Data processing method and system for intelligent road detection and related equipment

Abstract

The invention discloses a data processing method and system for intelligent road detection and related equipment, and relates to the technical field of traffic engineering and artificial intelligence crossing. In the road detection process, multi-sensor fusion equipment with dynamic adjustment is adopted to collect multi-source data; the real-time environment perception model based on deep reinforcement learning dynamically optimizes the parameters of the multi-modal sensor, and improves the data acquisition quality in severe weather and environment. Millimeter-level alignment of multi-source data is realized by adopting a spatio-temporal model, and a generative adversarial network is introduced to enhance the multi-source data after spatio-temporal alignment so as to supplement missing data details and improve data quality. Disease intelligent identification is carried out on multi-source data through a sample learning model and a meta-learning algorithm, the sample learning model solves the problem of insufficient labeling samples of micro-disease data, the meta-learning algorithm is quickly generalized on the basis of a small amount of labeling data, micro-diseases such as early cracks and hidden road structure layer defects are identified, and the accuracy of identification is improved. And the accuracy of road disease identification is improved.

Description

Technical Field

[0001] The present application relates to the interdisciplinary field of traffic engineering and artificial intelligence, and more specifically, to a data processing method, system, and related equipment for intelligent road detection. Background Art

[0002] Road inspection is an important part of ensuring road safety and efficient operation, including inspections of road thickness, flatness, deflection, structural strength and other aspects.

[0003] Road inspection involves road damage analysis. Existing data collection processes involve degradation of image quality due to environmental interference, such as rainy days and nighttime conditions. This leads to errors in road damage analysis and reduces the accuracy of road damage identification.

[0004] Therefore, how to improve the accuracy of road damage identification is an urgent problem to be solved in this application. Summary of the Invention

[0005] In view of this, the present application discloses a data processing method, system and related equipment for intelligent road detection, aiming to improve the accuracy of road disease identification.

[0006] In order to achieve the above purpose, the disclosed technical solutions are as follows:

[0007] In a first aspect, the present application discloses a data processing method for intelligent road detection, the method comprising:

[0008] Collect multi-source data through multi-sensor fusion equipment with dynamic adjustment;

[0009] Performing spatiotemporal alignment on the multi-source data at a preset spatiotemporal accuracy;

[0010] Generate adversarial networks to enhance the spatiotemporal aligned multi-source data to supplement missing data details;

[0011] Through the sample learning model and meta-learning algorithm, the enhanced multi-source data is used to perform intelligent disease identification and obtain the identification results;

[0012] If the recognition result indicates that the disease is recognized, the disease is analyzed through the knowledge graph to obtain the cause of the disease analysis;

[0013] The road condition prediction model constructed by the long short-term memory network and the graph neural network is used to dynamically predict the causes of the disease analysis and obtain the prediction results;

[0014] An optimal maintenance strategy is automatically generated based on the prediction results and maintenance resource constraints.

[0015] Preferably, the collecting of multi-source data by a multi-sensor fusion device with dynamic adjustment includes:

[0016] Dynamically adjust sensor parameters of a multi-sensor fusion device based on a reinforcement learning model and real-time environmental parameters; wherein the multi-sensor fusion device includes at least an infrared camera with a preset clarity, a lidar with a preset accuracy, and a ground-penetrating radar array; the sensor parameters include at least camera exposure parameters and radar scanning frequency;

[0017] Laser point cloud, radar echo and image data are collected through dynamically adjusted multi-sensor fusion equipment to complete the collection of multi-source data.

[0018] Preferably, performing spatiotemporal alignment on the multi-source data at a preset spatiotemporal accuracy includes:

[0019] With millimeter-level spatial accuracy and millisecond-level temporal accuracy, the spatiotemporal Transformer model is used to extract the spatiotemporal features of laser point clouds, radar data, and image data from multi-source data.

[0020] The attention mechanism is used to calculate the matching relationship between the spatiotemporal features of the laser point cloud, the spatiotemporal features of the radar data, and the spatiotemporal features of the image data; wherein the matching relationship represents the spatiotemporal mapping relationship between each data in the multi-source data;

[0021] In three-dimensional space, data fusion is performed according to the matching relationship to complete the spatiotemporal alignment of the multi-source data.

[0022] Preferably, the method of enhancing the spatiotemporally aligned multi-source data by generating an adversarial network to supplement missing data details includes:

[0023] The discriminator in the generative adversarial network is used to judge the authenticity of the multi-source data after spatiotemporal alignment, and the generator in the generative adversarial network is used to supplement the missing data details of the multi-source data, so as to complete the process of enhancing the multi-source data after spatiotemporal alignment through the generative adversarial network.

