar to the corresponding road condition description features, the degree of match between them is considered to be high. Based on the sample matching degree, the server calculates a cost parameter that reflects the performance of the model under the current parameters. For example, the cost parameter can be the average value of the matching degree of all samples or some form of loss function value. The server then uses an optimization algorithm (such as gradient descent) to adjust the initial raw auxiliary guidance data and the initial adjustment vector to minimize the cost parameter. This process is repeated until the model performance reaches a certain preset standard or the cost parameter converges. After multiple iterative optimizations, the server finally obtains a target road condition analysis model that has been trained based on the original road condition analysis model. This model has learned how to extract effective features from road monitoring vibration signals and accurately match them with road condition descriptions. Now, the model can be used to process new road monitoring vibration signals and generate corresponding road condition descriptions or classification results.

[0060] In an embodiment of the present invention, the sample road monitoring vibration signal and the sample road condition description set are loaded into an untrained original road condition analysis model to obtain sample road monitoring vibration signal features and sample road condition description set features, which can be implemented through the following example.

[0061] Acquire the initial original auxiliary guidance data, the initial vibration signal adjustment vector, and the initial road condition description adjustment vector, wherein the initial original auxiliary guidance data, the initial vibration signal adjustment vector, and the initial road condition description adjustment vector are adjustable parameters when training the original road condition analysis model;

[0062] Performing a linear adjustment operation on the initial raw auxiliary guidance data and the initial vibration signal adjustment vector to determine the original road monitoring vibration signal auxiliary guidance data, wherein the linear adjustment operation is used to use the third serial number component in the vector corresponding to the original road monitoring vibration signal auxiliary guidance data as the multiplication result of the third serial number component in the initial raw auxiliary guidance data and the third serial number component in the initial vibration signal adjustment vector;

[0063] Performing the linear adjustment operation on the initial original auxiliary guidance data and the initial road condition description adjustment vector to determine the initial road condition description auxiliary guidance data, wherein the linear adjustment operation is used to use the fourth sequence number component in the vector corresponding to the initial road condition description auxiliary guidance data as the multiplication result of the fourth sequence number component in the initial original auxiliary guidance data and the fourth sequence number component in the initial road condition description adjustment vector;

[0064] Merging the sample road monitoring vibration signal and the auxiliary guidance data of the original road monitoring vibration signal to determine a sample vibration signal merging feature, and merging the sample road condition description set and the auxiliary guidance data of the initial road condition description to determine a sample road condition description merging feature set;

[0065] The sample vibration signal combined feature and the sample road condition description combined feature set are converted to an initial feature domain to determine the sample road monitoring vibration signal feature and the sample road condition description set feature.

[0066] In an embodiment of the present invention, exemplarily, the server first obtains the initial original auxiliary guidance data, the initial vibration signal adjustment vector and the initial road condition description adjustment vector from the configuration file. These parameters are the parameters that can be adjusted by the original road condition analysis model during the training process, and are used to help the model better learn and adapt to the data. The server uses the initial original auxiliary guidance data and the initial vibration signal adjustment vector to perform a linear adjustment operation. For example, assuming that a component in the initial original auxiliary guidance data is 0.8 (the component on the third sequence number), and the corresponding component in the initial vibration signal adjustment vector is 1.2 (the component on the third sequence number), then through multiplication, the server obtains the value of the corresponding component in the original road monitoring vibration signal auxiliary guidance data as 0.96 (0.8*1.2). This process is performed on all components to obtain complete original road monitoring vibration signal auxiliary guidance data. The server uses the initial original auxiliary guidance data and the initial road condition description adjustment vector to perform a linear adjustment operation. For example, one component in the initial raw auxiliary guidance data is 0.7 (the component on the fourth sequence number), and the corresponding component in the initial road condition description adjustment vector is 1.1 (the component on the fourth sequence number). Through the multiplication operation, the server obtains the value of the corresponding component in the initial road condition description auxiliary guidance data as 0.77 (0.7*1.1). This process is also performed on all components to obtain the complete initial road condition description auxiliary guidance data. The server then merges the sample road monitoring vibration signal and the original road monitoring vibration signal auxiliary guidance data. For example, the server can perform weighted average of the energy of each frequency band of the vibration signal with the corresponding component in the auxiliary guidance data to obtain the sample vibration signal merge feature. Similarly, the server will also merge the sample road condition description set and the initial road condition description auxiliary guidance data, such as converting the text description into a vector through word embedding technology, and splicing or weighted averaging it with the auxiliary guidance data vector to obtain the sample road condition description merge feature set. Finally, the server converts the sample vibration signal merge feature and the sample road condition description merge feature set to the initial feature domain. This is usually achieved through a linear transformation or nonlinear transformation (such as a neural network layer). The converted features will be easier to learn and classify by the model. For example, the server can use a pre-trained neural network model to extract higher-level feature representations that can better reflect the complex relationship between road conditions and vibration signals. After the conversion is completed, the server obtains the final sample road monitoring vibration signal features and sample road condition description set features, which will be used in subsequent training and optimization processes.

