An intelligent system capable of quickly and easily detecting road conditions

Abstract

The application relates to the field of road detection and image processing, and discloses an intelligent system capable of quickly and simply detecting highway conditions, which comprises a data acquisition module, a data preprocessing module, an edge detection module, a multi-scale detection module, an optimization decision module and a result output module; highway pavement image data are collected through a high-resolution camera, and the images are preprocessed; edge information of cracks is extracted based on a variational energy functional theory and a partial differential equation model; multi-scale feature extraction is carried out on the crack edges by using a high-order continuous wavelet transform, and the specific position, size and type of the cracks are analyzed; the detection result is subjected to precision optimization through an optimization objective function, a detection report containing crack position, classification information and repair suggestions is finally generated, and visual display is provided. The application can realize high-precision, real-time and robust highway crack detection under complex pavement and illumination conditions, and provides a scientific basis for highway maintenance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of road detection and image processing, in particular to an intelligent system capable of quickly and simply detecting highway conditions. BACKGROUND

[0002] With the development of highway transportation, the construction scale and the driving distance of expressways are increasing, and the detection and maintenance of highways are becoming increasingly important. The condition of the road surface directly affects the safety and comfort of vehicle driving, among which cracks, ruts, and subsidence have a particularly significant impact on highway performance. In particular, cracks not only affect the service life of the road surface, but also can cause a series of chain diseases, such as damage to the roadbed caused by rain erosion, and even serious traffic accidents. Therefore, early detection and accurate identification of cracks have become a key problem that needs to be solved in the field of highway maintenance.

[0003] Traditional highway condition detection methods mainly rely on manual inspection and simple image processing technology. Although manual inspection has a certain flexibility, it has low detection efficiency, large errors, and is difficult to adapt to the large-scale and continuous detection needs of expressways. Existing image processing technology often only stays at the stage of simple edge detection and morphological analysis, and cannot effectively deal with noise interference and illumination changes in complex road surface environments. In addition, many systems cannot classify different types of cracks (such as surface cracks and fatigue cracks), making it difficult to provide targeted repair recommendations, resulting in resource waste and low maintenance efficiency.

[0004] At the same time, the existing technology is weak in multi-scale feature analysis and dynamic monitoring of cracks, and lacks the ability to evaluate the trend of crack changes. This makes it difficult for highway management departments to obtain accurate and comprehensive data support when developing long-term maintenance plans. In addition, the visualization and practicality of crack detection results are insufficient, and the output form is mostly a single numerical report, which fails to achieve intuitive expression of the crack condition and is difficult to meet the decision-making needs of different users.

[0005] In summary, the existing technology has significant shortcomings in the accuracy, robustness, classification ability, and practicality of crack detection, and there is an urgent need for an intelligent solution that can efficiently and accurately detect highway conditions in complex road surface environments. SUMMARY

[0006] In view of the shortcomings of the prior art, the present application provides an intelligent system capable of quickly and simply detecting highway conditions, which solves the technical problems of low crack detection accuracy, weak anti-interference ability, insufficient classification ability, and lack of practicality of detection results in the prior art.

[0007] To achieve the above purpose, the present application realizes the following technical scheme: an intelligent system capable of quickly and simply detecting highway conditions, comprising:

[0008] a data acquisition module configured to acquire image data of a road surface through sensors and cameras;

[0009] a data preprocessing module configured to perform denoising, grayscale transformation and edge enhancement processing on the image data;

[0010] an edge detection module configured to perform edge detection on the preprocessed image data based on variational energy functional theory and a partial differential equation model, and extract edge information of the cracks;

[0011] a multi-scale detection module configured to perform multi-scale analysis on the edge information through high-order continuous wavelet transform, and identify edge positions, size and types of the cracks;

[0012] an optimization decision module configured to perform precision optimization on the crack detection results based on an optimization objective function, and output optimized crack detection results;

[0013] a result output module configured to generate a detection report containing crack position coordinates, classification information and repair suggestions according to the optimized crack detection results.

[0014] Preferably, the denoising processing in the data preprocessing module is implemented through Gaussian filtering to remove noise in the image data and retain important edge features.

[0015] Preferably, the edge detection module performs edge extraction by minimizing an energy functional of the image data based on variational energy functional theory, and the energy functional includes an image gradient energy item and a regularization constraint item, specifically:

[0016]

[0017] wherein u is a grayscale intensity function of the image, is a gradient of the image grayscale, represents a p-norm of the image gradient, used to maintain edge features of the cracks, R(u) is a regularization constraint item, λ is a regularization coefficient, and Ω is a definition domain of the image;

[0018] The edge detection module solves a corresponding Euler-Lagrange equation by minimizing the energy functional:

[0019]

[0020] wherein, and are partial derivatives of the image gradient energy item in x and y directions, respectively, is a partial derivative of the energy functional with respect to the image grayscale intensity u.

[0021] Preferably, the edge detection module solves the minimum value of the variational energy functional by gradient descent method, and iteratively extracts the edge information of the crack.

