A vehicle-mounted ground penetrating radar intelligent detection and early warning method and system

Abstract

The application provides a kind of vehicle-mounted ground penetrating radar intelligent detection early warning method and system, comprising: obtaining ground penetrating radar echo data and positioning data in the process of inspection, time stamp alignment and space correlation are carried out to obtain fusion detection data;The fusion detection data is input into the preset disease identification model, and the disease identification result is obtained;According to disease identification result and positioning data, extract quantitative features, generate disease feature parameter set;Obtain the environmental parameter set corresponding to the disease location;The disease feature parameter set and the environmental parameter set are fused and input into the preset risk assessment model, the risk score is calculated and the risk level is determined, and the risk assessment result is generated;According to disease identification result, disease feature parameter set and risk assessment result, generate early warning information and output.The application can realize real-time detection, intelligent identification, quantitative evaluation and hierarchical warning of road underground disease, reduce artificial dependence, improve detection efficiency and accuracy, and provide scientific decision basis for urban lifeline safety engineering.

Description

Technical Field

[0001] This invention relates to the field of urban road underground defect detection technology, and in particular to a vehicle-mounted ground-penetrating radar intelligent detection and early warning method and system. Background Technology

[0002] Urban lifeline engineering is a vital infrastructure system for maintaining the normal operation of a city, encompassing key facilities such as roads, bridges, and underground pipelines. In recent years, with the increased intensity of underground space development and the growing prominence of aging pipelines, urban road collapses have become increasingly frequent, seriously threatening the safety of people's lives and property. The main causes of road collapses include underground pipeline leakage and erosion forming cavities, roadbed soil loss, and disturbance from underground construction projects. These hidden dangers are difficult to detect through routine inspections.

[0003] Ground-penetrating radar (GPR), as a highly efficient non-destructive testing technology, has been widely applied in the detection of underground defects in urban roads. This technology effectively identifies hidden defects such as underground cavities, voids, and loosening by analyzing radar echo signals. Integrating GPR into inspection vehicles allows for continuous scanning of the road's underground structure during normal driving, significantly improving detection efficiency and making it the mainstream method for detecting underground road defects.

[0004] However, existing vehicle-mounted ground-penetrating radar systems still have the following prominent problems: At the data acquisition level, the system only completes the data acquisition function, and cannot monitor the real-time location, running trajectory, and operating status of the detected vehicles in real time, making it difficult to achieve scientific scheduling and full-process management of detection tasks; At the data interpretation level, the raw radar data relies heavily on professional technicians to manually analyze and interpret each line, resulting in low interpretation efficiency and strong subjectivity, making it difficult to meet the data processing needs of large-scale, high-frequency inspections, leading to serious delays in the detection of defects; At the risk assessment level, existing systems only output detection data or result tables, and cannot intelligently assess and grade early warning of defect risks, resulting in a lack of scientific basis for maintenance decisions; At the information management level, detection data, defect information, and treatment records are stored in a scattered manner, lacking a unified management platform, making it difficult to form a complete road health record and full life cycle management.

[0005] Therefore, there is an urgent need in this field for a vehicle-mounted ground-penetrating radar intelligent monitoring and early warning solution to achieve the organic integration of real-time monitoring, intelligent interpretation, risk warning and information management, thereby solving the technical problems of low detection efficiency, strong interpretation subjectivity, rough assessment and information fragmentation in the existing technology. Summary of the Invention

[0006] In view of this, embodiments of the present invention provide a vehicle-mounted ground-penetrating radar intelligent detection and early warning method and system to solve the problems of low efficiency in detecting underground road defects, reliance on manual data interpretation, crude risk assessment, and fragmented information management in the prior art.

[0007] One aspect of the present invention provides an intelligent detection and early warning method for vehicle-mounted ground-penetrating radar, the method comprising: During the inspection, ground-penetrating radar echo data and positioning data of the target road area are acquired, and time stamp alignment and spatial correlation are performed to obtain fused detection data; The fused detection data is input into a preset disease identification model to obtain disease identification results. The disease identification results include at least disease type, disease area parameters, and identification confidence. The disease area parameters are used to characterize the corresponding location and range of the disease in the fused detection data. Based on the disease identification results and the location data, the quantitative features of the corresponding diseases are extracted to generate a disease feature parameter set, which is used to characterize the location, burial depth, scale and development degree of the diseases. Obtain a set of environmental parameters corresponding to the location coordinates of the disease, the set of environmental parameters being used to characterize the operating environment and surrounding facility conditions of the road area where the disease is located; The disease characteristic parameter set is fused with the environmental parameter set and input into a preset risk assessment model to calculate the risk score of the corresponding disease. Based on the risk score, the disease risk level is determined, and a risk assessment result is generated. Early warning information is generated based on the disease identification results, the disease feature parameter set, and the risk assessment results. The early warning information includes at least the disease type, disease location coordinates, risk level, and treatment suggestions. The early warning information is then output for early warning and treatment of underground diseases in the target road area.

[0008] In some embodiments of the present invention, the training steps of the preset disease identification model include: Obtain a training sample set, wherein each training sample in the training sample set includes a radar image sample and a corresponding disease type label and disease area label; The radar image samples are input into the initial disease identification model, and the disease type prediction result and disease area parameter prediction result are output. A first loss function is constructed based on the deviation between the disease type prediction result and the disease type label; a second loss function is constructed based on the deviation between the disease area parameter prediction result and the disease area label. The first loss function and the second loss function are weighted and summed to construct a joint loss function; The parameters of the initial disease identification model are updated by minimizing the joint loss function using the gradient descent algorithm. When the joint loss function satisfies the preset convergence condition or the training reaches the preset number of iterations, training stops, and the preset disease identification model is obtained.

[0009] In some embodiments of the present invention, the preset disease identification model is a deep learning model based on a convolutional neural network. The deep learning model adopts the YOLOX target detection algorithm, and the feature extraction network of the deep learning model is equipped with a spatial pyramid pooling module to enhance the feature representation of disease targets.

[0010] In some embodiments of the present invention, the extraction of quantitative features corresponding to the disease includes: The boundary box of the diseased area is determined based on the parameters of the diseased area. The boundary box is represented by the coordinates of the number of channels at the top left corner, the coordinates of the number of burial points at the top left corner, the length span of the number of channels, and the height span of the number of points. Based on the road number corresponding to the center position of the bounding box, the corresponding latitude and longitude coordinates are obtained from the positioning data as the coordinates of the disease location; The depth of the disease is calculated based on the coordinates of the burial depth point in the upper left corner, combined with the radar sampling rate and time-depth conversion coefficient. Calculate the length of the disease based on the length of the span and the spacing between the spans; When using multi-channel ground-penetrating radar, the width of the disease is calculated based on the number of channels the disease spans and the channel spacing. Based on the span of the height points, combined with the radar sampling rate and time-depth conversion coefficient, the net depth of the disease is calculated.