[0067] In an embodiment of the present invention, the target road monitoring vibration signal and the road condition description set are loaded into a pre-trained target road condition analysis model to obtain target road monitoring vibration signal features and a road condition description feature set, which can be implemented through the following example.

[0068] On the basis that the target road condition analysis model includes a plurality of road monitoring vibration signal analysis networks, the corresponding original auxiliary guidance data and adjustment vector are configured for each of the road monitoring vibration signal analysis networks, wherein the output of a first road monitoring vibration signal analysis network among the plurality of road monitoring vibration signal analysis networks and the road monitoring vibration signal auxiliary guidance data corresponding to a second road monitoring vibration signal analysis network are combined as inputs of the second road monitoring vibration signal analysis network, and the second road monitoring vibration signal analysis network is a subsequent network of the first road monitoring vibration signal analysis network;

[0069] On the basis that the target road condition analysis model includes multiple road condition analysis networks, the corresponding original auxiliary guidance data and the adjustment vector are configured for each of the road condition analysis networks, wherein the output of the first road condition analysis network among the multiple road condition analysis networks and the road condition description auxiliary guidance data corresponding to the second road condition analysis network are combined as the input of the second road condition analysis network.

[0070] In an embodiment of the present invention, exemplarily, the target road condition analysis model deployed in the server includes multiple road monitoring vibration signal analysis networks. In order to improve the model's processing capability for different types of vibration signals, the server configures corresponding original auxiliary guidance data and adjustment vectors for each road monitoring vibration signal analysis network. For example, there are two road monitoring vibration signal analysis networks in the model, namely network A and network B. Network A is responsible for processing low-frequency vibration signals, while network B processes high-frequency vibration signals. The server configures auxiliary guidance data and adjustment vectors that focus on low-frequency signals for network A, and configures auxiliary guidance data and adjustment vectors for high-frequency signals for network B. In the model, network A is the first road monitoring vibration signal analysis network, and its output not only includes the processed low-frequency vibration signal characteristics, but also combines the road monitoring vibration signal auxiliary guidance data corresponding to network B. These data are combined as the input of network B (i.e., the second road monitoring vibration signal analysis network, which is also the subsequent network of network A). For example, when the server receives a vibration signal containing low-frequency and high-frequency components, the low-frequency part is first processed by network A, and the processing result is merged with the auxiliary guidance data of network B, and then passed to network B for analysis of the high-frequency part. In addition to the vibration signal analysis network, the target road condition analysis model also includes multiple road condition analysis networks. These networks are used to infer road conditions based on vibration signal characteristics. The server also configures corresponding original auxiliary guidance data and adjustment vectors for each road condition analysis network. For example, the model contains two road condition analysis networks, namely network C and network D. Network C is responsible for judging whether the road is damaged based on the vibration signal characteristics, while network D further analyzes the specific type and degree of damage. The server configures corresponding auxiliary guidance data and adjustment vectors for these two networks to optimize their analysis performance. In the model, network C serves as the first road condition analysis network, and its output is combined with the road condition description auxiliary guidance data corresponding to network D as the input of network D (i.e., the second road condition analysis network). For example, when network C determines that the road is damaged based on the vibration signal characteristics, it will merge this result with the auxiliary guidance data of network D, and then pass it to network D for more in-depth analysis. Network D will further determine the specific type and degree of damage based on this information and generate a corresponding road condition description. In this way, the server can use multiple networks to conduct a step-by-step in-depth analysis and judgment of complex road monitoring vibration signals and road conditions.