[0022] Preferably, the edge detection module smoothes the non-edge region by a nonlinear diffusion equation based on a partial differential equation model, while enhancing the edge information of the crack.

[0023] Preferably, the multi-scale detection module performs multi-scale analysis based on a high-order continuous wavelet transform, extracts crack edge features of different scales, and determines the specific position and size of the crack by a local maximum value detection method.

[0024] Preferably, the high-order continuous wavelet transform selects a high-order Daubechies wavelet as a mother function to improve the extraction accuracy of the crack edge features.

[0025] Preferably, the optimization decision module optimizes the crack detection result based on an optimization objective function, and the objective function includes a crack region overlap loss, an edge gradient continuity loss and a regularization constraint term.

[0026] Preferably, the result output module further visually processes the optimized crack detection result, outputs a crack image labeling result, and automatically generates a detection report containing repair materials and resource allocation.

[0027] The application also provides an intelligent method for quickly and simply detecting the road condition, comprising the following steps:

[0028] Collecting road surface image data through a high-resolution camera;

[0029] Performing noise removal, gray scale conversion and edge enhancement processing on the image data;

[0030] Extracting edge information of the crack based on variational energy functional theory and partial differential equation model;

[0031] Performing multi-scale feature extraction on the edge information by using a high-order continuous wavelet transform to determine the specific position, size and type of the crack;

[0032] Optimizing the accuracy of the crack detection result based on an optimization objective function containing crack region overlap, edge continuity and regularization constraint;

[0033] Generating a detection report containing crack position coordinates, classification information and repair suggestions, and visually displaying the detection report.

[0034] The application provides an intelligent system for quickly and simply detecting the road condition. The system has the following beneficial effects:

[0035] 1.The present application can effectively extract the edge features of cracks by combining various algorithms such as variational energy functional, partial differential equation and high-order continuous wavelet transform, and can suppress the interference of background noise on the detection results. The optimization decision module further optimizes the accuracy of the crack detection results, ensuring the accuracy and continuity of the detection results, and maintaining high robustness even under complex road surface materials and lighting conditions.

[0036] 2.The present application uses high-order continuous wavelet transform to perform multi-scale analysis on the global features and local details of cracks, and can accurately extract the size, shape and edge information of cracks. Through energy distribution analysis of wavelet coefficients, the system can realize automatic classification of crack types such as surface cracks, fatigue cracks and micro-cracks, providing a scientific basis for maintenance decisions for different types of cracks.

[0037] 3.In the data acquisition and processing stage, the influence of different road surface materials (such as asphalt, cement and gravel pavement) and different lighting conditions (such as sunny day, night, strong light and shadow) is fully considered. Through automatic exposure adjustment, high dynamic range imaging and adaptive parameter optimization, the system has strong scene adaptability and can maintain good detection performance under various environmental conditions.

[0038] 4.The result output module outputs the crack detection results in the form of intuitive images and structured documents through visualization and report generation functions, with clear labeling of crack edge, size and classification information. The detection report contains maintenance recommendations and material resource allocation information, providing comprehensive decision support for highway maintenance departments and greatly improving the scientificity and pertinence of maintenance work. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 is a schematic diagram of the system architecture of the present application;

[0040] Figure 2 is a schematic diagram of the method flow of the present application. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings of the present application specification. Obviously, the described embodiments are only a part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0042] Please refer to the drawings of the present application Figure 1 The present application provides an intelligent system that can quickly and simply detect the condition of highways, which can collect, process, analyze and output crack information of highway pavement in real time, and provide accurate basis for the maintenance and management of highways. The modules of the system will be described in detail as follows.

[0043] The intelligent system for quickly and simply detecting the road condition can include a data acquisition module, a data preprocessing module, an edge detection module, a multi-scale detection module, an optimized decision module, and a result output module.

[0044] In this embodiment, the data acquisition module is mainly used for acquiring image data of the road surface through sensors and cameras, providing basic data input for subsequent image processing and crack detection. The data acquisition module is described in detail below in combination with the specific implementation.

[0045] In this embodiment, the data acquisition module includes a high-resolution industrial camera, a wide-angle lens, a GPS positioning module, and a related image acquisition control unit installed on the detection vehicle. Through the combination of hardware and software, the road surface can be comprehensively monitored under different driving conditions and environments.

[0046] As an option, the camera can use an industrial camera with a resolution of 1920x1080 and a frame rate of 30fps to ensure that the acquired images have sufficient clarity and detail information. The use of a wide-angle lens ensures that the camera can cover a larger detection range, for example, a lens with a viewing angle of 120° or more can be selected.

[0047] Specifically, the driving speed of the detection vehicle should match the acquisition frame rate of the camera. For example, when the vehicle drives at a speed of 20km / h, the acquisition frequency of the camera can be set to 10 frames per second to ensure the continuity of image acquisition and the integrity of road surface coverage.

[0048] In a possible implementation, to improve the adaptability of image acquisition, the data acquisition module can further include an automatic exposure and dynamic range adjustment function to maintain image quality under strong light, shadows, and other complex lighting conditions. In some embodiments, HDR (High Dynamic Range Imaging) technology can be used to improve the detail performance of shadow areas and highlight areas.

