Road crack detection system and method

Abstract

The invention relates to the technical field of road detection, in particular to a road crack detection system and method.The road crack detection system is mainly composed of an unmanned aerial vehicle module, an unmanned trolley module, a multi-sensor module, a control center module and a data processing and optimizing module; the method comprises the following steps: acquiring road image data, processing, detecting, classifying cracks, estimating the length and width of the cracks, transmitting crack position data to a control center module, generating a preliminary crack distribution diagram by a control center according to the data, sending crack coordinates to an unmanned trolley, indicating the unmanned trolley to accurately reach the crack positions, and marking; the unmanned trolley obtains the three-dimensional geometric data and extension information of the crack, generates a high-precision crack three-dimensional model, predicts the extension trend of the crack, and generates a crack health assessment report, so that the accuracy and reliability of crack identification can be effectively improved, the phenomena of false detection and missing detection are reduced, and the timeliness and effectiveness of road maintenance are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of road detection, and more specifically, it relates to a road crack detection system and method. Background Art

[0002] Cracks are one of the common diseases in infrastructure such as roads, bridges, and tunnels. Cracks not only affect the strength and service life of the structure but also pose a serious threat to traffic safety. Therefore, timely and accurately detecting cracks and evaluating their severity and development trend are important research directions in the field of structural health monitoring.

[0003] For example, a method for calculating the damage degree of a road surface disclosed in the patent publication number CN114298972B installs an image acquisition device on a road surface detection vehicle to obtain road surface images, thereby obtaining the road surface disease situation, and calculating the pavement condition index PCI in the background, eliminating

[0004] the trouble of manual on-site measurement and calculation, reducing labor costs, and improving calculation efficiency.

[0005] This method has good application value, but in actual use, the image quality is greatly affected by the environment, resulting in a decrease in the accuracy of crack detection. Relying solely on two-dimensional images, it is impossible to accurately measure the depth and internal expansion of cracks, especially there is a certain difficulty in identifying deep cracks. Although this method improves the efficiency of crack detection, in the application of large-scale or complex areas, the real-time performance of image processing is still limited.

[0006] There is also a road crack detection device and method based on drone inspection disclosed in the patent publication number CN110046584B. When using a drone for inspection, it collects pictures through a picture collection device, and then transmits them to a ground wireless image receiving platform and a backend central station in sequence for image processing, so as to automatically detect and extract the crack contour and can be used to analyze the road surface health condition. It is convenient to operate, has high detection efficiency, saves a lot of manpower and material resources, and at the same time has accurate analysis, and has high practicality and promotion value.

[0007] Although drones can cover large areas, the fine marking and depth measurement of cracks depend on image quality and processing algorithms. In a complex environment, the positioning accuracy of cracks is affected,

[0008] it is difficult to comprehensively evaluate the type and development trend of cracks, especially for cracks with complex shapes and deep cracks, accurate classification and evaluation are still difficult.

[0009] Based on the above-mentioned many existing problems, there is an urgent need for new road crack detection systems and methods in the current market. Summary of the Invention

[0010] For this reason, the purpose of the present invention is to provide a road crack detection system and method, which integrates high-precision detection, advanced data fusion, optimized path planning, and accurate trend prediction to improve the efficiency and accuracy of road crack detection.

[0011] To achieve the above object, the present invention provides the following technical solutions:

[0012] A road crack detection system, comprising:

[0013] A drone module, which is used to conduct large-scale inspections on roads, collect road image data, and perform preliminary crack detection and feature extraction through a frequency domain analysis algorithm based on Fourier transform;

[0014] An unmanned vehicle module, which is used to move to a specified position according to the crack coordinate information sent by the drone, and mark and detect the cracks;

[0015] A multi-sensor module, including a lidar, an infrared thermal imager, and a piezoelectric sensor, which is used to obtain three-dimensional information, depth data of the cracks, and dynamic response data of crack propagation;

[0016] A control center module, which is used to receive data from the drone and the unmanned vehicle, generate a three-dimensional crack model, and provide maintenance suggestions;

[0017] A data processing and optimization module, which is used to process crack detection data in real time, and perform path planning and task scheduling through an adaptive hybrid quantum genetic algorithm.