[0071] In an embodiment of the present invention, the target road condition analysis model is used to determine the degree of match between the target road monitoring vibration signal feature and each road condition description feature in the road condition description feature set, and the target road state type represented by the road condition description feature whose matching degree reaches a preset matching threshold is used as the road state type corresponding to the target road monitoring vibration signal feature. Implementation can be performed through the following examples.

[0072] Acquire a first Pearson correlation coefficient between the target road monitoring vibration signal feature and a first road condition description feature and a second Pearson correlation coefficient between the target road monitoring vibration signal feature and a second road condition description feature, wherein the road condition description feature set includes the first road condition description feature and the second road condition description feature;

[0073] On the basis that the first Pearson correlation coefficient does not exceed a Pearson correlation coefficient threshold, determining that the first road condition description feature is a target road condition description feature, and the road state type of the target road monitoring vibration signal is the target road state type;

[0074] On the basis that the second Pearson correlation coefficient exceeds the Pearson correlation coefficient threshold, it is determined that the second road condition description feature is not the target road condition description feature, and the road state type of the target road monitoring vibration signal is not the target road state type.

[0075] In an embodiment of the present invention, exemplarily, the server first calculates the Pearson correlation coefficient between the target road monitoring vibration signal feature and each feature in the road condition description feature set. For example, the server calculates the first Pearson correlation coefficient between the target road monitoring vibration signal feature and the first road condition description feature (such as "road surface damage"), assuming that the result obtained is 0.3. Next, the server calculates the second Pearson correlation coefficient between the target road monitoring vibration signal feature and the second road condition description feature (such as "slippery road surface"), assuming that the result obtained is 0.8. The server sets a Pearson correlation coefficient threshold, such as 0.5, to determine whether there is a strong correlation between the road monitoring vibration signal feature and the road condition description feature. Since the calculated first Pearson correlation coefficient is 0.3, which is lower than the set threshold value of 0.5, the server determines that the correlation between the target road monitoring vibration signal feature and the first road condition description feature (road surface damage) is not strong. Based on the above judgment, since the first Pearson correlation coefficient is lower than the threshold, the server determines that the first road condition description feature (road surface damage) is not a feature that is strongly correlated with the target road monitoring vibration signal feature. Therefore, the server will not use the state type of "pavement damage" as the road state type corresponding to the target road monitoring vibration signal feature. For the second Pearson correlation coefficient of 0.8, it is higher than the set Pearson correlation coefficient threshold of 0.5. This shows that there is a strong correlation between the target road monitoring vibration signal feature and the second road condition description feature (slippery road surface). Since the second Pearson correlation coefficient is higher than the threshold, the server determines that the second road condition description feature (slippery road surface) is a feature that is strongly correlated with the target road monitoring vibration signal feature. However, in this scenario, since it is looking for unmatched features, the server will exclude the state type of "slippery road surface" and will not use it as the road state type corresponding to the target road monitoring vibration signal feature. Ultimately, through this process, the server is able to filter out road condition description features that do not match the target road monitoring vibration signal feature, and determine the unmatched road state type based on this, thereby more accurately identifying the road state type that truly matches the target road monitoring vibration signal. In this example, the server can continue to look for other road condition description features until it finds a feature that matches the target road monitoring vibration signal feature.

