Method for predicting pavement service life based on road construction indentation distribution

Abstract

The invention discloses a method for predicting the service life of a road surface based on road construction indentation distribution, and belongs to the technical field of road data processing, and the method comprises the steps: obtaining three-dimensional point cloud data of the surface of the road surface, separating an indentation region from the three-dimensional point cloud data, and generating a first data set; acquiring image data from the first data set, and generating a second feature set containing indentation boundary data; calculating a point cloud height difference according to the height data of the first data set to obtain a third feature set containing depth feature data; inputting a current second feature set containing indentation boundary information and a third feature set containing depth features into a fourth prediction model, and outputting a material hardness prediction value; and calculating a pavement life prediction value according to the material hardness prediction value and the road load data. The method for predicting the road surface service life based on the road construction indentation distribution solves the problem that a traditional method is difficult to accurately evaluate the road surface indentation condition and predict the service life.

Description

Technical Field

[0001] The present invention relates to the technical field of road data processing, and in particular to a method for predicting the service life of a pavement based on road construction indentation distribution. Background Art

[0002] Road engineering is crucial to the durability and safety of infrastructure. Accurately predicting the service life of a pavement is directly related to road maintenance costs and traffic efficiency. Indentations are a common characteristic of pavement damage during road construction, and their distribution significantly impacts pavement life, making them the foundation for high-quality road management. However, existing methods for predicting pavement life rely primarily on single-source indentation statistics or macroscopic damage assessments, ignoring diverse characteristics such as indentation morphology, depth, and road loads. This results in insufficiently accurate prediction models, making them difficult to adapt to complex construction environments and variable traffic loads. Summary of the Invention

[0003] In order to overcome the defects of the prior art, the present invention provides a method for predicting the service life of a pavement based on the distribution of road construction indentations to solve the above problems.

[0004] The technical solution adopted by the present invention to solve the technical problem is: a method for predicting the service life of a pavement based on the distribution of road construction indentations, comprising the following steps: S1: Acquire three-dimensional point cloud data of a road surface, separate the indentation area from the three-dimensional point cloud data and generate a first data set; S2: Obtain image data from the first dataset and use an edge detection mechanism to generate a second feature set containing indentation boundary data; S3: Calculate the point cloud height difference according to the height data of the first data set to obtain a third feature set containing depth feature data; S4: Inputting the second feature set including indentation boundary data, the third feature set including depth feature data, and the measured value of material hardness corresponding to the same historical time into a convolutional neural network for training, establishing a mapping relationship between indentation features and material hardness, and generating a fourth prediction model, wherein the indentation features include indentation boundary data and depth feature data; inputting the current second feature set including indentation boundary information and the third feature set including depth features into the fourth prediction model, and outputting a predicted value of material hardness; S5: Calculating a pavement life prediction value based on the material hardness prediction value output by the fourth prediction model and combining it with the road load data.

[0005] Preferably, the step S1 includes the following sub-steps: S11: using laser scanning technology to obtain three-dimensional point cloud data containing spatial information of the road surface to obtain a first point cloud; S12: Multiple indentation regions are separated from the first point cloud by setting a growth threshold in the region growing segmentation algorithm, each indentation region corresponds to a sub-dataset containing indentation features, and a first data set is obtained, wherein the growth threshold includes the normal vector angle of the point cloud and the distance between two adjacent point clouds.

[0006] Optionally, step S13 is further included after step S12, and step S13 includes: The number of points in each sub-dataset is compared with the preset threshold. If the number of points in a sub-dataset is lower than the preset threshold, the missing points in the sub-dataset are supplemented by the bilinear interpolation algorithm to generate an optimized sub-dataset.

[0007] It is worth noting that step S2 includes the following sub-steps: S21: Acquire image data of each sub-dataset from the first data set; S22: converting the image data into a grayscale image using the cv2.cvtColor function in the OpenCV library to obtain a first grayscale image set; S23: for each grayscale image in the first grayscale image set, edge detection is performed using the cv2.Canny function in the OpenCV library, with a high threshold and a low threshold preset as a first threshold and a second threshold, respectively, to extract indentation boundary data including an indentation boundary position; S24: Integrate the indentation boundary data into the corresponding sub-datasets, and combine all the sub-datasets to obtain a second feature set.