[0049] It should be noted that in order to accurately geolocate the crack position, the data acquisition module integrates a GPS positioning module, which records the vehicle's driving trajectory and the corresponding coordinates of image acquisition through one-second positioning refresh. The accuracy of the GPS module can be selected to be sub-meter level (such as differential GPS technology) to ensure that the spatial positioning of the crack is accurate to the specific location of the road.

[0050] As an implementation, the data acquisition module can be combined with an inertial navigation system (INS) to further improve the robustness of vehicle positioning. In the case of weak satellite signals (such as tunnels, mountainous areas), inertial navigation can compensate for positioning errors to ensure that the trajectory of the detection vehicle and the spatial correspondence of the collected images are not affected.

[0051] It should be noted that the installation height and angle of the camera are crucial for image acquisition. For example, in some embodiments, the camera is installed on the top of the vehicle, with a height of 2-3 meters from the ground and an inclination angle of 30-45 degrees, which can cover a wide enough area of the road surface and maintain clear imaging of the cracks.

[0052] In this embodiment, image acquisition also includes an automatic triggering mechanism. For example, the vehicle speed and road area information can be detected by the mileage sensor and road detection sensor, and input into the image acquisition control unit in real time. According to the trigger condition (such as the threshold matching of vehicle speed and acquisition frame rate), the working state of the camera is automatically started and paused. This can reduce the collection of redundant data and improve system efficiency.

[0053] As a further extension, the data acquisition module can also integrate a multispectral camera or an infrared camera to enhance detection capability in complex environments. Specifically, the multispectral camera can collect information in the visible light and near-infrared waveband to enhance the road surface texture details at night or in insufficient light, while the infrared camera can be used to detect cracks in areas with significant temperature differences.

[0054] It can be understood that in order to adapt to different road materials (such as asphalt, cement or gravel pavement), the data acquisition module can flexibly adjust the imaging parameters by collecting images of different spectra or resolutions. For example, brightness control can be increased on cement pavement, while contrast can be enhanced on asphalt pavement to highlight the edge information of the cracks.

[0055] It should be noted that the data acquisition module in this embodiment has high real-time performance. Image data and location information are transmitted in real time to the processing unit of the system through wired or wireless means. Wireless transmission can use a 5G module to ensure low latency and high bandwidth of data transmission when the vehicle is driving at high speed.

[0056] The data acquisition module can be equipped with a binocular camera to realize three-dimensional modeling of the highway pavement. In specific implementation, the depth information of the cracks is calculated from the parallax map of the two cameras through a stereo vision algorithm to distinguish between superficial cracks and deep cracks, thereby providing more accurate data support for the classification and subsequent repair of the cracks.

[0057] In this embodiment, the image data acquisition can also be combined with optical flow analysis to identify dynamic changes in the road surface. For example, during vehicle travel, optical flow can detect local changes in road surface texture, thereby capturing subtle crack features.

[0058] In this embodiment, the data preprocessing module is mainly used for denoising, gray scale transformation and edge enhancement processing of the collected road surface image data, so as to provide high-quality input data for the subsequent edge detection and multi-scale analysis module. The data preprocessing module will be described in detail in combination with the specific implementation mode.

[0059] In this embodiment, the first step of the data preprocessing module is gray scale transformation. For the RGB color image collected by the data collection module, the weighted method is used to convert it into a gray scale image I(x, y). The specific formula is as follows:

[0060] I(x, y) = 0.299R(x, y) + 0.587G(x, y) + 0.114B(x, y)

[0061] Wherein, R(x, y), G(x, y), B(x, y) represent the intensity values of the image in the red, green and blue channels respectively, and I(x, y) represents the converted gray scale image.

[0062] It should be noted that the purpose of gray scale transformation is to simplify the image information while retaining the texture features of the road cracks. In some embodiments, the weight coefficients of each channel can also be adjusted according to the difference in road material, for example, for asphalt pavement, the weight of the green channel can be appropriately increased to better highlight the contrast of the road cracks.

[0063] As an option, the gray scale image is further processed for noise removal in this embodiment. Specifically, the image is smoothed by the method of Gaussian filtering to reduce the interference of random noise on crack detection. The mathematical expression of Gaussian filtering is:

[0064]

[0065] Wherein, σ is the standard deviation of Gaussian distribution, which determines the smoothing strength of the filter kernel. In a possible implementation mode, the typical value range of σ is 1.0 to 2.0 to achieve better noise suppression effect.

[0066] It should be noted that the size of the filter kernel also has an important influence on the denoising effect. For example, a 3x3, 5x5 or larger filter kernel can be used, and the specific size is selected according to the image noise level. In some embodiments, the kernel size can be adjusted by adaptive Gaussian filtering to adapt to the noise distribution of different scenes.