[0076] In the embodiments of the present invention, the following implementation modes are also provided.

[0077] determining that the target road state type is an abnormal road state type set, wherein the preset road state type includes the abnormal road state type set;

[0078] In response to determining that the target road monitoring vibration signal is the abnormal road monitoring vibration signal, abnormality auxiliary guidance data is generated, wherein the abnormality auxiliary guidance data is used to indicate that the target road monitoring vibration signal is an abnormal vibration signal.

[0079] In an embodiment of the present invention, exemplarily, the server has determined that the road state type corresponding to the target road monitoring vibration signal feature is "road collapse" through the analysis of the previous steps. An abnormal road state type set is preset inside the server, which contains abnormal road states such as "road collapse", "serious cracks", "road surface water" and so on that may affect traffic safety. The server compares the determined target road state type "road collapse" with the internal abnormal road state type set. After comparison, the server finds that "road collapse" does belong to one of the abnormal road state type sets. Since the target road state type "road collapse" belongs to the abnormal road state type set, the server determines that the corresponding target road monitoring vibration signal is an abnormal road monitoring vibration signal. This means that the vibration signal reflects the abnormal situation on the road and may require further processing and attention. After determining that the target road monitoring vibration signal is an abnormal signal, the server immediately generates abnormal auxiliary guidance data. These data contain detailed information about the abnormal vibration signal, such as the abnormal type ("road collapse"), the latitude and longitude coordinates of the abnormal position, the intensity and timestamp of the abnormal signal, etc. These data are mainly used to indicate and record the abnormal vibration signal for subsequent analysis, processing and reporting. The generated abnormal auxiliary guidance data will be used by the server in many aspects. First, these data can be sent to the road maintenance department so that they can quickly locate and deal with abnormal situations such as road collapse. Secondly, these data can also be used to update the road condition analysis model of the server to improve the model's recognition accuracy of similar abnormal situations in the future. Finally, these data can also be used to generate reports or alarms to notify relevant managers or the public to pay attention to road safety.

[0080] In the embodiments of the present invention, the following implementation modes are also provided.

[0081] Performing an element-by-element product operation on the original auxiliary guidance data and the vibration signal adjustment vector to obtain the road monitoring vibration signal auxiliary guidance data, wherein the adjustment vector includes the vibration signal adjustment vector;

[0082] The original auxiliary guidance data and the road condition description adjustment vector are subjected to the element-by-element product operation to obtain the road condition description auxiliary guidance data, wherein the adjustment vector includes the road condition description adjustment vector.

[0083] In an embodiment of the present invention, exemplarily, a large amount of original auxiliary guidance data is stored in the server, which is generated based on historical vibration signals and road condition information and is used to assist in analyzing the current road monitoring vibration signal. At the same time, the server also generates a vibration signal adjustment vector based on real-time road and environmental conditions. Now, the server needs to analyze a specific road monitoring vibration signal. In order to perform this analysis more accurately, the server first obtains the original auxiliary guidance data related to the vibration signal. These data may include vibration signal characteristics under similar road conditions before, the influence of environmental factors on the vibration signal, etc. Then, the server takes out the vibration signal adjustment vector generated for the current road and environmental conditions. This vector contains multiple elements, each of which corresponds to a feature in the original auxiliary guidance data, which is used to adjust the weight of the feature. The server performs an element-by-element product operation to multiply the original auxiliary guidance data with the element corresponding to the vibration signal adjustment vector. In this way, the server can dynamically adjust the importance of each feature in the original auxiliary guidance data according to the real-time road and environmental conditions, thereby generating more accurate road monitoring vibration signal auxiliary guidance data. For example, if the current road is slippery, then the features related to slippery road surface can have a higher weight in the vibration signal adjustment vector. Through the element-by-element product operation, the weights of these features will be correspondingly amplified in the generated auxiliary guidance data, thereby improving the accuracy of the server in analyzing the vibration signal on the slippery road. After processing the vibration signal, the server needs to further assist in the analysis based on the road condition description information. The server also stores a large amount of original auxiliary guidance data related to the road condition description, which can include historical road damage records, traffic flow information, weather conditions, etc. In order to more accurately reflect the current road conditions, the server also generates a road condition description adjustment vector. This vector is dynamically adjusted based on real-time road monitoring data, user feedback and other information to reflect the latest changes in the current road conditions. The server performs an element-by-element product operation on the original auxiliary guidance data and the road condition description adjustment vector. In this way, the server can adjust the weights of each feature in the original auxiliary guidance data in real time based on the latest information of the current road conditions, thereby generating more accurate road condition description auxiliary guidance data. For example, if the current road is undergoing maintenance construction, then the features related to construction can have a higher weight in the road condition description adjustment vector. Through element-by-element multiplication operations, the weights of these features will be amplified accordingly in the generated auxiliary guidance data, thereby improving the accuracy of the server in analyzing road conditions on construction sections.