[0008] Specifically, step S3 includes the following sub-steps: S31: Obtaining point cloud height data of each sub-dataset in the first data set, and calculating the height difference of adjacent point clouds for each sub-dataset to obtain a sub-dataset height difference set; S32: extracting the maximum value, minimum value and median from the height difference set of the sub-dataset, and integrating them into the corresponding sub-dataset to obtain a sub-dataset containing depth feature data; S33: Combining the sub-datasets containing the depth feature data into a third feature set.

[0009] Specifically, in step S32, the sub-dataset containing the depth feature data is classified using the K-means algorithm to determine the depth feature data category of the sub-dataset and obtain a classification feature set.

[0010] It is worth noting that step S4 includes the following sub-steps: S41: merging the indentation boundary data in the second feature set and the depth feature data in the third feature set into a feature input matrix; S42: forward propagating the feature input matrix through a convolutional neural network, using a ReLU activation function and an Adam optimizer to calculate feature maps of the convolution layer and the pooling layer to obtain an initial feature map set; S43: using a mean square error loss function to calculate the error between the initial feature mapping set and the actual value of the material hardness, adjusting the network parameters through back propagation to obtain optimized network parameters; and obtaining a mapping relationship based on the optimized network parameters and the convergence condition; S44: Constructing a fourth prediction model based on the mapping relationship, and calculating the material hardness prediction value by inputting new indentation boundary data and depth feature data.

[0011] Optionally, in step S43, if the optimized network parameters meet a preset convergence condition, the current feature map in the initial feature map set is associated with the material property label to generate a mapping relationship; If the optimized network parameters do not meet the preset convergence conditions, the forward propagation and the backward propagation are repeated to adjust the network parameters until the convergence conditions are met and the mapping relationship is obtained.

[0012] Preferably, step S5 includes the following sub-steps: S51: Obtaining the material hardness prediction value output by the fourth prediction model and the traffic flow intensity in the load data to obtain a pavement life prediction data set; performing standardization processing on the pavement life prediction data set to obtain a standardized data set; S52: extracting features from the standardized data set using a random forest algorithm to obtain a feature vector set; S53: Calculating the importance of each feature based on the feature vector set, and determining the weight of a weighted fusion mechanism based on the importance of each feature; processing the feature vector set using the weighted fusion mechanism to obtain a set of pavement life prediction values; S54: Selecting a median from the pavement life prediction value set to determine a final pavement life prediction value.

[0013] The beneficial effects of the present invention are as follows: in the method for predicting the service life of a pavement based on the distribution of road construction indentations, by acquiring the three-dimensional point cloud data of the pavement, a stereoscopic segmentation algorithm is used to separate the indentation area, and based on the indentation area, the corresponding indentation boundary data and depth feature data are obtained. The feature set is input into a convolutional neural network to establish a mapping relationship between the indentation features and the material hardness values. Finally, the material hardness prediction value output by the fourth prediction model is combined with the load data for analysis to achieve an accurate prediction of the pavement life. The present invention effectively solves the problem that traditional methods are difficult to accurately evaluate the pavement indentation condition and predict the life through multi-dimensional feature extraction and deep learning models, providing a reliable basis for pavement maintenance decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A flowchart of a method for predicting pavement service life based on road construction indentation distribution in one embodiment of the present invention; Figure 2 A step-by-step flow chart of step S1 in one embodiment of the present invention; Figure 3 A step-by-step flow chart of step S2 in one embodiment of the present invention; Figure 4 A step-by-step flow chart of step S3 in one embodiment of the present invention; Figure 5 A step-by-step flow chart of step S4 in one embodiment of the present invention; Figure 6 FIG. 1 is a step-by-step flow chart of step S5 in one embodiment of the present invention. DETAILED DESCRIPTION