[0018] The present invention is further configured as: the control center module includes:

[0019] A deep learning algorithm module, which uses a convolutional neural network and a graph neural network to process the images and depth data collected by the drone, and extract the position, shape, and depth features of the cracks;

[0020] A dynamic expansion prediction module, which combines a diffusion equation model and a variational autoencoder to predict the future development trend of the cracks.

[0021] The present invention is further configured as: the data processing and optimization module adopts an adaptive hybrid quantum genetic algorithm, and combines deep reinforcement learning to adjust task allocation based on real-time feedback, optimize the path planning of the drone and the unmanned vehicle, and avoid path overlap.

[0022] The present invention is further configured as: the unmanned vehicle module is provided with a paint spraying unit for spraying fluorescent paint on the crack position.

[0023] A road crack detection method, comprising the following steps:

[0024] Step S1: Use a drone to conduct a large - scale inspection of the target area and collect road image data;

[0025] Step S2: Use a frequency - domain analysis algorithm based on Fourier transform to process the collected image data, detect and classify cracks, and estimate the length and width of the cracks;

[0026] Step S3: The drone transmits the crack location data to the control center module, and the control center generates a preliminary crack distribution map based on the data;

[0027] Step S4: The control center sends the crack coordinates to the unmanned vehicle, instructing the unmanned vehicle to accurately reach the crack location and make a mark;

[0028] Step S5: The unmanned vehicle measures the depth of the crack through a lidar, an infrared thermal imager, and a piezoelectric sensor to obtain the three - dimensional geometric data and expansion information of the crack;

[0029] Step S6: Adopt a fusion algorithm that combines Bayesian inference and graph neural network to fuse multi - sensor data and generate a high - precision three - dimensional crack model;

[0030] Step S7: Use an adaptive hybrid quantum genetic algorithm combined with deep reinforcement learning to dynamically optimize the paths and tasks of the drone and the unmanned vehicle;

[0031] Step S8: Based on a prediction model that combines a diffusion equation model and a variational auto - encoder, predict the expansion trend of the crack;

[0032] Step S9: Generate a crack health assessment report, including the risk score of the crack, expansion prediction, and repair suggestions.

[0033] The present invention is further configured as: in the step S1, the drone collects images through a high - definition camera, uses a frequency - domain analysis algorithm based on Fourier transform to identify the location of the crack, and uses an RTK positioning system to record the geographical coordinates of the crack.

[0034] The present invention is further configured as: in the step S6, the multi - sensor data fusion includes the fusion of Bayesian inference and graph neural network. Bayesian inference is used to integrate the uncertainty information of multi - source data, and the graph neural network is used to capture the complex associations between data to improve the accuracy and robustness of data fusion.

[0035] The present invention is further configured as: in the step S7, the adaptive hybrid quantum genetic algorithm combined with deep reinforcement learning specifically includes the implementation of the following mathematical models:

[0036] The qubit represents path planning. The path nodes of the drone and the unmanned vehicle are represented as qubits, specifically:

[0037]

[0038] Among them, and are complex amplitudes that satisfy the normalization condition: ;

[0039] Genetic operations are performed by quantum gate operations. The path nodes are crossed and mutated by applying quantum gates. The specific operations include:

[0040] The role of the Hadamard gate H:

[0041] The role of the Pauli-X gate X: ;

[0042] Fitness function is defined as the weighted sum of the total path length and energy consumption:

[0043]

[0044] represents the distance between adjacent nodes in the path, represents the node 's energy consumption,

[0045] and are weight coefficients that respectively reflect the influence degrees of the path length and energy consumption on the fitness;

[0046] Adaptive mechanism, dynamically adjust the crossover rate and the mutation rate to meet the requirements of different optimization stages, which is defined as:

[0047]

[0048] and are the initial crossover rate and mutation rate respectively, is the adjustment rate, is the current iteration number;

[0049] Deep reinforcement learning, adopt the state-action value function of the deep Q network and define the reward function as:

[0050]

[0051] The weight coefficients of the task completion degree, energy consumption and time in the reward function,

[0052] is the current state, including the positions of the drone and the unmanned vehicle, the remaining tasks, and the energy state.