[0067] In this embodiment, in order to further enhance the edge features of the cracks, the image after denoising is processed for edge enhancement. Specifically, the edge enhancement method based on Laplacian operator is adopted. The enhanced image can be represented by the following formula:

[0068]

[0069] where, is the Laplacian operator of the image, representing a discrete approximation of the second-order gradient of the image. Laplacian enhancement can effectively enhance the contrast of high-frequency regions (such as crack edges) in the image and suppress the loss of details in flat regions.

[0070] In one possible implementation, high-pass filtering and Laplacian enhancement can be combined to further highlight the edge features of the cracks. Specifically, high-pass filtering can be implemented through Fourier transform to enhance the edge frequency components of the cracks.

[0071] As a further processing, the image is normalized in this embodiment to map the image data to a fixed range (e.g., [0, 1]), thereby improving the robustness of the subsequent edge detection algorithm. The normalization formula is as follows:

[0072]

[0073] where, I min and I max represent the minimum and maximum pixel values of the image, respectively.

[0074] It should be noted that the normalization process not only standardizes the pixel value range between different images, but also reduces the influence of lighting conditions on crack feature extraction. In some embodiments, the normalized image can be adjusted in combination with the contrast stretching technique to further optimize the separability of the crack region.

[0075] It can be understood that each processing step of the data preprocessing module can be combined according to actual needs. For example, for a road surface scene with low noise, the Gaussian filtering step can be omitted, and edge enhancement and normalization processing can be performed directly; for complex road conditions with unclear crack edges, multi-scale filtering and gradient enhancement methods can be combined.

[0076] As a possible extension, the data preprocessing module can use a machine learning-based adaptive enhancement method. Specifically, by training a neural network model, the de-noising, enhancement, and normalization parameters of the image are adaptively adjusted according to the road surface material and crack features. For example, a convolutional neural network (CNN) is used to extract multi-scale features, which are used as guide information for edge enhancement.

[0077] The data preprocessing module can also combine prior information of edge detection to adjust the preprocessing results. For example, by performing preliminary analysis on the image after edge enhancement, it is determined whether the contrast of the crack edge meets a set threshold, and if not, the weight of the Laplacian enhancement or the Gaussian filtering parameter is dynamically adjusted.

[0078] In this embodiment, the edge detection module is mainly used for further processing of the image data processed by the data preprocessing module. Through an edge detection method based on variational energy functional theory and partial differential equation (PDE) model, edge features of the highway cracks are extracted, and accurate edge information is provided for subsequent multi-scale analysis. The module is described in detail in combination with a specific implementation manner.

[0079] In this embodiment, the edge detection module adopts variational energy functional theory to construct an energy functional model of the image, and realizes edge extraction by minimizing the energy functional. Specifically, the defined energy functional is as follows:

[0080]

[0081] Wherein:

[0082] u represents a gray intensity function of the image;

[0083] is an image gradient energy term, which is used to maintain the significant features of the crack edges;

[0084] R(u) is a regularization term, which is used to smooth the non-edge area and suppress noise;

[0085] λ is a regularization coefficient, which is used to balance the influence of the gradient energy and the regularization term;

[0086] Ω represents the definition domain of the image;

[0087] p≥1 is a parameter for controlling the form of gradient calculation, and a typical value is 2.

[0088] As an option, in order to optimize the calculation efficiency and result accuracy, the energy functional is obtained by solving the corresponding Euler-Lagrange equation, and the equation is specifically as follows:

[0089]

[0090] It should be noted that the equation is numerically solved by gradient descent method to gradually approach the optimal solution. In a possible implementation manner, the iterative formula can be expressed as:

[0091]

[0092] Wherein:

[0093] u n is the gray value of the nth iteration;

[0094] τ is the step size of gradient descent, and a typical value is 0.1-0.01.

[0095] It should be noted that the selection of the step size affects the convergence speed and the stability of the result, and can be optimized through experiments.

[0096] In this embodiment, in order to further enhance the effect of edge extraction, the edge detection module combines a partial differential equation (PDE) diffusion model. Specifically, the model uses the following nonlinear diffusion equation for edge smoothing and enhancement:

[0097]

[0098] wherein:

[0099] I is the input image;

[0100] div represents the divergence operator;

[0101] is an edge stopping function used to control the diffusion rate, defined as:

[0102]

[0103] represents the gradient module of the image;

[0104] K is a gradient threshold parameter, which is usually set according to the gradient distribution of the image.

[0105] Specifically, the diffusion process is numerically solved by finite difference method. In one possible implementation, the discretized equation can be represented as:

[0106]

[0107] wherein:

[0108] represents the gray value of the pixel point (i, j) at the tthiteration;

[0109] Δt is the time step, used to control the iteration step size.

[0110] In some embodiments, in order to adapt to complex road surface features, the edge detection module can combine the advantages of variational energy functional and PDE model, and realize the smooth extraction and enhanced detection of crack edges through a hybrid model. The specific implementation is:

[0111] Use the energy functional to perform global gradient optimization to extract the preliminary edge;

[0112] Based on the PDE diffusion model, the high-noise area is locally smoothed.