[0084] In the embodiment of the present invention, the aforementioned step S205 can be implemented through the following example execution.

[0085] For the road monitoring surveillance image in the road monitoring event to be processed, similar images whose relevance reaches a relevance threshold are retrieved by an image retrieval model to obtain candidate surveillance images, wherein the image retrieval model is trained using historical road monitoring surveillance images as training instance data;

[0086] Based on the candidate monitoring image, for the current road state type in the road monitoring event to be processed, the candidate current road state type whose relevance reaches the relevance threshold is retrieved through the current road state type to obtain the candidate road monitoring event, wherein the current road state type retrieval is trained by using the historical current road state type as training instance data;

[0087] The road risk information evaluation model is used to evaluate the road monitoring event to be processed and the candidate road monitoring event to obtain risk information of the road monitoring event to be processed and risk information of each candidate road monitoring event, wherein the road risk information evaluation model is trained using monitoring events with preset labels in historical roadside monitoring data as training instance data;

[0088] taking the candidate road monitoring event that matches the risk information of the road monitoring event to be processed as a road condition monitoring matching result;

[0089] The road maintenance strategy for the road section to be monitored is called from a preset database according to the road condition monitoring matching result.

[0090] In an embodiment of the present invention, illustratively, a server receives a road monitoring event to be processed, which includes a road monitoring surveillance image. In order to find information associated with this image, the server first uses an image retrieval model to retrieve similar images whose correlation reaches a set threshold. This image retrieval model was previously trained through a large number of historical road monitoring surveillance images. For example, the server stores tens of thousands of historical road monitoring images, which include various road conditions and events. When the server receives a new surveillance image, it uses this image retrieval model to search for similar images in the historical image library. Through comparison and calculation, the model finds several historical surveillance images that are highly similar to the new image in visual features, and these selected images are called candidate surveillance images. After obtaining the candidate surveillance images, the server further analyzes the current road state type (such as "road damage" or "traffic congestion") in the road monitoring event to be processed. In order to find events of similar state types in history, the server uses another model-the current road state type retrieval model, which is trained through historical current road state type data. The server inputs the current road state type into the model, and the model retrieves the candidate current road state type with high correlation with the current state type from the historical events corresponding to the candidate monitoring images, thereby obtaining a series of candidate road monitoring events. These candidate events have a high similarity with the current event in terms of road state. Next, the server needs to evaluate the risk information of the current road monitoring event to be processed and the candidate road monitoring event. This is done through a road risk information evaluation model, which is trained with monitoring events with preset labels in historical roadside monitoring data as training instance data. The server inputs the monitoring images of the current event and the candidate event into the road risk information evaluation model, and the model analyzes the road conditions, traffic flow, environmental factors, etc. in the image, and generates a risk information evaluation report for each event. These reports describe in detail the risk level of the event, the possible scope of impact, and the potential consequences. After obtaining the risk information evaluation reports of each event, the server begins to match these reports with the risk information of the current road monitoring event to be processed. By comparing key indicators such as risk level and scope of impact, the server can find the candidate road monitoring event that best matches the risk information of the current event. These matching candidate events are regarded as the matching results of road condition monitoring, which provide the server with important references about the possible subsequent development, impact and response strategies of the current event. Finally, based on the matching results of road condition monitoring, the server will call the road maintenance strategy for the road section to be monitored from the pre-set database. This database stores various road conditions and the corresponding optimal maintenance strategies, which are formulated based on historical data and expert experience.For example, if the matching results show that the current event is highly similar to the historical "road damage causing traffic congestion" event, the server will retrieve the maintenance strategy for this situation from the database, such as immediately dispatching a maintenance team to repair the road surface, setting up temporary traffic signs to guide traffic, etc. These strategies will then be executed to quickly and effectively respond to current road problems.