[0015] The following is a further description of specific embodiments of the present invention in conjunction with the accompanying drawings. It should be noted that the description of these embodiments is intended to facilitate understanding of the present invention and does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0016] like Figure 1-6 As shown, a method for predicting the service life of a pavement based on the distribution of road construction indentations includes the following steps: S1: Acquire three-dimensional point cloud data of a road surface, separate the indentation area from the three-dimensional point cloud data and generate a first data set; S2: Obtain image data from the first dataset and use an edge detection mechanism to generate a second feature set containing indentation boundary data; S3: Calculate the point cloud height difference according to the height data of the first data set to obtain a third feature set containing depth feature data; S4: Inputting the second feature set including indentation boundary data, the third feature set including depth feature data, and the measured value of material hardness corresponding to the same historical time into a convolutional neural network for training, establishing a mapping relationship between indentation features and material hardness, and generating a fourth prediction model, wherein the indentation features include indentation boundary data and depth feature data; inputting the current second feature set including indentation boundary information and the third feature set including depth features into the fourth prediction model, and outputting a predicted value of material hardness; S5: Calculating a pavement life prediction value based on the material hardness prediction value output by the fourth prediction model and combining it with the road load data.

[0017] In the method for predicting the service life of a pavement based on the distribution of road construction indentations, three-dimensional point cloud data of the pavement is obtained, and the indentation area is separated using a stereoscopic segmentation algorithm. Based on the indentation area, the corresponding indentation boundary data and depth feature data are obtained. The feature set is input into a convolutional neural network to establish a mapping relationship between the indentation features and the material hardness values. Finally, the material hardness prediction value output by the fourth prediction model is combined with the load data for analysis to achieve an accurate prediction of the pavement life. Through multi-dimensional feature extraction and deep learning models, the present invention effectively solves the problem that traditional methods are difficult to accurately evaluate the pavement indentation condition and predict the life, providing a reliable basis for pavement maintenance decisions.

[0018] It is worth noting that if Figure 2 As shown, step S1 includes the following sub-steps: S11: Acquire three-dimensional point cloud data containing spatial information of a road surface using laser scanning technology to obtain a first point cloud; Denoise and downsample the first point cloud using Gaussian filtering to generate a smoothed point cloud dataset, thereby obtaining a preprocessed first point cloud; S12: Multiple indentation regions are separated from the first point cloud by setting a growth threshold in the region growing segmentation algorithm, each indentation region corresponds to a sub-dataset containing indentation features, and a first data set is obtained, wherein the growth threshold includes the normal vector angle of the point cloud and the distance between two adjacent point clouds.

[0019] For example, laser scanning technology can be used to acquire three-dimensional point cloud data of road surfaces using a vehicle-mounted LiDAR system. The LiDAR emits millions of laser pulses per second, recording the positions of reflection points on the road surface and generating a highly accurate spatial coordinate dataset. For example, during a highway scan, the LiDAR collected data at a point spacing of 0.01 meters, forming a first point cloud containing tens of millions of points that accurately reflect the road surface's geometric characteristics. This high-density data provides a reliable foundation for subsequent analysis.

[0020] In one possible implementation, Gaussian filtering is used to denoise and downsample the first point cloud. In principle, Gaussian filtering smoothes the point cloud through weighted averaging, reducing noise interference while also reducing the data volume through downsampling. For example, applying a Gaussian filter to the first point cloud with a standard deviation of 0.02 meters removes random noise points and then reduces the point spacing from 0.01 meters to 0.05 meters yields the preprocessed first point cloud. This reduces the data volume by approximately 80%, while still preserving key road features and improving computational efficiency.

[0021] Specifically, the region growing segmentation algorithm processes the first point cloud to isolate the indentation regions. This algorithm segments the point cloud into distinct regions based on spatial proximity and normal vector differences. For example, for road indentations, a growth threshold is set, such that points with a normal vector angle less than 15 degrees and a distance less than 0.03 meters are grouped together. In one scan, multiple indentation regions were successfully isolated, resulting in the first dataset. After segmentation, the indentation region point cloud is isolated from other road features, facilitating subsequent analysis.

[0022] Preferably, step S13 is further included after step S12, and step S13 includes: The number of points in each sub-dataset is compared with the preset threshold. If the number of points in a sub-dataset is lower than the preset threshold, the missing points in the sub-dataset are supplemented by the bilinear interpolation algorithm to generate an optimized sub-dataset.

[0023] It should be noted that if the number of points in the indentation area corresponding to a sub-dataset within the first dataset falls below a threshold, such as fewer than 1000 points, bilinear interpolation is used to fill in the missing points. Bilinear interpolation generates new points to fill sparse areas based on a weighted average of neighboring points. For example, if a specific indentation area has only 800 points, interpolation can be used to increase the number to 1500, forming an optimized sub-dataset. This method enhances the continuity of the point cloud and improves the accuracy of feature extraction.