[0113] It should be noted that this hybrid approach can take into account the continuity of the global edge and the saliency of the local edge, effectively reducing the cracking phenomenon of the crack edge.

[0114] It can be understood that the edge detection module can also adapt to different scenarios by adjusting the model parameters. For example, on the road surface with complex lighting conditions (such as shadow and highlight areas), the regularization coefficient λ and the gradient threshold K can be dynamically adjusted to improve the distinction between the crack edge and the background. In one possible implementation, the parameter values are automatically estimated in combination with the image contrast analysis method.

[0115] In an extended embodiment, the edge detection module can also be combined with a machine learning model. For example, by training a deep learning model (such as a convolutional neural network based on U-Net), guided detection of crack edges is performed, and the results are used as the initial boundary conditions for the variational calculation or PDE model, thereby further improving the detection effect.

[0116] In this embodiment, the multi-scale detection module is mainly used to analyze the multi-scale features of the crack edges based on the crack edge information output by the edge detection module using high-order continuous wavelet transform (HOCWT). Through multi-scale analysis, different sizes of crack edges can be effectively identified, and the position, size and type features of the cracks can be extracted. The following describes the module in detail in combination with specific technical implementation.

[0117] In this embodiment, the multi-scale detection module extracts and analyzes the multi-scale features of the cracks based on high-order continuous wavelet transform (HOCWT). Specifically, wavelet transform is used to decompose the image signal into high-frequency and low-frequency components at multiple scales, so as to highlight the detailed features of the cracks. Its mathematical definition is as follows:

[0118]

[0119] Where:

[0120] W ψ f(a,b) is the wavelet transform coefficient, representing the image features at scale a and position b;

[0121] f(x) is the input crack edge signal;

[0122] ψ(x) is the wavelet mother function, which adopts high-order Daubechies wavelet in this embodiment;

[0123] a is the scale parameter, which determines the resolution of decomposition;

[0124] b is the translation parameter, representing the position of the feature in space.

[0125] It should be noted that wavelet transform effectively distinguishes the overall shape of large cracks and the detailed information of micro-cracks by gradually extracting the local features of the edges at different scales.

[0126] As an alternative, the mother function of the wavelet transform can adopt a high-order Daubechies wavelet with good time-frequency localization performance. This wavelet function has compact support and high-order orthogonality, and can better preserve the detailed features of the crack in the high-frequency part.

[0127] Specifically, the high-order Daubechies wavelet can capture the tiny details of the crack, such as the refined edges of the crack, when the scale a→0 is small, and can analyze the overall structure and macroscopic shape of the crack when the scale a→∞ is large.

[0128] In a possible implementation, to further improve the effect of wavelet analysis, a multi-scale decomposition and reconstruction method can be used. By decomposing the original image signal f(x) into high-frequency part H(a,b) and low-frequency part L(a,b) of different scales, and then focusing on the analysis of the high-frequency part, the multi-scale features of the crack are obtained.

[0129] It should be noted that the multi-scale detection module not only extracts the position of the crack edge, but also determines the specific position and intensity of the crack through the local maximum value detection algorithm. The specific method is as follows:

[0130] First, the local maximum value of the wavelet transform coefficient is used to judge the intensity distribution of the crack edge:

[0131]

[0132] Where Ω represents the definition domain of the image, and the max operation is used to extract the maximum value point of the edge under the scale a.

[0133] Second, according to the spatial distribution of the maximum value point, the specific position of the crack edge is identified, and the width and depth of the crack are determined through scale correlation analysis.

[0134] In a possible implementation, the multi-scale detection module can also classify the crack according to the distribution of the wavelet transform coefficient. For example, by counting the energy distribution of the wavelet coefficient under different scales, the crack can be divided into the following types:

[0135] Surface crack: showing high-frequency component intensity on small scale;

[0136] Fatigue crack: showing smooth distribution on medium scale;

[0137] Fine crack: not significant on large scale, only has obvious local features on small scale.

[0138] As a further extension, the multiscale detection module in this embodiment can also combine Fourier transform for joint analysis. For example, for some special crack types (such as periodic cracks), Fourier analysis can complement the shortcomings of wavelet transform and extract the periodic characteristics and direction information of the cracks. Specifically, Fourier transform is used to analyze the global frequency distribution, while wavelet transform is used to extract local time-frequency features.

[0139] It should be noted that in some complex scenarios (such as uneven lighting or high noise background), the multiscale detection module can also combine an adaptive wavelet analysis method. By dynamically adjusting the wavelet mother function and scale parameters, the accuracy of crack feature extraction is optimized.

[0140] In this embodiment, in order to improve the real-time performance of the multiscale detection module, the wavelet transform is implemented using a fast wavelet transform (FWT) algorithm. This algorithm effectively reduces the computational complexity through hierarchical decomposition, enabling the module to be applied in real-time detection.

[0141] For example, in some embodiments, the fast wavelet transform algorithm can be further combined with GPU acceleration to improve the efficiency of processing large resolution images. For example, in a 4096x4096 pixel image, through GPU parallel computing, the time of wavelet analysis can be controlled within 200ms.