[0091] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the above-mentioned roadside digital road condition monitoring method based on deep learning. Figure 2 As shown, Figure 2 The block diagram of the computer device 100 provided in the embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112 and a communication unit 113. To achieve data transmission or interaction, the memory 111, the processor 112 and the communication unit 113 are directly or indirectly electrically connected to each other. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0092] For illustrative purposes, the foregoing description is made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise form disclosed. Numerous modifications and variations are possible based on the above teachings. These embodiments are selected and described in order to best illustrate the principles of the present disclosure and its practical application, so that those skilled in the art can best be based on the present disclosure and on various embodiments with different modifications to be suitable for the expected specific application.

Claims

1. A roadside digital road condition monitoring method based on deep learning, characterized in that: include: Acquire a target road monitoring vibration signal collected by a vibration sensor deployed on a road section to be monitored; Inputting the target road monitoring vibration signal into a previously trained target road condition analysis model to determine the current road condition type of the road section to be monitored; Acquire road monitoring surveillance images collected by cameras deployed on the road section to be monitored; Generate a road monitoring event to be processed according to the road monitoring surveillance image and the current road state type; Determining a road maintenance strategy for the road section to be monitored according to the road monitoring event to be processed; The step of inputting the target road monitoring vibration signal into a previously trained target road condition analysis model to determine the current road condition type of the road section to be monitored includes: Acquire the target road monitoring vibration signal and a preset road condition description set, wherein a road condition description in the road condition description set is used to indicate a road state type in preset road state types; The target road monitoring vibration signal and the road condition description set are loaded into a pre-trained target road condition analysis model to obtain target road monitoring vibration signal features and a road condition description feature set, wherein the target road condition analysis model includes a road condition description feature extraction network, a road monitoring vibration signal feature extraction network, road condition description auxiliary guidance data, and road monitoring vibration signal auxiliary guidance data, the road condition description auxiliary guidance data and the road monitoring vibration signal auxiliary guidance data are generated by calculation from the original auxiliary guidance data, and the road monitoring vibration signal auxiliary guidance data is used to be jointly loaded with the target road monitoring vibration signal to the road monitoring vibration signal feature extraction network, obtain the target road monitoring vibration signal feature, the road condition description auxiliary guidance data is used to be loaded into the road condition description feature extraction network together with the road condition description set, and obtain the road condition description feature set; the original auxiliary guidance data includes the measured values ​​of road surface hardness, temperature, and humidity; the original auxiliary guidance data and the vibration signal adjustment vector are subjected to element-by-element product operation processing to obtain the road monitoring vibration signal auxiliary guidance data; the original auxiliary guidance data and the road condition description adjustment vector are subjected to the element-by-element product operation processing to obtain the road condition description auxiliary guidance data; Acquire a first Pearson correlation coefficient between the target road monitoring vibration signal feature and a first road condition description feature and a second Pearson correlation coefficient between the target road monitoring vibration signal feature and a second road condition description feature, wherein the road condition description feature set includes the first road condition description feature and the second road condition description feature; On the basis that the first Pearson correlation coefficient does not exceed a Pearson correlation coefficient threshold, determining that the first road condition description feature is a target road condition description feature, and the road state type of the target road monitoring vibration signal is the target road state type; On the basis that the second Pearson correlation coefficient exceeds the Pearson correlation coefficient threshold, determining that the second road condition description feature is not the target road condition description feature, and the road state type of the target road monitoring vibration signal is not the target road state type; The road state type is used as the current road state type of the road section to be monitored.