[0011] In some embodiments of the present invention, the preset risk assessment model employs a multi-factor coupled assessment method, and the risk score is calculated in the following manner: The area score, net depth score, and span ratio score are determined based on the disease characteristic parameter set, where the span ratio is the ratio of burial depth to length. The area score, net depth score, and coverage ratio score are weighted and summed according to preset weights to obtain the basic risk score. The basic risk score is corrected by combining the road area's operating environment and surrounding facility conditions characterized by the environmental parameter set, to obtain the final risk score.

[0012] In some embodiments of the present invention, the generation of early warning information specifically includes: Based on the disease type, the disease characteristic parameter set, and the risk assessment results, a preset expert rule base is matched to generate corresponding causal analysis and treatment suggestions; wherein, the expert rule base includes combination rules of disease type, disease characteristic parameter range, and risk level, as well as causal analysis and treatment suggestions corresponding to each combination rule; The disease type, disease location coordinates, disease characteristic parameter set, risk assessment results, cause analysis, and treatment suggestions are integrated into a standardized disease information card, and the standardized disease information card is output as the early warning information.

[0013] In some embodiments of the present invention, the method further includes: The location, driving trajectory, disease distribution points, and risk heat map of the vehicles performing the inspection are displayed in real time on the electronic map. The disease distribution points are distinguished by different identifiers according to the disease risk level, and the risk heat map is generated based on the disease density and risk level to show the spatial distribution characteristics of the disease and the risk cluster area.

[0014] In some embodiments of the present invention, the method further includes: Before inputting the fused detection data into the preset disease identification model, the fused detection data is sequentially processed by bandpass filtering, gain compensation, and noise suppression. The bandpass filtering uses a Butterworth filter, the gain compensation uses a combination of exponential and linear gain, and the noise suppression uses a moving average filter.

[0015] In some embodiments of the present invention, the method further includes: Based on the warning information, a disposal task is created in the cloud and dispatched to the responsible unit corresponding to the location coordinates of the disease; the disposal record uploaded after disposal is completed is received and stored in association with the corresponding warning information to form a disease life cycle management file.

[0016] On the other hand, the present invention also provides a vehicle-mounted ground-penetrating radar intelligent detection and early warning system, the system comprising: The vehicle-mounted data acquisition terminal system is used to collect ground-penetrating radar echo data and positioning data of the target road area during the inspection process; The real-time data transmission system is bidirectionally connected to the vehicle-mounted data acquisition terminal system for transmitting the ground-penetrating radar echo data and the positioning data; The cloud-based data management system is bidirectionally connected to both the vehicle-mounted data acquisition terminal system and the intelligent interpretation system. It is used to receive data uploaded by the vehicle-mounted data acquisition terminal system and the results output by the intelligent interpretation system, and to perform vehicle monitoring, task management, data management, visualization, and statistical analysis. The intelligent interpretation system is bidirectionally connected to both the real-time data transmission system and the cloud data management system, and is used to execute the steps of the vehicle-mounted ground-penetrating radar intelligent detection and early warning method as described in any of the above-mentioned embodiments.

[0017] The intelligent detection and early warning method and system for vehicle-mounted ground-penetrating radar provided by this invention collects ground-penetrating radar echo data and positioning data simultaneously during the inspection process and performs spatiotemporal fusion. It automatically outputs the disease type, regional parameters and confidence level using a preset disease identification model, and then extracts quantitative features such as the location, burial depth and scale of the disease. Combined with environmental parameter set, it calculates risk score and determines risk level through risk assessment model, and finally generates early warning information containing treatment suggestions, realizing closed-loop management of underground diseases of urban roads from discovery to treatment.

[0018] Furthermore, this invention employs a multi-factor coupled assessment method, which calculates a basic risk score by weighting the area, net depth, and coverage ratio of the disease according to preset weights. The basic risk score is then corrected based on environmental parameters such as road grade, traffic flow, and information on adjacent pipelines, thereby classifying the disease risk into multiple levels and achieving a reasonable leap from qualitative judgment to quantitative grading.

[0019] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the description, or may be learned by practice of the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures specifically pointed out in the description and drawings.

[0020] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description

[0021] The accompanying drawings, which are provided to further illustrate the invention and form part of this application, are not intended to limit the scope of the invention.

[0022] Figure 1 This is a flowchart illustrating the intelligent detection and early warning method for vehicle-mounted ground-penetrating radar according to an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of the architecture of a vehicle-mounted ground-penetrating radar intelligent detection and early warning system according to an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0025] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0026] Currently, the detection and monitoring of underground road defects in urban areas mainly relies on manual inspections, sporadic point detection, or periodic surveys using traditional ground-penetrating radar, which presents the following prominent problems: (1) Low detection efficiency and poor real-time performance: Traditional underground road detection often uses walking or low-speed towing ground-penetrating radar, which is slow, occupies road resources, and is difficult to achieve high-frequency, large-scale routine inspections. Data often requires manual processing afterward and cannot be uploaded and analyzed online in real time, resulting in a serious delay in the discovery of road defects.

[0027] (2) Data interpretation relies on manual labor and is highly subjective: Traditional radar image analysis relies heavily on the personal experience of experts and relies on visual identification of abnormal features such as hyperbolas. This method is not only inefficient and time-consuming, but also suffers from inconsistent interpretation standards and is prone to omissions and misjudgments, making it difficult to meet the needs of rapid processing of large-scale data.

[0028] (3) The phenomenon of information silos is serious and there is a lack of linkage: detection data, location information and historical disease records are stored in different systems and have not been effectively integrated. Managers have difficulty intuitively grasping the spatial distribution of diseases and their evolution patterns, and cannot form a closed-loop management mechanism for detection, assessment and treatment.

[0029] (4) Risk assessment is lagging and crude: Existing technology can only identify the existence of diseases, but lacks an intelligent assessment model that combines geometric features and the surrounding environment, making it difficult to scientifically sort a large number of diseases, resulting in insufficient basis for maintenance decisions.