[0142] In this embodiment, the optimization decision module is mainly used to further improve the detection accuracy of the cracks based on the preliminary crack detection results of the multiscale detection module through an optimization objective function, and to generate high-precision crack feature data for the result output module. The optimization decision module is described in detail below in combination with specific technical implementation methods.

[0143] In this embodiment, the optimization decision module optimizes the crack detection results by constructing an objective function containing multiple evaluation indicators. Specifically, the objective function is composed of the following parts:

[0144] L = aL IoU + bL gradient + gL reg

[0145] Where:

[0146] L IoU : represents the overlap loss of the detection results and the actual crack area, ensuring the accuracy of the detection;

[0147] L gradient : represents the consistency loss of the crack edge gradient, used to maintain the smoothness and continuity of the crack edge;

[0148] L reg: regularization constraint term, used to suppress the interference of noise on crack edge;

[0149] α, β, γ: coefficients for balancing the loss weights, usually set by experiment or dynamically adjusted.

[0150] It should be noted that the target function can effectively improve the accuracy and robustness of the detection result by comprehensively considering the region overlap, edge continuity and noise suppression.

[0151] Specifically, the region overlap loss L IoU is defined as:

[0152]

[0153] wherein:

[0154] A is the predicted crack region;

[0155] B is the real crack region;

[0156] |A∩B| represents the intersection area of the predicted region and the real region;

[0157] |A∪B| represents the union area of the predicted region and the real region.

[0158] As an option, the crack edge gradient loss L gradient is defined based on the difference of the edge gradient field, and the specific formula is:

[0159]

[0160] wherein:

[0161] I pred is the predicted crack edge intensity;

[0162] I true is the real crack edge intensity;

[0163] represents the gradient operator;

[0164] Ω represents the image domain where the crack edge is located.

[0165] It should be noted that the gradient loss can effectively avoid the phenomenon of crack edge breaking or irregularity by constraining the local continuity of the edge.

[0166] The regularization constraint loss L reg is used to limit the complexity of the detection result and suppress the influence of background noise, and is defined as:

[0167] L reg =∫ Ω |ΔI pred |dx

[0168] wherein:

[0169] ΔI pred is the second-order gradient of the crack (Laplacian operator);

[0170] Ω denotes the image domain where the crack is located.

[0171] In one possible implementation, the regularization constraint can be combined with a data smoothing technique to further reduce the background noise through Gaussian filtering.

[0172] In this embodiment, the solution of the optimization objective function uses an iterative optimization algorithm. Specifically, the gradient descent method is used to iteratively minimize the objective function, and the crack edge prediction result is updated. The iterative update formula is:

[0173]

[0174] wherein:

[0175] θ t is the model parameter at the t-th iteration;

[0176] η is the learning rate, which controls the step size of each parameter update;

[0177] is the gradient of the objective function.

[0178] It should be noted that the selection of the learning rate η has an important influence on the optimization effect, and usually a dynamic adjustment method (such as cosine annealing or exponential decay) is used to improve the convergence performance.

[0179] As an option, in order to improve the efficiency and stability of the optimization, an adaptive gradient optimization algorithm (such as the Adam optimizer) can be combined in this embodiment. Specifically, the Adam optimizer effectively improves the convergence speed of the objective function by adaptively weighting the first and second moments of the gradient.

[0180] In some embodiments, the optimization decision module can also combine a multi-stage optimization strategy. The specific implementation is:

[0181] In the initial stage, the crack region is globally optimized to ensure the integrity of the region;

[0182] In the subsequent stage, the crack edge is locally optimized to improve the fineness of the edge.

[0183] In this embodiment, to adapt to different detection scenarios, the weight coefficients a, b, g of the objective function can be adaptively adjusted based on the scene characteristics. For example, in a high-noise scenario, the regularization weight g can be increased to enhance the noise suppression effect; in a complex crack morphology scenario, the gradient weight b can be increased to highlight the edge continuity.

[0184] The optimization decision module can also combine a deep learning model to generate a preliminary crack edge prediction result through a pre-trained neural network, which is used as the initial condition for optimization of the objective function. Specifically, the output result of the deep learning model is used to accelerate the convergence of the objective function and improve the detection accuracy in complex scenarios.

[0185] In this embodiment, the result output module is mainly used to receive the crack detection result output by the optimization decision module, and through data visualization and report generation functions, the detection result is converted into an intuitive output form to provide a scientific basis for subsequent road maintenance. The result output module is described in detail below in combination with a specific implementation manner.

[0186] In this embodiment, the result output module includes a visualization unit and a report generation unit. The visualization unit is mainly used to image mark the detection result to intuitively display the edge, position and size information of the crack; the report generation unit outputs the detection result in the form of a document, including the geographic coordinates, classification type and repair recommendations of the crack.

[0187] Specifically, the visualization unit marks the crack area on the original pavement image based on the input crack edge data. As an option, the crack edge can be highlighted with colored lines, and the color is associated with the crack type, for example:

[0188] Surface cracks are marked in red;

[0189] Fatigue cracks are marked in blue;

[0190] Microcracks are marked in green.