[0053] is the action taken, including the selection of the moving directions and speeds of the drone and the unmanned vehicle.

[0054] represents the network parameters of the deep Q-network.

[0055] The state-action value function ) is defined as:

[0056]

[0057] where is the discount factor of the is the reward obtained at time + is the current time step, is the current time and the th time step from now,

[0058] Parameter update rule: Use the Bellman equation and the gradient descent method to minimize the loss function:

[0059]

[0060] The objective function, defined as minimizing the task completion time and energy consumption, is specifically:

[0061]

[0062] and are the task completion times of the drone and the unmanned vehicle respectively, is the total energy consumption of the system,

[0063] respectively reflect the influence degrees of the task completion time and energy consumption on the objective function.

[0064] The present invention is further configured as: The crack propagation trend prediction model includes:

[0065] The diffusion equation model, whose mathematical expression is:

[0066]

[0067] where represents the crack propagation degree, is the diffusion coefficient, ​is the source term, representing the external influencing factors for crack propagation; is the partial derivative operator with respect to time ; is the second-order partial derivative operator with respect to the spatial coordinates ;

[0068] The variational autoencoder model is used to generate the latent representation of crack propagation, and its mathematical expression is:

[0069]

[0070] where is the crack propagation data, is the latent variable, is the encoder distribution, is the decoder distribution, is the Kullback-Leibler divergence.

[0071] The present invention is further configured such that: the crack health assessment report includes the risk score of the crack and predicts the crack propagation trend, and the scoring model is:

[0072]

[0073] is the length of the crack, is the width of the crack, is the depth of the crack, is the volume of crack propagation, is the maximum stress at the crack tip, is the strain value around the crack, is the reference weight coefficient, is the dynamically adjusted weight coefficient.

[0074] Comparing with the deficiencies of the prior art, the beneficial effects of the present invention are:

[0075] Adopting the frequency domain analysis algorithm based on Fourier transform combined with the three-dimensional convolutional neural network (3D-CNN) to achieve high-precision detection and depth estimation of road cracks. This method can effectively improve the accuracy and reliability of crack identification, reduce false detection and missed detection phenomena, and ensure the timeliness and effectiveness of road maintenance.

[0076] In particular, through the Bayesian inference and graph neural network (GNN) fusion algorithm, it is possible to integrate high-precision three-dimensional crack models generated from lidar, infrared thermal imagers, and piezoelectric sensors. This technology improves the accuracy and robustness of data fusion and ensures the comprehensiveness and accuracy of crack detection results.

[0077] In particular, by combining the diffusion equation model with the variational autoencoder (VAE), the future propagation trend of cracks can be accurately predicted. This predictive ability provides a scientific basis for road maintenance decisions, extends the service life of roads, and reduces maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 It is a flowchart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0079] Referring to Figure 1 , the present invention provides a road crack detection system and method integrating drones and unmanned vehicles, which combines advanced algorithm models and mathematical functions to improve the accuracy, real-time performance, and predictive ability of crack detection. The following will detail each component of the present invention and its collaborative working mechanism:

[0080] The road crack detection system mainly consists of the following modules: drone module, unmanned vehicle module, multi-sensor module, control center module, data processing and optimization module, and user interface module.

[0081] The road crack detection method includes the following steps:

[0082] Step S1: Use a drone to conduct a large-scale inspection of the target area and collect road image data;

[0083] Step S2: Process the collected image data using a frequency domain analysis algorithm based on Fourier transform to detect and classify cracks and estimate the length and width of the cracks;

[0084] Step S3: The drone transmits the crack position data to the control center module, and the control center generates a preliminary crack distribution map based on the data;

[0085] Step S4: The control center sends the crack coordinates to the unmanned vehicle, instructing the unmanned vehicle to accurately reach the crack position and mark it;

[0086] Step S5: The unmanned vehicle measures the depth of the crack through a lidar, an infrared thermal imager, and a piezoelectric sensor to obtain the three-dimensional geometric data and propagation information of the crack;

[0087] Step S6: Use a fusion algorithm that combines Bayesian inference and graph neural networks to fuse the multi-sensor data and generate a high-precision three-dimensional model of the crack;

[0088] Step S7: Use an adaptive hybrid quantum genetic algorithm combined with deep reinforcement learning to dynamically optimize the paths and tasks of the drone and the unmanned vehicle;

[0089] Step S8: Based on a prediction model that combines the diffusion equation model and the variational autoencoder, predict the propagation trend of the crack;

[0090] Step S9: Generate a crack health assessment report, including the risk score of the crack, expansion prediction, and repair suggestions.