[0030] (5) Insufficient system collaboration capability: The lack of a comprehensive platform that integrates vehicle scheduling, task assignment, data collection, intelligent interpretation and visualization results in a disconnect between front-end data collection and back-end management, making it impossible to achieve multi-vehicle collaborative operation and rapid response to emergency tasks.

[0031] In view of this, one aspect of the present invention provides an intelligent detection and early warning method for vehicle-mounted ground-penetrating radar, such as... Figure 1 As shown, the method includes the following steps: S101: During the inspection process, ground-penetrating radar echo data and positioning data of the target road area are acquired, and time stamp alignment and spatial correlation are performed to obtain fused detection data.

[0032] S102: Input the fused detection data into the preset disease identification model to obtain the disease identification results. The disease identification results include at least the disease type, disease area parameters, and identification confidence. The disease area parameters are used to characterize the corresponding location and range of the disease in the fused detection data.

[0033] S103: Based on the disease identification results and location data, extract the quantitative features of the corresponding diseases to generate a disease feature parameter set. The disease feature parameter set is used to characterize the location, burial depth, scale and development degree of the diseases.

[0034] S104: Obtain the set of environmental parameters corresponding to the location coordinates of the disease. The set of environmental parameters is used to characterize the operating environment and surrounding facility conditions of the road area where the disease is located.

[0035] S105: Integrate the disease characteristic parameter set with the environmental parameter set, input them into the preset risk assessment model, calculate the risk score of the corresponding disease, determine the disease risk level based on the risk score, and generate the risk assessment result.

[0036] S106: Generate early warning information based on the disease identification results, disease characteristic parameter set and risk assessment results. The early warning information shall include at least the disease type, disease location coordinates, risk level and treatment suggestions, and output the early warning information for early warning and treatment of underground diseases in the target road area.

[0037] In step S101, the detection vehicle simultaneously collects ground-penetrating radar echo data and positioning data while driving in the target road area. The two types of data are then time-stamped and spatially correlated to generate fused detection data with precise spatial coordinates. Simultaneously, the data is uploaded to the cloud in real time, providing a timely and reliable data source for subsequent intelligent interpretation. The fused detection data includes radar image data after time-stamp alignment and spatial correlation, along with its corresponding positioning information. Specifically, spatial correlation involves aligning the latitude and longitude coordinates in the positioning data with timestamps and then correlating them with each radar echo data point to form fused detection data with spatial coordinates.

[0038] Specifically, the vehicle-mounted data acquisition terminal system continuously collects two types of data during vehicle operation: ground-penetrating radar echo data, used to characterize the reflection features of the underground structure in the target road area; and positioning data, used to record the vehicle's spatial position in real time. To ensure strict temporal correspondence between the two types of data, the vehicle-mounted industrial control computer interpolates the timestamps of the positioning data to the sampling time of each radar data stream, ensuring that each radar echo data stream has a unique acquisition time. Based on this, the system extracts corresponding latitude and longitude coordinates and other positioning information from the positioning data using the aligned timestamps, and assigns these spatial attributes to each radar data stream, thereby generating fused detection data closely related to the road's spatial location. The vehicle-mounted industrial control computer encapsulates the radar echo data carrying latitude and longitude coordinates into data frames in chronological order and uploads them in real time to the cloud data management system via a 4G / 5G communication module and a real-time data transmission system. This allows management personnel to view the vehicle's location, trajectory, and the radar data being collected in real time at the backend.

[0039] In some embodiments, to accommodate both deep and shallow depth detection needs, the ground-penetrating radar (GPR) equipment can employ a multi-frequency antenna combination, such as configuring two frequency channels, 200MHz and 600MHz, for depth detection (0-5 meters) and shallow depth detection (0-2 meters), respectively. The radar antenna array is installed on the underside of the vehicle, enabling detection speeds exceeding 40 km / h. Positioning data is acquired via a BeiDou or GPS dual-mode positioning module, which supports RTK differential positioning mode, achieving a horizontal positioning accuracy better than 0.5 meters, and centimeter-level accuracy under suitable conditions. The preferred update frequency for positioning data is 10Hz, synchronized with the radar data acquisition frequency. Timestamp alignment is achieved by matching the timestamps of the positioning data to the sampling time of each radar data channel using a linear interpolation method. The communication module can employ a 4G or 5G industrial-grade module with an upload bandwidth of no less than 20Mbps to meet the real-time upload requirements of the radar data. In areas with poor network signal, the industrial control computer can cache data frames on a local solid-state drive, automatically resuming transmission once the network recovers, ensuring data integrity.

[0040] Thus, step S101 generates fused detection data that precisely corresponds to the spatial location and realizes real-time data uploading, providing a detection data source with coordinate information for subsequent identification of underground diseases.

[0041] In step S102, the fused detection data generated in step S101 is input into a preset disease identification model. The model identifies underground diseases in the radar image and outputs the disease type, disease area parameters, and identification confidence score. The disease area parameters describe the specific location and coverage of the disease in the radar image, and the identification confidence score quantifies the reliability of the model's judgment.

[0042] Specifically, the preset disease identification model is a deep learning-based neural network model capable of end-to-end inference on the input radar image. The model analyzes each frame of the fused detection data, extracting abnormal reflection features such as hyperbolic curves, and determines the presence of disease based on these features. If disease is detected, the model outputs the disease category, such as voids, cavities, loose areas, or water-rich areas, and provides the disease's location and extent in the image in the form of a bounding box, i.e., the disease region parameters. Furthermore, the model outputs a confidence score between 0 and 1 to indicate the reliability of the current identification result.

[0043] In some embodiments, the preset disease identification model is pre-trained using a large number of labeled samples. The training steps include: obtaining a training sample set, where each training sample includes a radar image sample and corresponding disease type and disease region labels; inputting the radar image samples into the initial disease identification model, and outputting disease type prediction results and disease region parameter prediction results; constructing a first loss function based on the deviation between the disease type prediction results and the disease type labels, and constructing a second loss function based on the deviation between the disease region parameter prediction results and the disease region labels; performing a weighted summation of the first and second loss functions to construct a joint loss function; minimizing the joint loss function using a gradient descent algorithm to update the parameters of the initial disease identification model; and stopping training when the joint loss function meets a preset convergence condition or the training reaches a preset number of iterations, thus obtaining the preset disease identification model.

[0044] In some embodiments, the preset disease identification model employs a deep learning model based on convolutional neural networks, specifically using the YOLOX target detection algorithm as its framework. To improve the detection capability of small-sized diseases in radar images, a spatial pyramid pooling module is added to the model's feature extraction network. This module can fuse feature information at different scales, strengthen the feature representation of small diseases, and thus improve the identification accuracy.