[0024] Optional, such as Figure 3 As shown, step S2 includes the following sub-steps: S21: Acquire image data of each sub-dataset from the first data set; S22: converting the image data into a grayscale image using the cv2.cvtColor function in the OpenCV library to obtain a first grayscale image set; S23: for each grayscale image in the first grayscale image set, edge detection is performed using the cv2.Canny function in the OpenCV library, with a high threshold and a low threshold preset as a first threshold and a second threshold, respectively, to extract indentation boundary data including an indentation boundary position; S24: Integrate the indentation boundary data into the corresponding sub-datasets, and combine all the sub-datasets to obtain a second feature set.

[0025] Specifically, the image data of each sub-dataset is obtained from the first data set, typically by projecting the 3D point cloud onto a 2D plane to generate a corresponding image representation. Exemplarily, point cloud projection can be achieved through orthogonal projection, mapping 3D coordinates to a 2D pixel grid. For example, in an analysis of indentations on a highway pavement, a sub-dataset contains an indentation area of 1,500 points, which is projected onto an image with a resolution of 512×512 pixels, with the pixel value reflecting the height information of the point. This projection preserves the geometric features of the indentation, facilitating subsequent image processing.

[0026] In one possible implementation, the image data is converted into a grayscale image using the cv2.cvtColor function of the OpenCV library to generate a first set of grayscale images. In principle, cv2.cvtColor converts an RGB image into a single-channel grayscale image, reducing color interference and highlighting brightness differences in the indentation area. For example, the projected image of an indentation area contains road surface texture and shadows. By converting it to a grayscale image, the texture details are simplified, making the outline of the indentation area easier to identify. Preferably, the image data can be preprocessed before conversion, such as adjusting the contrast, to enhance the characteristic expression of the grayscale image.

[0027] It should be noted that, for the first grayscale image set, the cv2.Canny function is used to perform edge detection and extract indentation boundary data. The Canny algorithm is based on gradient calculation and combines high and low thresholds to screen edge points. Exemplarily, the first threshold is set to 200 and the second threshold is set to 100 to detect significant edges in grayscale images. For example, a grayscale image contains a concave outline of the indentation area. After detection, the Canny algorithm generates a binary image and only retains edge pixels. This method effectively separates the indentation boundary from the background and improves the accuracy of boundary positioning. For example, for different indentation shapes, the threshold of Canny detection can be dynamically adjusted. If the indentation outline is blurred, the second threshold can be lowered to 80 to increase the sensitivity of edge detection.

[0028] It is understandable that in order to determine the boundary position information corresponding to the indentation boundary data and generate the second feature set, the edge pixels need to be mapped back to the three-dimensional space. In one embodiment, by recording the coordinate correspondence during projection, the two-dimensional coordinates of the edge pixels are back-projected onto the three-dimensional point cloud to obtain the spatial position of the boundary points. For example, an indentation edge contains 300 pixels, and after back-projection, 300 three-dimensional coordinates are generated to form the second feature set. These coordinates accurately describe the position of the indentation boundary, which is convenient for subsequent geometric analysis.

[0029] Specifically, if Figure 4 As shown, step S3 includes the following sub-steps: S31: Obtaining point cloud height data of each sub-dataset in the first data set, and calculating the height difference of adjacent point clouds for each sub-dataset to obtain a sub-dataset height difference set; S32: extracting the maximum value, minimum value and median from the height difference set of the sub-dataset, and integrating them into the corresponding sub-dataset to obtain a sub-dataset containing depth feature data; S33: Combining the sub-datasets containing the depth feature data into a third feature set.

[0030] Specifically, in the processing of road surface point cloud data, each sub-dataset in the first data set contains the height information of the three-dimensional point cloud, and the road surface height is usually expressed as the Z-axis coordinate. The process of obtaining point cloud height data involves extracting the Z value from the point cloud file to form a height data set. For example, a highway road surface sub-dataset contains 2,000 points, each with X, Y, and Z coordinates, where the Z value reflects the change in road surface height, such as the height difference of indentation or raised areas. In one possible implementation, by traversing the point cloud data in each sub-dataset in the first data set, the Z value is directly extracted to form a sub-dataset height difference set corresponding to each sub-dataset, which is convenient for subsequent analysis.