[0091] The specific implementation of the above 9 steps is as follows:

[0092] The drone module is responsible for conducting large-scale and high-efficiency inspections of the target road area, collecting high-resolution road image data and depth information. This module is equipped with a high-definition camera and a stereo vision sensor, capable of capturing crack information on the road surface in real time.

[0093] Image acquisition: The drone is equipped with a high-definition camera and flies at a fixed height and speed to collect continuous images of the road surface.

[0094] Depth information acquisition: Through stereo vision technology or structured light devices, obtain the depth information of the crack to provide data support for subsequent 3D reconstruction.

[0095] Real-time positioning: Equipped with an RTK (Real-Time Kinematic) system to accurately record the position coordinates of each crack, ensuring the spatial accuracy of the data.

[0096] The unmanned vehicle module moves to the specified location according to the crack coordinate information sent by the drone, marks and finely detects the crack. This module has a self-repair function and can automatically release a repair agent for preliminary repair when detecting high-risk cracks.

[0097] Precise positioning and navigation: Use GNSS and inertial navigation systems to ensure that the unmanned vehicle accurately reaches the specified crack location.

[0098] Marking and detection: Equipped with high-contrast, long-lasting fluorescent paint to clearly identify the crack location, and equipped with a lidar, infrared thermal imager, piezoelectric sensor, and structured light device to conduct 3D scanning and collect dynamic response data.

[0099] The multi-sensor module integrates multiple sensors to obtain 3D information of the crack, depth data, and dynamic response data of crack expansion. Through the fusion analysis of multi-source data, a high-precision 3D model of the crack is generated.

[0100] Lidar: Used to accurately measure the geometry and spatial distribution of the crack.

[0101] Infrared thermal imager: Detect the temperature change around the crack and analyze the thermodynamic characteristics of crack expansion.

[0102] Piezoelectric sensor: Monitor the dynamic response of the crack and capture the vibration and stress changes during crack expansion.

[0103] Structured light device: Obtain the surface structure information of cracks to assist in 3D reconstruction.

[0104] The control center module is responsible for receiving data from the drone and the unmanned vehicle, generating a 3D crack model and providing maintenance suggestions through advanced data fusion algorithms. This module is also responsible for optimizing path planning and task scheduling to achieve the efficient operation of the system.

[0105] Data reception and storage: Receive images, depth information, and sensor data from the drone and the unmanned vehicle in real time through a wireless communication protocol and store them in a high-performance database.

[0106] Data fusion algorithm: Adopt a fusion algorithm of Bayesian inference and graph neural network (GNN) to comprehensively analyze multi-source data and generate a high-precision 3D crack model.

[0107] Dynamic task scheduling: Combine the adaptive hybrid quantum genetic algorithm (AHQGA) with deep reinforcement learning (DRL) to optimize the working paths and task assignments of the drone and the unmanned vehicle, ensuring the efficient collaborative operation of the system.

[0108] The data processing and optimization module is the core part of the system, responsible for real-time processing of crack detection data and optimizing path planning and task scheduling through advanced algorithm models to minimize task completion time and energy consumption.

[0109] Adaptive hybrid quantum genetic algorithm (AHQGA):

[0110] The qubit represents path planning, and the path nodes of the drone and the unmanned vehicle are represented as qubits, specifically:

[0111]

[0112] Among them, and are complex amplitudes that satisfy the normalization condition: .

[0113] Quantum gate operation: Perform crossover and mutation operations on path nodes by applying quantum gates (such as Hadamard gate HHH and Pauli-X gate XXX) to enhance the diversity and global search ability of genetic operations.