[0045] In some embodiments, to ensure the model's recognition performance, the fused detection data undergoes necessary preprocessing before being input into the model. This includes bandpass filtering, gain compensation, and noise suppression. Specifically, Butterworth filters are used for bandpass filtering to preserve effective frequency bands; a combination of exponential and linear gain is used for gain compensation to balance shallow and deep signals; and moving average filtering is used for noise suppression to reduce random interference. These preprocessing operations help highlight disease characteristics, improve the signal-to-noise ratio, and thus enhance the model's recognition accuracy.

[0046] At this point, step S102 outputs the disease type, disease area parameters, and identification confidence level, providing input basis for subsequent steps to extract quantitative features such as the location, burial depth, and scale of the disease.

[0047] In step S103, based on the disease identification results output in step S102 and the location data obtained in step S101, quantitative features of the corresponding diseases are extracted from the fused detection data to generate a disease feature parameter set. This feature parameter set is used to characterize the spatial location, burial depth, planar scale, and vertical development degree of the diseases.

[0048] Specifically, the disease identification results include disease area parameters, which are typically given in the form of bounding boxes. First, the bounding box of the disease in the radar image is determined based on the disease area parameters. This bounding box is represented by four parameters: the top-left corner channel number coordinates, the top-left corner depth point number coordinates, the length channel number span, and the height point number span. The channel number coordinates reflect the sampling channel position of the disease in the horizontal direction, the depth point number coordinates reflect the sampling point position of the top of the disease in the depth direction, the length channel number span indicates the number of channels the disease spans horizontally, and the height point number span indicates the number of sampling points the disease spans vertically. Then, using these bounding box parameters, combined with positioning data and radar acquisition parameters, the quantitative characteristics of the disease are calculated one by one: the disease location coordinates are obtained from the positioning data through the channel number corresponding to the center position of the bounding box, and the disease location coordinates are preferably latitude and longitude coordinates; the disease burial depth is calculated based on the coordinates of the burial point in the upper left corner, combined with the radar sampling rate and the time-depth conversion relationship; the disease length is calculated based on the length channel span and the horizontal distance between adjacent channels; the disease net depth is calculated based on the height point span, combined with the radar sampling rate and the time-depth conversion relationship.

[0049] In some embodiments, the specific steps for extracting quantitative features include: determining the four parameters of the bounding box based on the parameters of the disease area; obtaining the latitude and longitude coordinates as the disease location coordinates based on the track number corresponding to the center position of the bounding box; calculating the burial depth based on the coordinates of the number of burial depth points in the upper left corner, the radar sampling rate, and the time-depth conversion coefficient; calculating the length based on the length track span and track spacing; and calculating the net depth based on the height point span, the radar sampling rate, and the time-depth conversion coefficient.

[0050] In some embodiments, when using multi-channel ground-penetrating radar, the width can also be calculated based on the number of channels the defect crosses and the channel spacing. It should be noted that the time-depth conversion factor, channel spacing, and channel interval can be determined based on the actual acquisition parameters.

[0051] At this point, step S103 outputs a set of disease characteristic parameters, including the coordinates of the disease location, burial depth, length, width, and net depth, providing quantitative geometric indicators for subsequent risk assessment.

[0052] In step S104, a set of environmental parameters corresponding to the coordinates of the disease location is obtained. This set of environmental parameters is used to characterize the operating environment and surrounding facility conditions of the road area where the disease is located. These environmental factors affect the disaster risk and severity of the disease, providing an important basis for subsequent risk assessment.

[0053] Specifically, based on the location coordinates of the defect obtained in step S103, the road segment and surrounding area where the defect is located are determined, and then environmental information for that location is obtained from relevant data sources. The set of environmental parameters may include road traffic conditions, traffic load status, and the distribution and status of surrounding underground pipelines, buildings, and other facilities. Combining this environmental information with the geometric characteristics of the defect itself allows for a more accurate assessment of the potential safety risks posed by the defect.

[0054] In some embodiments, the environmental parameter set specifically includes at least one of road class, traffic flow, and adjacent pipeline information. Road class reflects the design standards and importance of the road segment, such as expressway, arterial road, secondary arterial road, or local road; different road classes correspond to different load requirements and safety risk weights. Traffic flow characterizes the vehicle density and load frequency of the road segment and can be obtained through traffic management data or real-time monitoring. Adjacent pipeline information includes the type, age, material, and potential leakage risk of underground pipelines near the defect location; this information can be obtained from urban underground pipeline databases or on-site detection data.

[0055] In some embodiments, the environmental parameter set is obtained as follows: First, based on the coordinates of the defect location, the road grade to which the location belongs is queried through a geographic information system or a road asset management system; then, the average daily traffic flow of the road segment is obtained from a traffic flow monitoring system or historical statistics; finally, pipeline data within a certain range around the defect location is extracted from the urban underground pipeline integrated management system, including information such as pipeline type, pipe age, material and historical maintenance records, and its leakage risk level is assessed.

[0056] Thus, step S104 has obtained a set of environmental parameters associated with the location of the disease, providing data support for risk assessment in step S105 by combining environmental factors.

[0057] In step S105, the disease characteristic parameter set generated in step S103 is fused with the environmental parameter set obtained in step S104, input into a preset risk assessment model, the risk score of the corresponding disease is calculated, and the disease risk level is determined based on the risk score, generating a risk assessment result. This step realizes the conversion from disease geometric characteristics and environmental factors to quantitative risk levels.

[0058] Specifically, the pre-set risk assessment model employs a multi-factor coupled assessment method, comprehensively considering the scale, burial state, and developmental stage of the road defect, as well as the operating environment and surrounding infrastructure conditions of the road area. The model first calculates a basic risk score reflecting the inherent danger of the defect based on a set of defect characteristic parameters. Then, it uses a set of environmental parameters to correct this basic score, obtaining the final risk score. Based on the risk score, the defects are classified into different risk levels. A higher risk level indicates a greater urgency and severity of road collapse, requiring priority for intervention.

[0059] In some embodiments, the risk score is calculated as follows: first, the area score, net depth score, and span ratio score are determined based on the disease characteristic parameter set, where the span ratio is the ratio of burial depth to length; then, these scores are weighted and summed according to preset weights to obtain the basic risk score; and finally, the basic risk score is corrected by combining the environmental parameter set to obtain the final risk score.