[0024] Preferably, if the recognition result indicates that a disease is recognized, analyzing the disease through a knowledge graph to obtain a cause of the disease analysis includes:

[0025] If the recognition result indicates that a disease is identified, extracting preset disease knowledge from a knowledge graph; wherein the knowledge graph is constructed based on road disease knowledge; the road disease knowledge at least includes the cause of the disease, its development pattern, and treatment standards;

[0026] The disease analysis cause reasoning is performed on the disease using the preset disease knowledge to obtain the disease analysis cause.

[0027] Preferably, the road condition prediction model constructed by the long short-term memory network and the graph neural network dynamically predicts the cause of the disease analysis to obtain prediction results, including:

[0028] By combining the long short-term memory network and graph neural network in the road condition prediction model, the temporal characteristics and spatial correlation of the corresponding data of the disease analysis cause are analyzed;

[0029] By integrating the time series features, the spatial associations, historical detection data, meteorological data and traffic flow data, the development trend of road diseases and changes in carrying capacity at a preset future time are analyzed to complete the process of dynamic prediction of the causes of the disease analysis and obtain prediction results.

[0030] Preferably, it also includes:

[0031] In the process of implementing the optimal maintenance strategy, the value of data assets is updated in real time through a multi-dimensional dynamic evaluation model, and blockchain technology is combined to achieve data asset ownership confirmation and traceability.

[0032] A second aspect of the present application discloses a data processing system for intelligent road detection, the system comprising:

[0033] An acquisition unit for collecting multi-source data through a multi-sensor fusion device with dynamic adjustment;

[0034] A spatiotemporal alignment unit, configured to perform spatiotemporal alignment on the multi-source data at a preset spatiotemporal accuracy;

[0035] The enhancement unit is used to enhance the multi-source data after spatiotemporal alignment through a generative adversarial network to supplement the missing data details;

[0036] Intelligent recognition unit, used to perform intelligent disease recognition on enhanced multi-source data through sample learning models and meta-learning algorithms to obtain recognition results;

[0037] an analysis unit configured to analyze the disease through a knowledge graph to obtain a cause of the disease if the recognition result indicates that the disease has been identified;

[0038] A dynamic prediction unit is used to dynamically predict the cause of the disease analysis by using a road condition prediction model constructed by a long short-term memory network and a graph neural network to obtain a prediction result;

[0039] The automatic generation unit is used to automatically generate an optimal maintenance strategy based on the prediction results and maintenance resource constraints.

[0040] A third aspect of the present application discloses a storage medium, which includes stored instructions, wherein when the instructions are executed, the device where the storage medium is located is controlled to execute the data processing method for intelligent road detection as described in any one of the first aspects.

[0041] The fourth aspect of the present application discloses an electronic device comprising a memory and one or more instructions, wherein the one or more instructions are stored in the memory and configured to be executed by one or more processors as the data processing method for intelligent road detection as described in any one of the first aspects.

[0042] It can be seen from the above technical solution that the present application discloses a data processing method, system and related equipment for intelligent road detection, which collects multi-source data through a multi-sensor fusion device with dynamic adjustment, and performs spatiotemporal alignment on the multi-source data under preset spatiotemporal accuracy. The multi-source data after spatiotemporal alignment is enhanced by a generative adversarial network to supplement the missing data details, and the enhanced multi-source data is intelligently identified for defects through a sample learning model and a meta-learning algorithm to obtain an identification result. If the identification result indicates that a defect has been identified, the defect is analyzed through a knowledge graph to obtain the cause of the defect analysis. The road condition prediction model constructed by a long short-term memory network and a graph neural network is used to dynamically predict the cause of the defect analysis to obtain a prediction result. According to the prediction result and maintenance resource constraints, the optimal maintenance strategy is automatically generated.

[0043] The beneficial effects of this application are as follows: During road detection, a multi-sensor fusion device with dynamic adjustment is used to collect multi-source data. A real-time environmental perception model based on deep reinforcement learning is used to dynamically optimize the parameters of multimodal sensors to improve the quality of data collection in severe weather and environments. A spatiotemporal model is used to achieve millimeter-level alignment of multi-source data, and a generative adversarial network is introduced to enhance the multi-source data after spatiotemporal alignment to supplement missing data details and improve data quality. The enhanced multi-source data is intelligently identified for defects through a sample learning model and a meta-learning algorithm. The sample learning model solves the problem of insufficient labeled samples for minor defect data. Through the meta-learning algorithm, it can quickly generalize on the basis of a small amount of labeled data to identify minor defects such as early cracks below 0.5 mm and hidden road structure layer defects, thereby improving the accuracy of road defect identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0045] Figure 1 A flowchart of a data processing method for intelligent road detection disclosed in an embodiment of the present application;

[0046] Figure 2 This is a data fusion flow chart of the spatiotemporal Transformer model disclosed in the embodiments of this application;

[0047] Figure 3 A schematic diagram of the process of generating the knowledge graph disclosed in the embodiments of this application;

[0048] Figure 4 This is an architecture diagram of the intelligent road detection data asset management system based on a large model disclosed in an embodiment of this application;

[0049] Figure 5 This is a schematic diagram of the structure of a data processing system for intelligent road detection disclosed in an embodiment of the present application;

[0050] Figure 6 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0052] In this application, the terms "comprises," "comprising," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0053] As can be seen from the background technology, in the existing road detection data collection process, traditional collection equipment is subject to environmental interference, which will lead to a decrease in the quality of road detection image data, resulting in errors in road disease analysis, thereby reducing the accuracy of road disease identification.