[0114] The role of Hadamard gate H:

[0115] The role of Pauli-X gate X: ;

[0116] Fitness function Is defined as the weighted sum of the total path length and energy consumption:

[0117]

[0118] represents the distance between adjacent nodes in the path, Representation Node Energy consumption,

[0119] and are weight coefficients, which respectively reflect the influence of path length and energy consumption on fitness;

[0120] Adaptive mechanism to dynamically adjust the crossover rate and mutation rate , to meet the needs of different optimization stages, is defined as:

[0121]

[0122] and are the initial crossover rate and mutation rate, respectively. To adjust the rate, is the current iteration number;

[0123] Deep reinforcement learning, using deep Q-network state-action-value functions ,

[0124]

[0125] is the current state of the system at time t, including the positions, remaining tasks, and energy states of the drone and the unmanned vehicle,

[0126] is the action taken at time t, including the moving direction and speed selection of the drone and the unmanned car,

[0127] Represents all trainable parameters of a deep Q-network (including network weights and biases).

[0128] in, Discount factor of , controlling the degree of decay of future rewards; For in time + Instant rewards, is the current time step, For current time Next time steps,

[0129] The reward function is defined as a weighted combination of task completion, energy consumption, and time:

[0130]

[0131] are the weight coefficients of task completion, energy consumption, and time in the reward function.

[0132] Policy selection and parameter update:

[0133] The ε-greedy policy is adopted to balance exploration and exploitation:

[0134]

[0135] The parameters are updated using the Bellman equation and gradient descent method to minimize the loss function :

[0136]

[0137] are the target network parameters, which are updated regularly from to stabilize the training process.

[0138] Comprehensive optimization process:

[0139] Initializing the population: Use AHQGA to randomly generate the qubit representation paths of the initial population.

[0140] Fitness evaluation: Calculate the fitness f(X) of each individual.

[0141] Quantum gate operations and genetic operations: Apply Hadamard gates and Pauli-X gates to perform crossover and mutation of paths to generate new individuals.

[0142] Selection and evolution: Select excellent individuals according to the fitness values to enter the next generation.

[0143] Adaptive adjustment: Dynamically adjust the crossover rate C(t) and mutation rate M(t).

[0144] Deep reinforcement learning optimization: Use DQN to locally optimize the selected paths and adjust the path details to adapt to real-time environmental changes.

[0145] Iteration termination: When the preset number of iterations is reached or the fitness converges, output the optimal path plan.

[0146] By combining the global search ability of AHQGA with the real-time policy optimization of DRL, the efficient collaborative operation of unmanned aerial vehicles and unmanned vehicles can be achieved, significantly minimizing the task completion time and energy consumption, and improving the overall operation efficiency and response speed of the system.

[0147] The crack propagation trend prediction model mainly consists of the following two parts: the diffusion equation model and the variational autoencoder.

[0148] Diffusion equation model:

[0149]

[0150] represents the degree of crack propagation at position and time ; is the diffusion coefficient, reflecting the diffusion rate of crack propagation; is the source term, representing the external influencing factors of crack propagation, such as external loads, environmental conditions, etc.; is the spatial coordinate, describing the distribution of cracks in one-dimensional space; is the time variable, describing the change process of crack propagation over time.

[0151] The diffusion equation model is used to describe the propagation behavior of cracks along the spatial coordinate inside or on the surface of the material. The time derivative term on the left represents the rate of change of the crack propagation degree over time; the diffusion term on the right describes the spatial diffusion characteristics of crack propagation, and the source term considers the influence of external factors on crack propagation.

[0152] Variational autoencoder:

[0153]

[0154] is the crack propagation data, is the latent variable, is the encoder distribution: representing the distribution of the latent variable given the crack propagation data ; is the decoder distribution: representing the generative distribution of the crack propagation data given the latent variable ; The prior distribution: usually set to the standard normal distribution N(0, I), is the Kullback-Leibler divergence: used to measure the difference between two probability distributions.

[0155] The VAE maps the crack propagation data to the latent space through the encoder, and then reconstructs the crack propagation data from the latent space . By maximizing the lower bound (ELBO), the VAE learns the distribution of the latent space such that the reconstructed data is as close as possible to the original data while the distribution of the latent space is close to the prior distribution. .