[0060] In some embodiments, the risk level is divided into five levels, with each level corresponding to a different probability of occurrence.

[0061] At this point, step S105 outputs the risk level and risk assessment results of the disease, providing a core basis for the subsequent generation of early warning information in terms of risk quantification.

[0062] In step S106, based on the disease identification results output in step S102, the disease characteristic parameter set output in step S103, and the risk assessment results output in step S105, early warning information is generated and output for use in the early warning and handling of underground diseases in the target road area. This step integrates the structured information generated in the preceding steps into early warning content that can directly guide maintenance actions, thus connecting the final link from detection to treatment.

[0063] Specifically, the early warning information includes at least the disease type, disease location coordinates, risk level, and corresponding treatment recommendations. The disease type and location coordinates are derived from disease identification results and feature parameter sets, the risk level is derived from risk assessment results, and the treatment recommendations are generated based on the disease type, feature parameters, and risk level, combined with preset rules. The early warning information can be output in a standardized format for easy storage, retrieval, and distribution, and can also be pushed to the terminals of relevant management personnel.

[0064] In some embodiments, the specific method for generating early warning information is as follows: First, based on the disease type, disease characteristic parameter set, and risk assessment results, a preset expert rule base is matched to generate corresponding causal analysis and corresponding treatment suggestions. The expert rule base contains multiple rule entries, each rule entry being associated with a disease type, a set of disease characteristic parameter ranges, and a corresponding risk level, and pre-set causal analysis and treatment suggestions for that combination. Then, the disease type, disease location coordinates, disease characteristic parameter set, risk assessment results, corresponding causal analysis, and corresponding treatment suggestions are integrated into a standardized disease information card. Finally, this information card is output as early warning information.

[0065] In some embodiments, after the early warning information is output, the system also creates a disposal task in the cloud based on the warning information and assigns the disposal task to the responsible unit corresponding to the coordinates of the disease location. After receiving the task, the maintenance personnel of the responsible unit carry out on-site handling according to the disposal suggestions. After the handling is completed, the disposal record is uploaded. The system associates and stores these disposal records with the corresponding early warning information to form a full lifecycle management archive from disease discovery to disposal archiving, which facilitates subsequent traceability and statistical analysis.

[0066] At this point, step S106 generates and outputs early warning information containing the type, location, risk level, and treatment recommendations for the road maintenance department, providing clear action instructions.

[0067] In some embodiments, the vehicle-mounted ground-penetrating radar intelligent detection and early warning method provided by the present invention further includes a visualization display step.

[0068] Specifically, the electronic map displays the real-time location of inspection vehicles, their travel trajectories, the distribution points of defects, and a risk heat map. Defect distribution points are distinguished by different identifiers based on their risk level, and the risk heat map is generated based on a combination of defect density and risk level, visually illustrating the spatial distribution characteristics of defects and areas of risk concentration. This visualization is dynamically updated based on a real-time communication protocol, ensuring synchronization between front-end data collection and back-end display.

[0069] On the other hand, the present invention also provides a vehicle-mounted ground-penetrating radar intelligent detection and early warning system, comprising: The vehicle-mounted data acquisition terminal system is used to collect ground-penetrating radar echo data and positioning data of the target road area during the inspection process; The real-time data transmission system is bidirectionally connected to the vehicle-mounted data acquisition terminal system for transmitting ground-penetrating radar echo data and positioning data. The cloud-based data management system is bidirectionally connected to both the vehicle-mounted data acquisition terminal system and the intelligent interpretation system. It is used to receive data uploaded by the vehicle-mounted data acquisition terminal system and the results output by the intelligent interpretation system, and to perform vehicle monitoring, task management, data management, visualization, and statistical analysis. The intelligent interpretation system is bidirectionally connected to both the real-time data transmission system and the cloud data management system to execute the steps of the vehicle-mounted ground-penetrating radar intelligent detection and early warning method described above.

[0070] The present invention will now be described with reference to a specific embodiment: This embodiment will elaborate on the complete technical solution of the vehicle-mounted ground-penetrating radar intelligent detection and early warning method proposed in this invention. Specifically, it will be discussed in turn from the aspects of hardware configuration of vehicle-mounted data acquisition terminal system, real-time data transmission mechanism, cloud data management architecture, structure and training method of deep learning model, quantitative calculation formula of disease characteristic parameters, multi-factor coupled risk assessment model and its weight setting, generation rules of early warning information card and system closed-loop management process.

[0071] The following is the specific technical solution followed in this embodiment: 1. System Overall Architecture The vehicle-mounted ground-penetrating radar intelligent monitoring and early warning platform of this invention adopts an end-edge-cloud collaborative architecture, consisting of four main parts: a vehicle-mounted data acquisition terminal system, a real-time data transmission system, a cloud-based data management system, and an intelligent interpretation system. The vehicle-mounted data acquisition terminal system is deployed on the detection vehicle, serving as the underlying device for data acquisition and perception; the real-time data transmission system achieves vehicle-to-cloud interconnection via 4G / 5G mobile communication networks; the cloud-based data management system is deployed on a cloud server, serving as the cloud center for data storage and business management; and the intelligent interpretation system can be deployed in the cloud to achieve intelligent real-time detection of radar data. The overall system architecture is as follows: Figure 2 As shown.

[0072] 2. Specific configuration of the vehicle-mounted data acquisition terminal system The vehicle-mounted data acquisition terminal system includes the following core components: (1) Ground penetrating radar equipment: A multi-frequency combined ground penetrating radar is adopted, with two frequency channels of 200MHz and 600MHz, which are used for detection at depths of 0 to 5 meters and shallow depths of 0 to 2 meters, respectively. The radar antenna array is installed on the bottom of the vehicle, and the detection speed is not less than 40km / h.

[0073] (2) Beidou / GPS Dual-mode Positioning Module: A high-precision Beidou / GPS dual-mode positioning terminal is adopted, with a horizontal positioning accuracy better than 0.5 meters. It supports RTK differential positioning mode, with an accuracy of up to centimeter level. The positioning data update frequency is 10Hz, which is synchronously collected with radar data to achieve accurate spatial positioning of the detection point.

[0074] (3) 4G / 5G communication module: An industrial-grade 4G / 5G communication module is adopted to ensure the reliability and continuity of data transmission. The upload bandwidth is not less than 20Mbps to meet the real-time upload requirements of radar data.

[0075] (4) Vehicle-mounted industrial control computer: responsible for radar data acquisition, positioning data fusion, local caching and preprocessing.