[0054] To address the above-mentioned issues, the present application discloses a data processing method, system, and related equipment for intelligent road detection. During the road detection process, a multi-sensor fusion device with dynamic adjustment is used to collect multi-source data. A real-time environmental perception model based on deep reinforcement learning is used to dynamically optimize the parameters of multimodal sensors to improve the quality of data collection in adverse weather and environments. A spatiotemporal model is used to achieve millimeter-level alignment of multi-source data, and a generative adversarial network is introduced to enhance the multi-source data after spatiotemporal alignment to supplement missing data details and improve data quality. The enhanced multi-source data is intelligently identified for defects using a sample learning model and a meta-learning algorithm. The sample learning model solves the problem of insufficient labeled samples for minor defect data. The meta-learning algorithm rapidly generalizes based on a small amount of labeled data to identify minor defects such as early cracks less than 0.5 mm and hidden road structure layer defects, thereby improving the accuracy of road defect identification. The specific implementation method is described in detail in the following embodiments.

[0055] It should be noted that the data processing method, system, and related equipment for intelligent road detection provided in this application can be used in the interdisciplinary field of traffic engineering and artificial intelligence to achieve intelligent management of road detection data throughout its lifecycle. The above is merely illustrative and does not limit the application areas of the data processing method, system, and related equipment for intelligent road detection provided in this application.

[0056] All models used in this application are open source models.

[0057] refer to Figure 1 As shown in FIG, a data processing method for intelligent road detection disclosed in an embodiment of the present application is provided. The data processing method for intelligent road detection mainly includes the following steps:

[0058] S101: Collect multi-source data through a multi-sensor fusion device with dynamic adjustment.

[0059] Among them, multi-sensor fusion equipment can be deployed on road detection vehicles. Multi-sensor fusion equipment includes but is not limited to high-definition infrared cameras, high-precision lidar, ground-penetrating radar arrays, etc.

[0060] A real-time environmental perception model based on a deep reinforcement learning algorithm (Proximal Policy Optimization, PPO) dynamically optimizes multimodal sensor parameters. At night or in inclement weather, the large model automatically optimizes camera exposure parameters and radar scanning frequency to ensure data collection quality. This improves road data collection quality in inclement weather, for example, increasing the signal-to-noise ratio of nighttime images by at least 40%.

[0061] Multi-sensor fusion equipment is used to collect multi-source data, including road data. A reinforcement learning model is used to dynamically adjust sensor parameters based on real-time environmental parameters to ensure data quality. Simultaneously, a lightweight Transformer model is used to perform noise reduction, format normalization, and feature extraction on the edge. For example, real-time dehazing and enhancement processing is performed on image data, and laser point cloud data is rapidly filtered and compressed to reduce data transmission pressure.

[0062] The specific process of collecting multi-source data through a multi-sensor fusion device with dynamic adjustment is shown in A1-A2:

[0063] A1: Dynamically adjust the sensor parameters of the multi-sensor fusion device based on the reinforcement learning model and real-time environmental parameters.

[0064] Among them, real-time environmental parameters include light intensity, rainfall, vehicle speed, etc.

[0065] The multi-sensor fusion equipment includes at least an infrared camera that meets the preset clarity, a lidar that meets the preset accuracy, and a ground-penetrating radar array; the sensor parameters include at least camera exposure parameters and radar scanning frequency.

[0066] A2: Collect laser point cloud, radar echo and image data through dynamically adjusted multi-sensor fusion equipment to complete multi-source data collection.

[0067] Multi-sensor fusion equipment with a built-in edge computing unit is deployed on road inspection vehicles. Once the system is activated, a large reinforcement learning model dynamically adjusts sensor parameters such as the ISO and shutter speed of the high-definition infrared camera and the scanning angle and frequency of the lidar based on real-time environmental parameters such as light intensity, rainfall, and vehicle speed.

[0068] The built-in edge computing unit receives sensor data in real time, denoises and enhances the image data through a lightweight Transformer model, voxelizes and filters the laser point cloud data, and compresses the processed data to 1 / 5 of the original data volume. The compressed data, i.e., multi-source data, can be transmitted to the data center via the 5G network.

[0069] S102: Performing spatiotemporal alignment on multi-source data at a preset spatiotemporal accuracy.

[0070] The preset space-time accuracy includes but is not limited to millimeter-level space accuracy and millisecond-level time accuracy.