[0191] It should be noted that the size information of the crack is directly displayed on the image in the form of marked text, such as crack width, length and other key information. In addition, a scale line can be superimposed on the image to provide a spatial scale reference for the crack area.

[0192] In one possible implementation manner, the visualization unit supports dynamic scaling and interactive functions, and users can view detailed information of the crack through operations such as dragging and zooming in. For example, by zooming in on the crack area, the morphological characteristics of the crack, such as branching or expansion area, can be further observed.

[0193] For example, to enhance the visualization effect, the result output module can also generate a three-dimensional visualization image of the road cracks by combining 3D modeling technology. By combining crack depth information with edge detection results, a three-dimensional shape of the crack can be generated, providing a basis for evaluating deep cracks.

[0194] In this embodiment, the report generation unit outputs the detection results in the form of a structured document, including the following key contents:

[0195] Crack location: Based on the geographic coordinate information of the GPS positioning module, the location of the crack is marked in the form of latitude and longitude;

[0196] Crack type: The specific type of the crack is indicated by the classification result of the multi-scale detection module, such as surface cracks, fatigue cracks, etc.

[0197] Crack size: Including the width, length and depth of the crack;

[0198] Crack status: Evaluate the severity of the crack through the detection result, and indicate whether emergency repair is needed;

[0199] Repair recommendations: According to the type and size of the crack, generate the corresponding repair plan, including the required materials, tools and human resources, etc.

[0200] As an option, the report generation unit supports output in multiple document formats, such as PDF, Excel and HTML format, to meet the use requirements in different scenarios.

[0201] It should be noted that, in order to improve the practicality of the report, in this embodiment, the report generation unit combines statistical analysis functions. For example, by summarizing the crack detection results, a crack distribution map and statistical table of the entire road are generated to show the overall distribution of cracks and the type proportion.

[0202] In one possible implementation, the report generation unit also supports time series analysis functions. For example, by comparing and analyzing multiple detection results, a crack change trend chart is generated to evaluate the expansion speed and deterioration trend of the crack. This function has important reference value for long-term planning of highway maintenance.

[0203] In this embodiment, in order to improve the flexibility of the system, the result output module supports real-time data stream output function. Specifically, the detection results can be transmitted in real time to the remote monitoring terminal through wireless network (such as 5G communication module), realizing remote crack monitoring and management.

[0204] The result output module can also integrate AI voice broadcast function to voice prompt key crack information. For example, when a serious crack is detected, the system can issue an alarm sound and voice broadcast the specific location and type of the crack.

[0205] In general, the system of the present application is composed of a data acquisition module, a data preprocessing module, an edge detection module, a multi-scale detection module, an optimized decision module, and a result output module. Through sensors and cameras, image data of the road is collected, and advanced image processing algorithms and mathematical models (such as variational energy functional, high-order continuous wavelet transform, and nonlinear partial differential equation) are used to extract, analyze, and optimize the edges of the cracks, combined with visualization and report generation, to realize real-time detection, classification, positioning, and maintenance suggestion generation of the cracks. The system has high precision, high robustness, and real-time performance, and provides scientific and comprehensive technical support for road maintenance and management.

[0206] Please refer to the attached Figure 2 The present application also provides an intelligent method for quickly and simply detecting the road conditions, comprising the following steps:

[0207] Step S1: Collecting image data of the road surface through a high-resolution camera.

[0208] In this embodiment, an industrial-grade high-resolution camera installed on a detection vehicle collects image data of the road surface in real time. The resolution, frame rate, and viewing angle of the camera are adjusted according to the road surface detection requirements, and at the same time, a GPS positioning module is used to record the geographical coordinates corresponding to the images in real time, providing spatial information support for crack location calibration.

[0209] Step S2: Noise removal, gray scale conversion, and edge enhancement processing of the image data.

[0210] In this embodiment, the collected raw image data is first subjected to noise removal through Gaussian filtering, which removes random noise while preserving the edge features of the image. Then, the image is subjected to gray scale conversion, converting the RGB image into a gray scale image, simplifying the data structure and highlighting the crack texture. Finally, an edge enhancement algorithm is used to enhance the saliency of the crack edges, providing high-quality input data for the subsequent edge detection module.

[0211] Step S3: Extracting edge information of the cracks based on variational energy functional theory and partial differential equation model.

[0212] In this embodiment, the preprocessed image is subjected to edge detection to extract the edge information of the cracks. The module uses an optimization method based on variational energy functional theory to extract the salient edges of the cracks, and at the same time, a partial differential equation model is used to smooth the non-edge regions, enhancing the continuity and stability of the edges. The data after edge extraction retains the accurate outline of the cracks, providing a basis for multi-scale analysis.

[0213] Step S4: Using high-order continuous wavelet transform to extract multi-scale features of the edge information, and determining the specific position, size, and type of the cracks.