2. The method according to claim 1, characterized in that: The step of loading the target road monitoring vibration signal and the road condition description set into a previously trained target road condition analysis model to obtain target road monitoring vibration signal features and a road condition description feature set includes: Performing a linear adjustment operation on the original auxiliary guidance data and the vibration signal adjustment vector to determine the road monitoring vibration signal auxiliary guidance data, wherein the linear adjustment operation is used to use the component with the first sequence number in the vector corresponding to the road monitoring vibration signal auxiliary guidance data as the multiplication result of the component with the first sequence number in the original auxiliary guidance data and the component with the first sequence number in the vibration signal adjustment vector; dividing the target road monitoring vibration signal into a plurality of vibration signal segments; Mapping the multiple vibration signal segments to a vibration signal feature domain to obtain multiple local vibration signal features; Combining the plurality of local vibration signal features and the road monitoring vibration signal auxiliary guidance data to determine the vibration signal merging feature; The vibration signal combined features are converted into a target feature domain to determine the target road monitoring vibration signal features.

3. The method according to claim 1, characterized in that The step of loading the road condition description set into a previously trained target road condition analysis model to obtain a road condition description feature set includes: Performing a linear adjustment operation on the original auxiliary guidance data and the road condition description adjustment vector to determine the road condition description auxiliary guidance data, wherein the linear adjustment operation is used to use the component with the second sequence number in the vector corresponding to the road condition description auxiliary guidance data as the multiplication result of the component with the second sequence number in the original auxiliary guidance data and the component with the second sequence number in the road condition description adjustment vector; The road condition description set and the road condition description auxiliary guidance data are merged and processed respectively by the following process to determine a road condition description merge feature set, wherein each road condition description that is merged is regarded as the current road condition description, and each road condition description merge feature obtained by the merge process is regarded as the current road condition description merge feature: performing text segmentation parsing on the current road condition description to obtain a road condition parsing result set; mapping the road condition parsing result set to a road condition description feature domain to obtain the current road condition description merge feature; The road condition description combined feature set is converted into a target feature domain respectively to determine the road condition description feature set.

4. The method according to claim 1, characterized in that: The method further comprises: Acquire a sample road monitoring vibration signal and a sample road condition description set, wherein the sample road monitoring vibration signal is pre-labeled to correspond to a target sample road condition description, and the sample road condition description set includes the target sample road condition description; Acquire initial raw auxiliary guidance data, an initial vibration signal adjustment vector, and an initial road condition description adjustment vector, wherein the initial raw auxiliary guidance data, the initial vibration signal adjustment vector, and the initial road condition description adjustment vector are adjustable parameters when training an original road condition analysis model; Performing a linear adjustment operation on the initial raw auxiliary guidance data and the initial vibration signal adjustment vector to determine the original road monitoring vibration signal auxiliary guidance data, wherein the linear adjustment operation is used to use the third serial number component in the vector corresponding to the original road monitoring vibration signal auxiliary guidance data as the multiplication result of the third serial number component in the initial raw auxiliary guidance data and the third serial number component in the initial vibration signal adjustment vector; Performing the linear adjustment operation on the initial original auxiliary guidance data and the initial road condition description adjustment vector to determine the initial road condition description auxiliary guidance data, wherein the linear adjustment operation is used to use the fourth sequence number component in the vector corresponding to the initial road condition description auxiliary guidance data as the multiplication result of the fourth sequence number component in the initial original auxiliary guidance data and the fourth sequence number component in the initial road condition description adjustment vector; Merging the sample road monitoring vibration signal and the auxiliary guidance data of the original road monitoring vibration signal to determine a sample vibration signal merging feature, and merging the sample road condition description set and the auxiliary guidance data of the initial road condition description to determine a sample road condition description merging feature set; Converting the sample vibration signal merged feature and the sample road condition description merged feature set to an initial feature domain, determining the sample road monitoring vibration signal feature and the sample road condition description set feature, wherein the original road condition analysis model includes the road condition description feature extraction network, the road monitoring vibration signal feature extraction network, initial road condition description auxiliary guidance data and original road monitoring vibration signal auxiliary guidance data, and the initial road condition description auxiliary guidance data and the original road monitoring vibration signal auxiliary guidance data are generated by computing the initial original auxiliary guidance data; Determining the sample matching degree between the sample road monitoring vibration signal feature and each sample road condition description feature in the sample road condition description set feature by using the original road condition analysis model; A cost parameter is calculated based on the sample matching degree, and the initial original auxiliary guidance data and the initial adjustment vector are optimized based on the cost parameter until the target road condition analysis model trained based on the original road condition analysis model is obtained.