[0071] Build a large spatiotemporal Transformer model to achieve accurate spatiotemporal alignment of multi-source data with millimeter-level spatial accuracy and millisecond-level temporal accuracy.

[0072] In S102, after receiving multi-source data, the data center uses deep multi-source data fusion technology to perform spatiotemporal alignment of the multi-source data with millimeter-level spatial accuracy and millisecond-level temporal precision. This means that data fusion is completed with millimeter-level spatial accuracy and millisecond-level temporal precision. Specifically, the process of spatiotemporal alignment of multi-source data at the preset spatiotemporal accuracy is shown in Figures A1-A3.

[0073] A1: With millimeter-level spatial accuracy and millisecond-level temporal accuracy, the spatiotemporal Transformer model is used to extract the spatiotemporal features of laser point clouds, radar data, and image data from multi-source data.

[0074] A2: The attention mechanism is used to calculate the matching relationship between the spatiotemporal features of the laser point cloud, the spatiotemporal features of the radar data, and the spatiotemporal features of the image data. The matching relationship represents the spatiotemporal mapping relationship between each data in the multi-source data.

[0075] A3: In three-dimensional space, data fusion is performed based on matching relationships to complete the spatiotemporal alignment of multi-source data.

[0076] By learning the spatiotemporal mapping relationship between laser point clouds, radar data, and image data, data fusion is achieved with millimeter-level spatial accuracy and millisecond-level temporal precision. For example, the location of underground cavities detected by radar can be precisely matched with corresponding road surface images to form a three-dimensional visualization of disease information.

[0077] After the data center receives multi-source data, the spatiotemporal Transformer model performs spatiotemporal alignment of the laser point cloud, radar echo, and image data. The spatiotemporal Transformer model first extracts the spatiotemporal features of each data, calculates the matching relationship between the data through the attention mechanism, and completes data fusion in three-dimensional space. The specific data fusion process of the spatiotemporal Transformer model is as follows: Figure 2 shown.

[0078] Figure 2 In the ,Transformer model collects multi-source data with a dynamically adjusted multi-sensor fusion device, such as aurora point cloud data, radar echo data, and image data, and extracts spatiotemporal features respectively;

[0079] The Transformer model calculates the matching relationship of various spatiotemporal features through the attention mechanism;

[0080] The Transformer model completes data fusion in three-dimensional space and sends the fused data to the Generative Adversarial Network (GAN).

[0081] S103: Enhance the multi-source data after spatiotemporal alignment through the GAN network to supplement the missing data details.

[0082] In S103, the authenticity of the multi-source data after spatiotemporal alignment is judged by the discriminator in the GAN network, and the missing data details of the multi-source data are supplemented by the generator in the generative adversarial network, so as to complete the process of enhancing the multi-source data after spatiotemporal alignment through the generative adversarial network.

[0083] A generative adversarial network optimizes the fused data, employing a GAN network to enhance the quality of the fused data. Through adversarial training, missing data details are supplemented. For example, images of occluded road cracks are generated based on the image features of adjacent areas, improving data integrity and usability. For another example, for road areas obscured by shadows, the generator in the adversarial network generates reasonable image content based on the surrounding road texture features, thereby supplementing missing data details and improving data integrity. Experimental verification demonstrates that the accuracy of supplementing missing data details, such as crack images in occluded areas, exceeds 93%.

[0084] S104: Intelligently identify diseases on the enhanced multi-source data using the sample learning model and the meta-learning algorithm (Model-Agnostic Meta-Learning, MAML) to obtain identification results.

[0085] In S104, a large model architecture based on small-sample learning was developed to address the issue of insufficient labeled samples for minor road defects. This architecture, combined with a meta-learning algorithm, can identify minor road defects based on a small amount of labeled data. This meta-learning algorithm rapidly generalizes from a small amount of labeled data and can identify minor defects such as early cracks less than 0.5 mm and hidden road structural defects.

[0086] The collected road images and point cloud data are input into a large model based on sample learning. The sample learning model quickly learns the characteristics of minor defects through meta-learning and identifies early cracks, tiny potholes, etc.

[0087] S105: If the recognition result indicates that the disease is recognized, the disease is analyzed through the knowledge graph to obtain the cause of the disease analysis.

[0088] In S105, if the recognition result indicates that a disease is recognized, preset disease knowledge is extracted from the knowledge graph, and the disease analysis cause reasoning is performed on the disease using the preset disease knowledge to obtain the disease analysis cause, thereby completing the semantic understanding implementation.

[0089] Among them, the knowledge graph is constructed through road disease knowledge; road disease knowledge at least includes the causes of diseases, development laws and treatment standards.

[0090] The specific knowledge graph construction process is as follows Figure 3 shown.