[0156] To improve the accuracy and robustness of crack growth trend prediction, a diffusion equation model is combined with VAE to form a hybrid prediction model. The specific implementation steps are as follows:

[0157] Data collection: High-resolution images, depth information, and multi-sensor data of road cracks are collected through the UAV and unmanned vehicle modules, and data such as crack length, width, depth, expansion volume, stress-strain data, etc. are recorded.

[0158] Data preprocessing: Preprocessing steps such as denoising and enhancement are performed on the image data;

[0159] The multi-sensor data is standardized to unify the data format and scale;

[0160] Geometric features and physical properties of the crack are extracted to prepare data for subsequent model input.

[0161] Crack growth dynamic simulation: Based on historical crack growth data, the diffusion equation model is used to simulate the crack growth behavior at different time steps.

[0162] The diffusion equation is solved by numerical methods (such as the finite difference method or the finite element method) to predict the crack growth extent at future time steps. .

[0163] Source term Modeling: Considering external influencing factors such as traffic load, climate change, material fatigue, etc., the source term is modeled, and historical data and environmental monitoring data are used to dynamically adjust the source term to make the model more in line with the actual situation.

[0164] Encoder design: A deep neural network is constructed as the encoder to map the crack growth data to the latent space .

[0165] The encoder outputs the mean and variance of the latent variable for sampling the latent variable .

[0166] A deep neural network is constructed as the decoder to generate reconstructed crack growth data from the latent variable .

[0167] Loss function optimization:

[0168] ​Maximize the Evidence Lower Bound (ELBO) to train the VAE:

[0169]

[0170] Use backpropagation and gradient descent to optimize the network parameters, enabling the VAE to accurately reconstruct crack propagation data while the latent space distribution approaches the standard normal distribution.

[0171] Construction of the hybrid model: Use the prediction results of the diffusion equation model as the input of the VAE, and use the VAE to extract the latent variables of crack propagation , where the latent variables capture the high-dimensional features and latent patterns of crack propagation, providing rich information for trend prediction. The prediction results are fed back to the control center module, and combined with other detection data to generate a comprehensive health assessment report.

[0172] To more precisely reflect the impact of crack characteristics on risk, a multi-dimensional, non-linear and interactive risk scoring model is proposed:

[0173]

[0174] is the length of the crack, is the width of the crack, is the depth of the crack, is the volume of crack propagation, is the maximum stress at the crack tip, is the strain value around the crack, is the reference weight coefficient, is the dynamically adjusted weight coefficient.

[0175] Model analysis:

[0176]

[0177] Numerator: Combine the power function and the exponential function to non-linearly enhance the impact of crack length and width on the risk score.

[0178] Denominator: Adjust the negative impact of crack depth on the risk score through the exponential decay function to prevent the score from being too low when the depth increases.

[0179] Sine and cosine interaction term:

[0180] Sine function: Capture the maximum stress at the crack tip The periodic change reflects the non-linear effect of stress on the risk score.

[0181] Cosine function: reflecting the strain values around the crack The periodic change reflects the dynamic effect of strain on the risk score.

[0182] Logarithmic transformation and interaction term:

[0183] Logarithmic function: dealing with the crack growth volume The interaction term with stress and strain controls the growth rate of the score in the case of large volume and high stress and strain, ensuring the rationality of the score.

[0184] Interaction term: combining with reflects the comprehensive effect of multi-factor interaction on the risk score.

[0185] The weight coefficient for dynamic adjustment can be dynamically adjusted through machine learning algorithms to adapt to different road conditions and material properties, improving the adaptability and accuracy of the model.

[0186] The final data result content is displayed in the user interface module, which is a display screen for data presentation.

[0187] Example 1, crack detection on a certain highway:

[0188] Crack data table 1 collected by drones:

[0189]

[0190] Three-dimensional crack data table 2 collected by unmanned vehicles:

[0191]

[0192] Table 3 of the optimization results of path planning and task scheduling:

[0193]

[0194] Table 4 of the prediction results of crack propagation trend:

[0195]

[0196] Table description:

[0197] Table 1, crack data collected by drones: This table records the basic information of the cracks detected by drones during the highway inspection, including crack number, position coordinates, length, width, depth, and collection time.