5. The method according to claim 1, characterized in that The step of loading the target road monitoring vibration signal and the road condition description set into a previously trained target road condition analysis model to obtain target road monitoring vibration signal features and a road condition description feature set includes: On the basis that the target road condition analysis model includes a plurality of road monitoring vibration signal analysis networks, the corresponding original auxiliary guidance data and adjustment vector are configured for each of the road monitoring vibration signal analysis networks, wherein the output of a first road monitoring vibration signal analysis network among the plurality of road monitoring vibration signal analysis networks and the road monitoring vibration signal auxiliary guidance data corresponding to a second road monitoring vibration signal analysis network are combined as inputs of the second road monitoring vibration signal analysis network, and the second road monitoring vibration signal analysis network is a subsequent network of the first road monitoring vibration signal analysis network; On the basis that the target road condition analysis model includes multiple road condition analysis networks, the corresponding original auxiliary guidance data and the adjustment vector are configured for each of the road condition analysis networks, wherein the output of the first road condition analysis network among the multiple road condition analysis networks and the road condition description auxiliary guidance data corresponding to the second road condition analysis network are combined as the input of the second road condition analysis network.

6. The method according to claim 1, characterized in that The method further comprises: determining that the target road state type is an abnormal road state type set, wherein the preset road state type includes the abnormal road state type set; In response to determining that the target road monitoring vibration signal is the abnormal road monitoring vibration signal, abnormality auxiliary guidance data is generated, wherein the abnormality auxiliary guidance data is used to indicate that the target road monitoring vibration signal is an abnormal vibration signal.

7. The method according to claim 1, characterized in that Determining a road maintenance strategy for the road section to be monitored according to the road monitoring event to be processed includes: For the road monitoring surveillance image in the road monitoring event to be processed, similar images whose relevance reaches a relevance threshold are retrieved by an image retrieval model to obtain candidate surveillance images, wherein the image retrieval model is trained using historical road monitoring surveillance images as training instance data; Based on the candidate monitoring image, for the current road state type in the road monitoring event to be processed, the candidate current road state type whose relevance reaches the relevance threshold is retrieved through the current road state type to obtain the candidate road monitoring event, wherein the current road state type retrieval is trained by using the historical current road state type as training instance data; The road risk information evaluation model is used to evaluate the road monitoring event to be processed and the candidate road monitoring event to obtain risk information of the road monitoring event to be processed and risk information of each candidate road monitoring event, wherein the road risk information evaluation model is trained using monitoring events with preset labels in historical roadside monitoring data as training instance data; taking the candidate road monitoring event that matches the risk information of the road monitoring event to be processed as a road condition monitoring matching result; The road maintenance strategy for the road section to be monitored is called from a preset database according to the road condition monitoring matching result.

8. A server system, characterized in that: The method comprises a server, wherein the server is used to execute the method described in any one of claims 1 to 7.

Images