[0091] Figure 3 In the process of acquiring historical detection data, meteorological data, and traffic flow data;

[0092] Send historical inspection data, meteorological data, traffic flow data and other data to the road disease knowledge extraction model 1 (BERT), and respond to the expert rules obtained by domain experts based on the above data;

[0093] Obtain road disease entities and relationships through road disease knowledge extraction model 1, and obtain a road disease knowledge extraction decision table based on expert rules;

[0094] Send road damage entities, relationships, road damage knowledge extraction decision tables, and road damage knowledge sources (papers, inspection reports, etc.) to the self-learning road damage knowledge extraction model 2 for processing to obtain a primary knowledge graph for road damage detection and maintenance.

[0095] Determine the decision effect detection result table based on the primary knowledge map of road disease detection and maintenance and the road disease knowledge extraction decision table;

[0096] Based on the primary knowledge graph of road defect detection and maintenance and the decision-making effect detection result table, a large model self-learning reinforcement knowledge extraction is performed to obtain a road defect graph, namely a knowledge graph.

[0097] Introducing knowledge graphs to enhance semantic understanding. Road deterioration knowledge, such as its causes, development patterns, and treatment standards, is constructed into a knowledge graph. By combining the knowledge graph with test data and a sample learning model, the sample learning model can not only identify deterioration but also analyze its causes, predict its development trends, and provide maintenance recommendations.

[0098] For example, when horizontal cracks are identified in a certain road section, the sample learning model combines knowledge of geology, traffic flow, etc. to determine that the cracks may be caused by heavy-loaded vehicles and foundation settlement, and recommends corresponding repair plans; for example, when road cracks are identified, combined with the geological conditions and historical maintenance records of the road section in the knowledge graph, it is determined that they may be caused by the aging of the base material and heavy traffic, and recommends corresponding repair processes and materials.

[0099] S106: A road condition prediction model built using a long short-term memory network (LSTM) and a graph neural network (GNN) is used to dynamically predict the causes of road damage and obtain prediction results.

[0100] In S106, through the LSTM network and GNN network in the road condition prediction model, combined with the analysis of the temporal characteristics and spatial correlation of the data corresponding to the disease analysis cause, the temporal characteristics, spatial correlation, historical detection data, meteorological data and traffic flow data are integrated to perform the road disease development trend and carrying capacity changes in the preset future time, so as to complete the long-term dynamic prediction process of the disease analysis cause and obtain the prediction results.

[0101] The preset future time may be the next 3 months, the next 6 months, etc. This application does not make any specific limitation on the preset future time.

[0102] The road condition prediction model regularly receives historical inspection data, meteorological data, and traffic flow data. The LSTM network and GNN network are combined to analyze the temporal characteristics and spatial correlations of the data to predict the development trend of road diseases and changes in carrying capacity in the next 6 months.

[0103] A large road condition prediction model built based on LSTM networks and GNN graph neural networks integrates historical detection data, meteorological data, and traffic flow data to conduct long-term dynamic predictions of road disease development and carrying capacity changes, improving prediction accuracy by more than 30%.

[0104] S107: Automatically generate an optimal maintenance strategy based on the prediction results and maintenance resource constraints.

[0105] In S107, a reinforcement learning decision model is established to automatically generate an optimal maintenance strategy based on the model, prediction results, and maintenance resource constraints. Specifically, the reinforcement learning decision model automatically generates the optimal maintenance strategy based on the prediction results, combined with maintenance resource constraints such as maintenance funds, personnel, and equipment. For example, if a road section is predicted to have large-scale potholes during the rainy season, the reinforcement learning decision model can prioritize preventive seal construction and plan construction schedules and resource allocation.

[0106] The reinforcement learning decision-making model dynamically adjusts the optimal maintenance strategy by continuously interacting with the maintenance practice environment. For example, when maintenance funds are limited, it prioritizes the maintenance of high-risk sections to maximize resource benefits.

[0107] In the process of implementing the optimal maintenance strategy, the value of data assets is updated in real time through a multi-dimensional dynamic evaluation model, and blockchain technology is combined to realize data asset ownership and traceability to complete the implementation of dynamic value evaluation of data assets.

[0108] A multi-dimensional dynamic evaluation model was designed, taking into account factors such as data scarcity (e.g., data from road sections with unique geological conditions), timeliness (real-time monitoring data is more valuable than historical data), and business relevance (its support for key maintenance decisions). This model was then used to update the value of data assets in real time. This model, combined with blockchain smart contracts, enabled data traceability and secure transactions.

[0109] A multi-dimensional dynamic assessment model based on the entropy weight method regularly calculates the value of data assets. It calculates a real-time value score based on factors such as data update frequency, usage frequency, contribution to business decisions, as well as data scarcity, timeliness, and business relevance. The data asset value score is then adjusted in real time based on factors such as data update frequency, usage frequency, and contribution to business decisions. For example, real-time monitoring data on a specific geological section of a road would receive a higher data asset value score due to its scarcity and strong business relevance.