[0198] Table 2, 3D crack data collected by the unmanned vehicle: This table shows the 3D data collected by the unmanned vehicle during the fine detection at the specified crack positions, including crack number, measurement point number, spatial coordinates, stress and strain values, and measurement time.

[0199] Table 3, Optimization results of path planning and task scheduling: This table records the iterative optimization results of the adaptive hybrid quantum genetic algorithm and deep reinforcement learning in the process of path planning and task scheduling, including the crossover rate, mutation rate, optimal path length, total energy consumption, and task completion time for each iteration.

[0200] Table 4, Prediction results of crack propagation trend: This table shows the prediction results of crack propagation trend based on the diffusion equation model and VAE, including crack number, current length, predicted lengths for the next 1 day, 3 days, and 7 days, and the corresponding risk scores.

[0201] Example 2, Crack detection on an urban road

[0202] Table 5, Crack data of urban road collected by the drone

[0203]

[0204] Table 6, 3D crack data of urban road collected by the unmanned vehicle

[0205]

[0206] Table 7, Optimization results of path planning and task scheduling for urban road

[0207]

[0208] Table 8, Prediction results of crack propagation trend

[0209]

[0210] Table description:

[0211] Table 5, Crack data of urban road collected by the drone: This table records the basic information of the cracks detected by the drone during the inspection of the urban road, including crack number, position coordinates, length, width, depth, and collection time.

[0212] Table 6, 3D crack data of urban road collected by the unmanned vehicle: This table shows the 3D data collected by the unmanned vehicle during the fine detection at the specified urban road crack positions, including crack number, measurement point number, spatial coordinates, stress and strain values, and measurement time.

[0213] Table 7, Optimization Results of Urban Road Path Planning and Task Scheduling: This table records the iterative optimization results of the adaptive hybrid quantum genetic algorithm and deep reinforcement learning in the process of urban road crack detection and maintenance, including the crossover rate, mutation rate, optimal path length, total energy consumption, and task completion time for each iteration.

[0214] Table 8, Prediction Results of Crack Propagation Trend: This table shows the prediction results of crack propagation trend based on the diffusion equation model and VAE, including crack number, current length, predicted lengths for the next 1 day, 3 days, and 7 days, and corresponding risk scores.

[0215] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any ordinary changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included in the protection scope of the present invention.

Claims

1. A road crack detection system, characterized in that: include: The drone module is used to conduct large-scale road inspections, collect road image data, and perform preliminary crack detection and feature extraction using a frequency domain analysis algorithm based on Fourier transform; The unmanned vehicle module is used to move to the designated location according to the crack coordinate information sent by the drone, and mark and detect the cracks; Multi-sensor module, including lidar, infrared thermal imager, and piezoelectric sensor, used to obtain three-dimensional information, depth data, and dynamic response data of crack extension; The control center module is used to receive data from drones and unmanned vehicles, generate three-dimensional crack models and provide maintenance recommendations; The data processing and optimization module is used to process crack detection data in real time and perform path planning and task scheduling through an adaptive hybrid quantum genetic algorithm.

2. A road crack detection system according to claim 1, characterized in that: The control center module includes: The deep learning algorithm module uses convolutional neural networks and graph neural networks to process the images and depth data collected by drones to extract the location, shape, and depth features of cracks; The dynamic expansion prediction module combines the diffusion equation model with the variational autoencoder to predict the future development trend of cracks.

3. A road crack detection system according to claim 1, characterized in that: The data processing and optimization module adopts an adaptive hybrid quantum genetic algorithm and combines it with deep reinforcement learning to adjust task allocation based on real-time feedback, optimize the path planning of drones and unmanned vehicles, and avoid path overlap.

4. A road crack detection system according to claim 1, characterized in that: The unmanned vehicle module is provided with a paint spraying unit for spraying fluorescent paint on the crack position.