[0076] 3. Implementation methods of real-time data transmission systems The real-time data transmission system enables bidirectional data transmission between the vehicle-mounted data acquisition terminal and the cloud-based data management system. In the uplink direction, the system supports real-time uploading of raw radar data, positioning data, and vehicle status data; in the downlink direction, the system supports remote distribution of task commands and parameter configurations. The real-time data transmission system achieves vehicle-to-cloud interconnection via 4G / 5G mobile communication networks, ensuring the reliability and continuity of data transmission.

[0077] 4. Functional modules of the cloud-based data management system The cloud-based data management system is developed based on a microservice architecture. The functional modules are described below: Vehicle monitoring module: Displays the real-time location, trajectory, current speed, and operational status of monitored vehicles on an electronic map. It supports functions such as historical trajectory playback, detection coverage statistics, and vehicle dispatch command issuance. The system automatically triggers an alarm when a vehicle deviates from the planned route or stops abnormally.

[0078] Task Management Module: Supports the creation, editing, allocation, and scheduling of detection tasks. Task information includes the detection section, detection time, priority, and responsible vehicle. The system automatically optimizes the scheduling plan based on vehicle location and task distribution, improving detection efficiency and resource utilization.

[0079] Data Management Module: Enables full lifecycle management of radar raw data, interpretation results, damage information, and treatment records. Supports multi-dimensional searching by time, road segment, damage type, etc., and provides data version control and change tracking functions. Data storage adopts distributed object storage to ensure efficient access to massive amounts of data.

[0080] Data display module: Overlays detection trajectories, disease distribution points, and risk heat maps onto an electronic map. The risk heat map is generated based on factors such as disease density, risk level, and development trend, intuitively displaying the spatial distribution characteristics and risk cluster areas of underground road diseases.

[0081] Statistical Analysis Module: Performs statistical analysis on task management information, road underground defects information, and data management information, and displays the results on the electronic map using pie charts, line graphs, and other methods.

[0082] 5. Workflow of the Intelligent Interpretation System The intelligent interpretation system is the core technology module of this invention, and its workflow is as follows: S1: Data Preprocessing: The raw radar data undergoes preprocessing such as Butterworth bandpass filtering, gain compensation combining exponential and linear gain, and moving average noise suppression to improve data quality and signal-to-noise ratio. These preprocessing operations can be performed in real time by the onboard industrial control computer before data upload to reduce transmission noise and improve the signal-to-noise ratio; alternatively, they can be performed by the intelligent interpretation system after the data is received by the cloud-based data management system. In this embodiment, it is preferred to perform the preprocessing in the cloud to reduce the computational load on the onboard device.

[0083] S2: Deep Learning-Based Disease Identification: This technical solution employs the YOLOX target detection algorithm for high-precision detection and identification of disease targets in radar images. Addressing the core pain points of weak signals for small-sized diseases in radar images and the tendency for traditional detection algorithms to miss detections, the YOLOX network structure is specifically improved: the original YOLOX CSPDarknet backbone feature extraction network is replaced with a ResNet-50 deep residual network. ResNet-50, through residual connections and bottleneck layer design, solves the gradient vanishing problem in deep network training, enabling more thorough extraction of multi-scale features of diseases in radar images. During the replacement process, the output feature layers of ResNet-50 (layer 2, layer 3, and layer 4) correspond to the three scale feature maps required by the YOLOX detection head. Layer 2 corresponds to 8x downsampling, layer 3 to 16x downsampling, and layer 4 to 32x downsampling. For small-sized lesions, the feature layer adaptation was optimized: the high-resolution characteristics of shallow features of ResNet-50 were preserved to avoid the loss of small lesion features during downsampling; a Spatial Pyramid Pooling (SPPF) module was added to the shallow feature layer to fuse multi-scale spatial information and enhance the feature representation of small-sized lesions; the number of input channels of the YOLOX Neck feature fusion module was adjusted to adapt to the 512, 1024, and 2048 channel outputs of ResNet-50 to ensure the effectiveness of feature fusion. The model input is a preprocessed radar B-scan image, and the output is the lesion category and confidence score. The lesion categories include voids, cavities, and loosening.

[0084] S3: Disease Feature Extraction: Extract feature parameters such as location coordinates, burial depth, length, width, and net depth of the identified diseased areas.

[0085] The detected diseased area is a rectangular frame, through The expression, where X is the current channel number and Y is the number of points corresponding to the current burial depth. The length span of the disease is the number of lesions. This represents the number of points corresponding to the net depth; because the radar collects coordinate data simultaneously during the data acquisition process, each data point corresponds to a set of latitude and longitude coordinates. Location coordinate parameter acquisition method: Read the coordinate data of the channel number corresponding to the center position of the rectangular frame of the diseased area to obtain the location coordinate information of the diseased area. The channel number corresponding to the center position is... ; Methods for obtaining burial depth parameters: ,in, This is the time-depth conversion factor, for example, set to 0.05 m / ns based on the dielectric constant of the road material.

[0086] like If the sampling rate is 4G, then the burial depth is ; How to obtain the length parameter: ,like , Then the length is ; The width parameter is obtained as follows: Width = Number of channels in the defect * Channel spacing. For example, if the defect spans 6 channels in a 3D ground-penetrating radar and the channel spacing is 0.12m, then the width of the defect is... ; The method for obtaining the net depth parameter is similar to that for burial depth: ,in, For example, the time-depth conversion factor, If the sampling rate is 4G, then the net depth is ; S4: Risk assessment: A multi-factor coupled assessment method is used to assess the disease risk level.

[0087] The risk level assessment factors mainly include the diseased area, net depth, and coverage ratio, with weights of 0.3, 0.35, and 0.35 respectively. Risk assessment is conducted in accordance with relevant industry standards.

[0088] The risk levels are divided into 5 levels: A, B, C, D, and E, as shown in Table 1: Table 1 Risk Level Classification Standards S5: Disease Information Card Generation: Automatically generates standardized disease information cards, including disease number, location information, disease type, characteristic parameters such as burial depth, length, width, and net depth, risk level, cause analysis, treatment recommendations, responsible department, and deadline for treatment. The information cards use a unified field format, facilitating storage, retrieval, circulation, and push to management terminals or mobile devices, providing a direct and actionable basis for maintenance decisions.

[0089] 6. Alternative technical solutions In addition to the above-described embodiments, the present invention can also be implemented using various other alternative solutions, which will now be described from five aspects: system architecture, data transmission, AI interpretation, disease feature extraction and evaluation, visualization and interaction.