[0110] Using blockchain technology to confirm the ownership of data assets, record the entire process of data collection, processing, and use, and ensure the traceability and security of data assets during transactions and sharing. By combining blockchain technology to achieve data asset ownership confirmation and traceability, we can ensure the security and credibility of data transactions and sharing, and promote the market circulation of road inspection data assets.

[0111] Experiments have shown that this application is significantly superior to traditional methods in terms of road defect identification accuracy (F1 value 92.5%), maintenance costs (reduced by 18%), and data asset utilization (increased by 35%).

[0112] In order to facilitate the understanding of the data processing process of intelligent road detection, combined with Figure 4 To illustrate, Figure 4 The architecture diagram of the intelligent road detection data asset management system based on the big model is shown.

[0113] Figure 4 In the paper, the intelligent road detection data asset management system based on the big model includes data acquisition layer, edge processing layer, data center layer and application layer.

[0114] Among them, the data acquisition layer includes high-definition infrared cameras, high-precision lidar, ground-penetrating radar arrays, etc.

[0115] The edge processing layer includes multiple Transformer edge computing units.

[0116] Through the data center layer, deep fusion of multi-source data, semantic understanding of minor diseases, dynamic prediction and adaptive decision-making, and dynamic value assessment of data assets are performed.

[0117] The application layer includes road maintenance departments.

[0118] Advantages of this application:

[0119] Intelligent data collection: Multimodal adaptive collection and edge pre-processing technology improves data collection quality by 40%, reduces invalid data transmission by 30%, and lowers data storage costs;

[0120] Precise disease identification: The accuracy of identifying minor diseases has been increased to over 95%. Through semantic understanding, more targeted maintenance plans are provided, and preventive maintenance efficiency is increased by 50%.

[0121] Scientific decision support: Dynamic prediction and adaptive decision-making systems reduce road maintenance costs by 25%, extend road service life by 15%, and effectively improve the management level of road assets.

[0122] Data asset valuation: Dynamic value assessment and blockchain rights confirmation technology promote the market-oriented operation of road inspection data assets and provide an innovative model for industry data sharing and transactions.

[0123] This application aims to break through the limitations of traditional road inspection data management through technological innovation, build an intelligent management system based on large models, achieve efficient, accurate and intelligent processing of road inspection data from collection to application, and improve the level of digital management of road assets.

[0124] The beneficial effects of the method embodiments of the present application are as follows: in the process of road detection, a multi-sensor fusion device with dynamic adjustment is used to collect multi-source data, a real-time environmental perception model based on deep reinforcement learning is used to dynamically optimize the parameters of multimodal sensors, and the data collection quality in severe weather and environment is improved. A spatiotemporal model is used to achieve millimeter-level alignment of multi-source data, and a generative adversarial network is introduced to enhance the multi-source data after spatiotemporal alignment to supplement missing data details and improve data quality. The enhanced multi-source data is intelligently identified for defects through a sample learning model and a meta-learning algorithm. The sample learning model solves the problem of insufficient labeled samples for minor defect data, and the meta-learning algorithm is used to quickly generalize based on a small amount of labeled data to identify minor defects such as early cracks below 0.5 mm and hidden road structure layer defects, thereby improving the accuracy of road defect identification.

[0125] Based on the above embodiment Figure 1 The disclosed data processing method of intelligent road detection, the embodiment of the present application also discloses a data processing system of intelligent road detection, such as Figure 5 As shown, the data processing system of the intelligent road detection includes:

[0126] The acquisition unit 501 is used to collect multi-source data through a multi-sensor fusion device with dynamic adjustment;

[0127] A spatiotemporal alignment unit 502 is configured to perform spatiotemporal alignment on multi-source data at a preset spatiotemporal accuracy;

[0128] An enhancement unit 503 is used to enhance the multi-source data after spatiotemporal alignment by generating an adversarial network to supplement missing data details;

[0129] Intelligent identification unit 504, used to perform intelligent disease identification on the enhanced multi-source data through the sample learning model and meta-learning algorithm to obtain identification results;

[0130] An analysis unit 505 is configured to analyze the disease using a knowledge graph to obtain a cause of the disease if the recognition result indicates that the disease has been identified;

[0131] A dynamic prediction unit 506 is used to dynamically predict the cause of the disease analysis using a road condition prediction model constructed by a long short-term memory network and a graph neural network to obtain a prediction result;

[0132] The automatic generation unit 507 is used to automatically generate an optimal maintenance strategy based on the prediction results and maintenance resource constraints.

[0133] Furthermore, the acquisition unit 501 includes:

[0134] A dynamic adjustment module for dynamically adjusting sensor parameters of a multi-sensor fusion device based on a reinforcement learning model and real-time environmental parameters; wherein the multi-sensor fusion device includes at least an infrared camera meeting a preset clarity, a lidar meeting a preset accuracy, and a ground-penetrating radar array; the sensor parameters include at least camera exposure parameters and radar scanning frequency;

[0135] The acquisition module is used to collect laser point cloud, radar echo and image data through a dynamically adjusted multi-sensor fusion device to complete the acquisition of multi-source data.