[0214] In this embodiment, the data after edge detection is subjected to multi-scale feature extraction through high-order continuous wavelet transform. Wavelet transform extracts global features and local details of the crack at different scales, locates the specific position of the crack through the local maximum detection algorithm, and determines the size and morphology of the crack according to the energy distribution of the wavelet coefficients. Finally, the module classifies the crack as a surface crack, a fatigue crack, or a micro-crack according to the feature results, providing a basis for subsequent optimization decisions.

[0215] Step S5: Based on the optimization objective function containing crack area overlap, edge continuity and regularization constraints, the accuracy of the crack detection result is optimized.

[0216] In this embodiment, the preliminary detection results output by the multi-scale detection module are subjected to accuracy optimization through the optimization decision module. The module comprehensively evaluates the crack area overlap, edge continuity and background noise suppression according to the objective function, and gradually adjusts the edge positioning and crack morphology of the detection results through iterative optimization algorithms. The optimized detection results have higher accuracy and robustness, and can accurately reflect the actual condition of the crack.

[0217] Step S6: Generate a detection report containing crack position coordinates, classification information and repair suggestions, and perform visual display.

[0218] In this embodiment, the optimized crack detection results are generated into a structured detection report through the result output module. The report contains the geographical position, classification type, size information and repair suggestions of the crack, and can be output in PDF, HTML or other common document formats. At the same time, the module superimposes the crack information on the original image in an intuitive visual form for display, with different colors used to mark the crack edges, so that users can quickly identify and evaluate the distribution and severity of the crack.

[0219] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent system that can quickly and easily detect road conditions, characterized by: include: A data acquisition module is used to collect image data of the road surface through sensors and cameras; A data preprocessing module, used for performing denoising, grayscale conversion and edge enhancement on the image data; The edge detection module is used to perform edge detection on the preprocessed image data based on variational energy functional theory and partial differential equation model to extract the edge information of the crack; A multi-scale detection module is used to perform multi-scale analysis on the edge information through high-order continuous wavelet transform to identify the edge position, size and type of the crack; An optimization decision module, configured to optimize the accuracy of the crack detection result based on an optimization objective function and output the optimized crack detection result; A result output module is used to generate a detection report including crack location coordinates, classification information and repair suggestions based on the optimized crack detection results; The edge detection module is based on the variational energy functional theory and performs edge extraction by minimizing the energy functional of the image data. The energy functional includes the image gradient energy term and the regularization constraint term, specifically: Where μ is the grayscale intensity function of the image, is the gradient of image grayscale, Represents the p-norm of the image gradient, which is used to maintain the edge features of the crack, R(u) is the regularization constraint term, λ is the regularization coefficient, and Ω is the domain of the image; The edge detection module solves the corresponding Euler-Lagrange equation by minimizing the energy functional: in, and are the partial derivatives of the image gradient energy term in the x and y directions, is the partial derivative of the energy functional with respect to the image grayscale intensity u; The optimization decision module optimizes the crack detection results based on an optimization objective function, wherein the objective function includes a crack region overlap loss, an edge gradient continuity loss, and a regularization constraint term.

2. The intelligent system for quickly and easily detecting road conditions according to claim 1, characterized in that: The denoising process in the data preprocessing module is implemented by Gaussian filtering to remove noise in the image data and retain important edge features.

3. The intelligent system for quickly and easily detecting road conditions according to claim 1, characterized in that: The edge detection module solves the minimum value of the variational energy functional by the gradient descent method and iteratively extracts the edge information of the crack.

4. The intelligent system for quickly and easily detecting road conditions according to claim 1, characterized in that: The edge detection module is based on a partial differential equation model and smoothes non-edge areas through a nonlinear diffusion equation while enhancing edge information of cracks.

5. The intelligent system for quickly and easily detecting road conditions according to claim 1, characterized in that: The multi-scale detection module performs multi-scale analysis based on high-order continuous wavelet transform, extracts crack edge features of different scales, and determines the specific location and size of the cracks through local maximum detection method.

6. The intelligent system for quickly and easily detecting road conditions according to claim 5, characterized in that: The high-order continuous wavelet transform uses high-order Daubechies wavelet as a mother function to improve the extraction accuracy of crack edge features.

7. The intelligent system for quickly and easily detecting road conditions according to claim 1, characterized in that: The result output module further performs visualization processing on the optimized crack detection results, outputs crack image annotation results, and automatically generates a detection report including repair materials and resource allocation.

8. An intelligent method for quickly and easily detecting road conditions, applied to the system according to any one of claims 1 to 7, characterized in that: The following steps are involved: Collect road surface image data through high-resolution cameras; performing noise removal, grayscale conversion and edge enhancement processing on the image data; Extract the edge information of the crack based on variational energy functional theory and partial differential equation model; Using high-order continuous wavelet transform to perform multi-scale feature extraction on the edge information to determine the specific location, size and type of the crack; The crack detection results are optimized based on an optimization objective function that includes crack area overlap, edge continuity, and regularization constraints. Generate an inspection report containing crack location coordinates, classification information and repair suggestions, and display it visually.

Images