[0090] (1) Alternative solutions at the system architecture level The above embodiments adopt an end-edge-cloud collaborative architecture, with the vehicle terminal mainly responsible for data collection and simple fusion, and intelligent interpretation mainly deployed in the cloud.

[0091] An alternative is an edge-computing-first architecture. This approach partially or entirely deploys the deep learning-based disease identification module from the intelligent interpretation subsystem onto an onboard industrial computer. This is achieved using a more powerful onboard computing unit, such as an industrial computer equipped with a GPU module. The radar data undergoes AI inference on-board, directly outputting disease identification results, with only the identified disease information and key maps uploaded to the cloud. The advantages of this approach are a significant reduction in uplink data transmission, eliminating the need to transmit the entire original data, higher real-time interpretation down to millisecond levels, and no impact on basic detection functions even during network interruptions.

[0092] Another alternative is a pure cloud-based thin client architecture. This solution greatly simplifies the functions of the in-vehicle terminal, making it a pure data acquisition and relay unit. The implementation involves the in-vehicle industrial control computer not performing any data fusion or caching; it directly packages the raw radar and GPS data and pushes it to the cloud in real time via a dedicated 5G mobile communication network. All preprocessing, data alignment, and intelligent interpretation are completed in the cloud. The advantages of this solution are the lowest cost for the in-vehicle terminal, simple maintenance, and rapid algorithm updates that only need to be performed in the cloud.

[0093] (2) Alternative solutions at the data transmission layer The above embodiments achieve real-time uploading based on the fourth-generation or fifth-generation mobile communication public network.

[0094] An alternative is a multi-mode adaptive transmission strategy. This approach doesn't pursue a one-size-fits-all real-time upload of all data. Instead, the vehicle-mounted terminal dynamically switches transmission modes based on network signal quality and task urgency. Specifically, in 5G or Wi-Fi environments, a high-speed mode is used to upload all raw data and decoding results; in 4G environments, a standard-speed mode is used, prioritizing the upload of compressed critical data and AI decoding results; and when the network is unavailable, an offline mode is used to store data locally, automatically synchronizing it once the network is restored. This solution addresses the issues of poor signal in tunnels and remote areas, ensures data integrity, and saves on data usage costs.

[0095] (3) Alternative solutions at the intelligent interpretation level The above embodiments are based on a convolutional neural network model improved from ResNet-50 for image recognition.

[0096] An alternative is a temporal image fusion model based on the Transformer architecture. This approach changes the core of the deep learning algorithm, treating radar data not simply as two-dimensional images, but as a spatiotemporal sequence of data. It employs a visual Transformer or temporal convolutional network to simultaneously process the radar waveform sequence and the continuous motion trajectory of the vehicle, directly mining defect features from the underlying data and enabling earlier detection of subtle anomalies.

[0097] Another alternative is a semi-supervised or self-supervised learning framework. This approach changes the way models are trained and applied, reducing the reliance on massive amounts of labeled data. Specifically, it utilizes a large amount of unlabeled radar data for self-supervised pre-training to learn the intrinsic representations of the data. Afterward, only a small number of labeled samples are needed to fine-tune the model to adapt to new regions and new disease types. Simultaneously, a human feedback mechanism is introduced to allow the system to continuously evolve during use.

[0098] (4) Alternative solutions for disease feature extraction and assessment The above embodiments extract the geometric parameters of the disease based on image segmentation algorithms.

[0099] An alternative approach is feature extraction based on 3D geological body modeling. This approach upgrades 2D image analysis to 3D spatial reconstruction. It utilizes multi-channel, multi-frequency radar data, combined with continuous, dense sampling from vehicles, to generate a 3D model of the underground anomaly in the cloud through interpolation and 3D reconstruction algorithms. Using this approach, the scale, orientation, and volume of the anomaly can be calculated more accurately, leading to more precise risk assessment.

[0100] Another alternative is knowledge graph-driven risk assessment. This approach upgrades the risk assessment model from traditional formula-based calculations to graph-based calculations. It involves constructing a knowledge graph of urban underground space, using defects, pipelines, roads, historical accidents, and other elements as nodes. When a new defect is discovered, instead of simply scoring it using a few factors, the system retrieves information about its associated pipelines, construction disturbances, etc., from the knowledge graph, and uses reasoning to arrive at a more accurate causal analysis and risk level.

[0101] (5) Alternatives for visualization and interaction The above embodiments overlay and display the detection trajectory, disease distribution points, and risk heat map on a two-dimensional electronic map.

[0102] An alternative is digital twins combined with augmented reality (AR) visualization. This approach upgrades two-dimensional displays to a three-dimensional immersive experience. On the platform side, a digital twin scenario of urban roads is constructed, displaying the defects underground as a three-dimensional model, supporting cross-sectional viewing. On the field side, AR applications for mobile phones or tablets are developed. When managers arrive on site, they can use cameras to overlay virtual defect outlines and risk levels onto the real road surface, achieving a perspective-like effect.

[0103] All the above alternative solutions fall within the protection scope of this invention, and those skilled in the art can flexibly select and combine them according to actual application needs.

[0104] In summary, the vehicle-mounted ground-penetrating radar intelligent detection and early warning method and system provided by this invention simultaneously collects ground-penetrating radar echo data and positioning data during the inspection process and performs spatiotemporal fusion. It automatically outputs the disease type, regional parameters and confidence level using a preset disease identification model, and then extracts quantitative features such as the location, burial depth and scale of the disease. Combined with environmental parameter sets, it calculates risk scores and determines risk levels through a risk assessment model, and finally generates early warning information containing treatment suggestions, realizing closed-loop management of underground diseases on urban roads from discovery to treatment.

[0105] Furthermore, this invention employs a multi-factor coupled assessment method, which calculates a basic risk score by weighting the area, net depth, and coverage ratio of the disease according to preset weights. The basic risk score is then corrected based on environmental parameters such as road grade, traffic flow, and information on adjacent pipelines, thereby classifying the disease risk into multiple levels and achieving a reasonable leap from qualitative judgment to quantitative grading.

[0106] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.