[0136] Furthermore, the spatiotemporal alignment unit 502 includes:

[0137] The first extraction module is used to extract the spatiotemporal features of laser point clouds, radar data, and image data from multi-source data using a spatiotemporal Transformer model with millimeter-level spatial accuracy and millisecond-level temporal accuracy;

[0138] A calculation module is used to calculate the matching relationship between the spatiotemporal features of the laser point cloud, the spatiotemporal features of the radar data, and the spatiotemporal features of the image data through an attention mechanism; wherein the matching relationship represents the spatiotemporal mapping relationship between each data in the multi-source data;

[0139] The data fusion module is used to perform data fusion in three-dimensional space based on matching relationships to complete the spatiotemporal alignment of multi-source data.

[0140] Furthermore, the enhancement unit 503 includes:

[0141] The discriminator in the generative adversarial network is used to judge the authenticity of the multi-source data after spatiotemporal alignment, and the generator in the generative adversarial network is used to supplement the missing data details of the multi-source data, so as to complete the process of enhancing the multi-source data after spatiotemporal alignment through the generative adversarial network.

[0142] Furthermore, the analysis unit 505 includes:

[0143] A second extraction module is configured to extract preset disease knowledge from a knowledge graph if the recognition result indicates that a disease has been identified; wherein the knowledge graph is constructed based on road disease knowledge; and the road disease knowledge includes at least the causes of the disease, its development patterns, and treatment standards.

[0144] The reasoning module is used to perform disease analysis and cause reasoning on the disease through preset disease knowledge to obtain the cause of the disease analysis.

[0145] Furthermore, the dynamic prediction unit 506 includes:

[0146] The analysis module is used to analyze the temporal characteristics and spatial correlation of the corresponding data of disease analysis causes by combining the long short-term memory network and graph neural network in the road condition prediction model;

[0147] The dynamic prediction module is used to integrate time series characteristics, spatial correlation, historical detection data, meteorological data and traffic flow data to predict the development trend of road diseases and changes in carrying capacity at preset future times, so as to complete the process of dynamic prediction of the causes of disease analysis and obtain prediction results.

[0148] Furthermore, the data processing system for intelligent road detection also includes:

[0149] The update unit is used to update the value of data assets in real time through a multi-dimensional dynamic evaluation model during the execution of the optimal maintenance strategy, and to realize the ownership confirmation and traceability of data assets in combination with blockchain technology.

[0150] The beneficial effects of the embodiment of this system are as follows: During the road detection process, a multi-sensor fusion device with dynamic adjustment is used to collect multi-source data. A real-time environmental perception model based on deep reinforcement learning is used to dynamically optimize the parameters of multimodal sensors to improve the data collection quality in severe weather and environments. A spatiotemporal model is used to achieve millimeter-level alignment of multi-source data, and a generative adversarial network is introduced to enhance the multi-source data after spatiotemporal alignment to supplement missing data details and improve data quality. The enhanced multi-source data is intelligently identified for defects through a sample learning model and a meta-learning algorithm. The sample learning model solves the problem of insufficient labeled samples for minor defect data. The meta-learning algorithm quickly generalizes based on a small amount of labeled data to identify minor defects such as early cracks below 0.5 mm and hidden road structure layer defects, thereby improving the accuracy of road defect identification.

[0151] An embodiment of the present application further provides a storage medium, which includes stored instructions, wherein when the instructions are executed, the device where the storage medium is located is controlled to execute the data processing method for intelligent road detection as described above.

[0152] The present application also provides an electronic device, the structure of which is shown in FIG. Figure 6 As shown, it specifically includes a memory 601 and one or more instructions 602, wherein the one or more instructions 602 are stored in the memory 601 and are configured to be executed by one or more processors 603 to perform the data processing method for the above-mentioned intelligent road detection.

[0153] For the sake of simplicity, the aforementioned method embodiments are described as a series of action combinations. However, those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application.

[0154] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similarities between the various embodiments can be referred to in conjunction with each other. For system-related embodiments, since they are generally similar to method-related embodiments, their description is relatively simple. For relevant details, refer to the description of the method-related embodiments.

[0155] The steps in the methods of the various embodiments of the present application can be adjusted in sequence, combined, and deleted according to actual needs.

[0156] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0157] The above description of the disclosed embodiments will enable those skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

[0158] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A data processing method for intelligent road detection, characterized in that: The method comprises: Collect multi-source data through multi-sensor fusion equipment with dynamic adjustment; Performing spatiotemporal alignment on the multi-source data at a preset spatiotemporal accuracy; Generate adversarial networks to enhance the spatiotemporal aligned multi-source data to supplement missing data details; Through the sample learning model and meta-learning algorithm, the enhanced multi-source data is used to perform intelligent disease identification and obtain the identification results; If the recognition result indicates that the disease is recognized, the disease is analyzed through the knowledge graph to obtain the cause of the disease analysis; The road condition prediction model constructed by the long short-term memory network and the graph neural network is used to dynamically predict the causes of the disease analysis and obtain the prediction results; An optimal maintenance strategy is automatically generated based on the prediction results and maintenance resource constraints.