5. A road crack detection method, characterized in that: The following steps are involved: Step S1: Use a drone to conduct a large-scale inspection of the target area and collect road image data; S2: Use the frequency domain analysis algorithm based on Fourier transform to process the collected image data, detect and classify cracks, and estimate the length and width of cracks; Step S3: The UAV transmits the crack location data to the control center module, and the control center generates a preliminary crack distribution map based on the data; Step S4: The control center sends the crack coordinates to the unmanned vehicle, instructing the unmanned vehicle to accurately reach the crack location and mark it; Step S5: the unmanned vehicle measures the depth of the crack by using a laser radar, an infrared thermal imager, and a piezoelectric sensor to obtain three-dimensional geometric data and extension information of the crack; Step S6: using Bayesian reasoning and graph neural network fusion algorithm to fuse multi-sensor data and generate a high-precision three-dimensional crack model; Step S7: Use an adaptive hybrid quantum genetic algorithm combined with deep reinforcement learning to dynamically optimize the paths and tasks of the drone and the unmanned vehicle; Step S8: predicting the crack expansion trend based on a prediction model combining a diffusion equation model and a variational autoencoder; Step S9: Generate a crack health assessment report, including crack risk score, crack extension prediction and repair suggestions.

6. A road crack detection method according to claim 5, characterized in that: In step S1, the UAV collects images through a high-definition camera, uses a frequency domain analysis algorithm based on Fourier transform to identify the location of the crack, and uses an RTK positioning system to record the geographic coordinates of the crack.

7. A road crack detection method according to claim 6, characterized in that: In step S6, the multi-sensor data fusion includes the fusion of Bayesian reasoning and graph neural network. Bayesian reasoning is used to integrate the uncertainty information of multi-source data, and the graph neural network is used to capture the complex correlation between data to improve the accuracy and robustness of data fusion.

8. A road crack detection method according to claim 7, characterized in that: In step S7, the adaptive hybrid quantum genetic algorithm combined with deep reinforcement learning specifically includes the implementation of the following mathematical model: The qubit represents the path planning, and the path nodes of the drone and the unmanned car are represented as qubits, specifically: ； in, and is the complex amplitude, satisfying the normalization condition: ; Quantum gate operations are used to perform genetic operations. By applying quantum gates, path nodes are crossed and mutated. The specific operations include: The role of Hadamard gate H: ; Pauli-XT door X function: ; Fitness function It is defined as the weighted sum of the total path length and energy consumption: ； represents the distance between adjacent nodes in the path, Representation Node Energy consumption, and are weight coefficients, which respectively reflect the influence of path length and energy consumption on fitness; Adaptive mechanism to dynamically adjust the crossover rate and mutation rate , to meet the needs of different optimization stages, is defined as: ； and are the initial crossover rate and mutation rate, respectively. To adjust the rate, is the current iteration number; Deep reinforcement learning, using deep Q-network state-action-value functions And define the reward function as: ； 2 is the weight coefficient of task completion, energy consumption and time in the reward function, is the current state, including the location, remaining tasks and energy status of the drone and unmanned vehicle. The actions to be taken, including the direction and speed of movement of drones and unmanned vehicles, represents the network parameters of the deep Q network; State-Action Value Function ) is defined as: ； in, Discount factor of square, For in time + Rewards received, is the current time step, For current time Next time steps, Parameter update rule: Use the Bellman equation and gradient descent to minimize the loss function: ； The objective function is defined as minimizing the task completion time and energy consumption, specifically: ； and are the task completion times of the UAV and the unmanned vehicle, is the total energy consumption of the system, Respectively reflect the impact of task completion time and energy consumption on the objective function.

9. A road crack detection method according to claim 8, characterized in that: The crack growth trend prediction model includes: Diffusion equation model, its mathematical expression is: ； in, Indicates the extent of crack expansion. is the diffusion coefficient, is the source term, which represents the external factors affecting crack propagation; For time The partial derivative operator, For space coordinates The second-order partial derivative operator of ; The variational autoencoder model is used to generate the potential representation of crack extension, and its mathematical expression is: ； in, is the crack growth data, is a latent variable, is the encoder distribution, is the decoder distribution, is the Kullback-Leibler divergence.

10. A road crack detection method according to claim 9, characterized in that: The crack health assessment report includes a risk score for the crack and predicts the crack expansion trend. The scoring model is: ； is the length of the crack, is the crack width, is the crack depth, is the crack extension volume, is the maximum stress at the crack tip, is the strain value around the crack, is the base weight coefficient, is the dynamically adjusted weight coefficient.

Images