[0107] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0108] In this invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0109] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent detection and early warning using vehicle-mounted ground-penetrating radar, characterized in that, The method includes: During the inspection, ground-penetrating radar echo data and positioning data of the target road area are acquired, and time stamp alignment and spatial correlation are performed to obtain fused detection data; The fused detection data is input into a preset disease identification model to obtain disease identification results. The disease identification results include at least disease type, disease area parameters, and identification confidence. The disease area parameters are used to characterize the corresponding location and range of the disease in the fused detection data. Based on the disease identification results and the location data, the quantitative features of the corresponding diseases are extracted to generate a disease feature parameter set, which is used to characterize the location, burial depth, scale and development degree of the diseases. Obtain a set of environmental parameters corresponding to the location coordinates of the disease, the set of environmental parameters being used to characterize the operating environment and surrounding facility conditions of the road area where the disease is located; The disease characteristic parameter set is fused with the environmental parameter set and input into a preset risk assessment model to calculate the risk score of the corresponding disease. Based on the risk score, the disease risk level is determined, and a risk assessment result is generated. Early warning information is generated based on the disease identification results, the disease feature parameter set, and the risk assessment results. The early warning information includes at least the disease type, disease location coordinates, risk level, and treatment suggestions. The early warning information is then output for early warning and treatment of underground diseases in the target road area.

2. The intelligent detection and early warning method for vehicle-mounted ground-penetrating radar according to claim 1, characterized in that, The training steps for the preset disease identification model include: Obtain a training sample set, wherein each training sample in the training sample set includes a radar image sample and a corresponding disease type label and disease area label; The radar image samples are input into the initial disease identification model, and the disease type prediction result and disease area parameter prediction result are output. A first loss function is constructed based on the deviation between the disease type prediction result and the disease type label; a second loss function is constructed based on the deviation between the disease area parameter prediction result and the disease area label. The first loss function and the second loss function are weighted and summed to construct a joint loss function; The parameters of the initial disease identification model are updated by minimizing the joint loss function using the gradient descent algorithm. When the joint loss function satisfies the preset convergence condition or the training reaches the preset number of iterations, training stops, and the preset disease identification model is obtained.

3. The intelligent detection and early warning method for vehicle-mounted ground-penetrating radar according to claim 1, characterized in that, The preset disease identification model is a deep learning model based on convolutional neural networks. The deep learning model adopts the YOLOX target detection algorithm. The feature extraction network of the deep learning model is equipped with a spatial pyramid pooling module to enhance the feature representation of disease targets.

4. The intelligent detection and early warning method for vehicle-mounted ground-penetrating radar according to claim 1, characterized in that, The extraction of quantitative features corresponding to the disease includes: The boundary box of the diseased area is determined based on the parameters of the diseased area. The boundary box is represented by the coordinates of the number of channels at the top left corner, the coordinates of the number of burial points at the top left corner, the length span of the number of channels, and the height span of the number of points. Based on the road number corresponding to the center position of the bounding box, the corresponding latitude and longitude coordinates are obtained from the positioning data as the coordinates of the disease location; The depth of the disease is calculated based on the coordinates of the burial depth point in the upper left corner, combined with the radar sampling rate and time-depth conversion coefficient. Calculate the length of the disease based on the length of the span and the spacing between the spans; When using multi-channel ground-penetrating radar, the width of the disease is calculated based on the number of channels the disease spans and the channel spacing. Based on the span of the height points, combined with the radar sampling rate and time-depth conversion coefficient, the net depth of the disease is calculated.

5. The intelligent detection and early warning method for vehicle-mounted ground-penetrating radar according to claim 1, characterized in that, The preset risk assessment model employs a multi-factor coupled assessment method, and the risk score is calculated in the following manner: The area score, net depth score, and span ratio score are determined based on the disease characteristic parameter set, where the span ratio is the ratio of burial depth to length. The area score, net depth score, and coverage ratio score are weighted and summed according to preset weights to obtain the basic risk score. The basic risk score is corrected by combining the road area's operating environment and surrounding facility conditions characterized by the environmental parameter set, to obtain the final risk score.

6. The intelligent detection and early warning method for vehicle-mounted ground-penetrating radar according to claim 1, characterized in that, The generation of early warning information specifically includes: Based on the disease type, the disease characteristic parameter set, and the risk assessment results, a preset expert rule base is matched to generate corresponding causal analysis and treatment suggestions; wherein, the expert rule base includes combination rules of disease type, disease characteristic parameter range, and risk level, as well as causal analysis and treatment suggestions corresponding to each combination rule; The disease type, disease location coordinates, disease characteristic parameter set, risk assessment results, cause analysis, and treatment suggestions are integrated into a standardized disease information card, and the standardized disease information card is output as the early warning information.

7. The intelligent detection and early warning method for vehicle-mounted ground-penetrating radar according to claim 1, characterized in that, The method further includes: The location, driving trajectory, disease distribution points, and risk heat map of the vehicles performing the inspection are displayed in real time on the electronic map. The disease distribution points are distinguished by different identifiers according to the disease risk level, and the risk heat map is generated based on the disease density and risk level to show the spatial distribution characteristics of the disease and the risk cluster area.

8. The intelligent detection and early warning method for vehicle-mounted ground-penetrating radar according to claim 1, characterized in that, The method further includes: Before inputting the fused detection data into the preset disease identification model, the fused detection data is sequentially processed by bandpass filtering, gain compensation, and noise suppression. The bandpass filtering uses a Butterworth filter, the gain compensation uses a combination of exponential and linear gain, and the noise suppression uses a moving average filter.

9. The intelligent detection and early warning method for vehicle-mounted ground-penetrating radar according to claim 1, characterized in that, The method further includes: Based on the warning information, a disposal task is created in the cloud and dispatched to the responsible unit corresponding to the location coordinates of the disease; the disposal record uploaded after disposal is completed is received and stored in association with the corresponding warning information to form a disease life cycle management file.

10. A vehicle-mounted ground-penetrating radar intelligent detection and early warning system, characterized in that, The system includes: The vehicle-mounted data acquisition terminal system is used to collect ground-penetrating radar echo data and positioning data of the target road area during the inspection process; The real-time data transmission system is bidirectionally connected to the vehicle-mounted data acquisition terminal system for transmitting the ground-penetrating radar echo data and the positioning data; The cloud-based data management system is bidirectionally connected to both the vehicle-mounted data acquisition terminal system and the intelligent interpretation system. It is used to receive data uploaded by the vehicle-mounted data acquisition terminal system and the results output by the intelligent interpretation system, and to perform vehicle monitoring, task management, data management, visualization, and statistical analysis. The intelligent interpretation system is bidirectionally connected to the real-time data transmission system and the cloud data management system, and is used to execute the steps of the vehicle-mounted ground-penetrating radar intelligent detection and early warning method as described in any one of claims 1 to 9.

Images