2. The method according to claim 1, characterized in that The method of collecting multi-source data through a multi-sensor fusion device with dynamic adjustment includes: Dynamically adjust sensor parameters of a multi-sensor fusion device based on a reinforcement learning model and real-time environmental parameters; wherein the multi-sensor fusion device includes at least an infrared camera with a preset clarity, a lidar with a preset accuracy, and a ground-penetrating radar array; the sensor parameters include at least camera exposure parameters and radar scanning frequency; Laser point cloud, radar echo and image data are collected through dynamically adjusted multi-sensor fusion equipment to complete the collection of multi-source data.

3. The method according to claim 1, characterized in that The performing spatiotemporal alignment on the multi-source data at a preset spatiotemporal accuracy includes: With millimeter-level spatial accuracy and millisecond-level temporal accuracy, the spatiotemporal Transformer model is used to extract the spatiotemporal features of laser point clouds, radar data, and image data from multi-source data. The attention mechanism is used to calculate the matching relationship between the spatiotemporal features of the laser point cloud, the spatiotemporal features of the radar data, and the spatiotemporal features of the image data; wherein the matching relationship represents the spatiotemporal mapping relationship between each data in the multi-source data; In three-dimensional space, data fusion is performed according to the matching relationship to complete the spatiotemporal alignment of the multi-source data.

4. The method according to claim 1, wherein The multi-source data after spatiotemporal alignment is enhanced by generating adversarial networks to supplement missing data details, including: The discriminator in the generative adversarial network is used to judge the authenticity of the multi-source data after spatiotemporal alignment, and the generator in the generative adversarial network is used to supplement the missing data details of the multi-source data, so as to complete the process of enhancing the multi-source data after spatiotemporal alignment through the generative adversarial network.

5. The method according to claim 1, wherein If the recognition result indicates that a disease is recognized, the disease is analyzed through the knowledge graph to obtain the cause of the disease analysis, including: If the recognition result indicates that a disease is identified, extracting preset disease knowledge from a knowledge graph; wherein the knowledge graph is constructed based on road disease knowledge; the road disease knowledge at least includes the cause of the disease, its development pattern, and treatment standards; The disease analysis cause reasoning is performed on the disease using the preset disease knowledge to obtain the disease analysis cause.

6. The method according to claim 1, characterized in that The road condition prediction model constructed by the long short-term memory network and the graph neural network dynamically predicts the cause of the disease analysis and obtains prediction results, including: By combining the long short-term memory network and graph neural network in the road condition prediction model, the temporal characteristics and spatial correlation of the corresponding data of the disease analysis cause are analyzed; By integrating the time series features, the spatial associations, historical detection data, meteorological data and traffic flow data, the development trend of road diseases and changes in carrying capacity at a preset future time are analyzed to complete the process of dynamic prediction of the causes of the disease analysis and obtain prediction results.

7. The method according to claim 1, characterized in that Also includes: In the process of implementing the optimal maintenance strategy, the value of data assets is updated in real time through a multi-dimensional dynamic evaluation model, and blockchain technology is combined to achieve data asset ownership confirmation and traceability.

8. A data processing system for intelligent road detection, characterized in that: The system comprises: An acquisition unit for collecting multi-source data through a multi-sensor fusion device with dynamic adjustment; A spatiotemporal alignment unit, configured to perform spatiotemporal alignment on the multi-source data at a preset spatiotemporal accuracy; The enhancement unit is used to enhance the multi-source data after spatiotemporal alignment through a generative adversarial network to supplement the missing data details; Intelligent recognition unit, used to perform intelligent disease recognition on enhanced multi-source data through sample learning models and meta-learning algorithms to obtain recognition results; an analysis unit configured to analyze the disease through a knowledge graph to obtain a cause of the disease if the recognition result indicates that the disease has been identified; A dynamic prediction unit is used to dynamically predict the cause of the disease analysis by using a road condition prediction model constructed by a long short-term memory network and a graph neural network to obtain a prediction result; The automatic generation unit is used to automatically generate an optimal maintenance strategy based on the prediction results and maintenance resource constraints.

9. A storage medium, characterized in that: The storage medium includes stored instructions, wherein when the instructions are executed, the device where the storage medium is located is controlled to execute the data processing method for intelligent road detection according to any one of claims 1 to 7.

10. An electronic device, characterized in that: The device comprises a memory and one or more instructions, wherein the one or more instructions are stored in the memory and configured to execute the data processing method for intelligent road detection according to any one of claims 1 to 7 by one or more processors.

Images