[0031] Exemplarily, when calculating the height difference of adjacent point clouds, the definition of adjacent points needs to be determined. Usually, the nearest neighbor of each point is determined based on the spatial adjacency of the point cloud, such as by the k-nearest neighbor algorithm. For example, the Z value of a point in a sub-dataset is 10.5 cm, and the Z value of its nearest neighbor is 10.8 cm, with a height difference of 0.3 cm. Traverse all points in the sub-dataset to generate a height difference set. It should be noted that the height difference reflects the local undulating characteristics of the road surface, such as the steepness of the indentation edge. In one embodiment, the sub-dataset contains 1000 height difference values, ranging from -0.5 cm to 0.7 cm, indicating that there are slight bumps on the road surface.

[0032] It can be understood that after extracting the maximum, minimum, and median values from the sub-dataset height difference set, these values are integrated into the corresponding sub-dataset to obtain the sub-dataset containing depth feature data. For example, the maximum height difference of a sub-dataset is 0.7 cm, the minimum is -0.5 cm, and the median is 0.1 cm. These statistical features summarize the magnitude and distribution of local road height variations.

[0033] It is worth noting that, in step S32, the K-means algorithm is used to classify the sub-dataset containing the depth feature data, determine the depth feature data category of the sub-dataset, and obtain a classification feature set.

[0034] In one possible implementation, the K-means algorithm is used to classify sub-datasets containing depth feature data to determine the categories of the depth feature data. For example, the depth feature data categories are divided into three categories: flat, slight indentation, and significant indentation. The feature vector (maximum value 0.7, minimum value -0.5, median 0.1) of a sub-dataset containing depth feature data is classified as the slight indentation category. The classification feature set records the category label of each sub-dataset containing depth feature data, so that the depth feature data of each sub-dataset can be more intuitively reflected. It should be noted that K-means iteratively optimizes the cluster center to ensure that the classification results reflect the inherent differences in road surface characteristics.

[0035] Preferably, Figure 5 As shown, step S4 includes the following sub-steps: S41: merging the indentation boundary data in the second feature set and the depth feature data in the third feature set into a feature input matrix; adjusting the distribution of the feature input matrix by a Z-score normalization method to obtain a normalized feature input matrix; S42: forward propagating the feature input matrix through a convolutional neural network, using a ReLU activation function and an Adam optimizer to calculate feature maps of the convolution layer and the pooling layer to obtain an initial feature map set; S43: using a mean square error loss function to calculate the error between the initial feature mapping set and the actual value of the material hardness, adjusting the network parameters through back propagation to obtain optimized network parameters; and obtaining a mapping relationship based on the optimized network parameters and the convergence condition; S44: Constructing a fourth prediction model based on the mapping relationship, and calculating the material hardness prediction value by inputting new indentation boundary data and depth feature data.

[0036] In one embodiment, a highway point cloud subdataset contains 1,000 points. The boundary data records the X, Y, and Z coordinates of the indentation edge points, such as (50, 30, 20) and (52, 32, 22), and the depth features, such as a median of 0.2 cm. When merging, the boundary coordinates and depth features are aligned point by point to form a feature input matrix, where each row contains the boundary and depth information of the point.

[0037] Specifically, Z-score normalization adjusts the distribution of the feature input matrix to ensure that features of different dimensions are comparable. For example, boundary coordinate values range widely, such as 50 to 100, while depth feature values range from small, such as 0.1 to 0.5 centimeters. By calculating the mean and standard deviation of each feature, the normalized values are concentrated in the range [-1, 1], making them easier for neural networks to process and thus reflecting the relative changes in boundary and depth.

[0038] In one embodiment, a convolutional neural network processes a standardized feature matrix, employing the ReLU activation function to enhance nonlinear representations and the Adam optimizer to accelerate gradient descent. The convolutional layer extracts local features, such as the continuity of indentation edges; the pooling layer compresses data to preserve key information. For example, a feature input matrix is processed with a 3x3 convolution kernel to generate a feature map that reflects the spatial relationships between edge points.

[0039] Subsequently, based on this mapping, new data is input into the fourth prediction model to predict material hardness. For example, the feature map of a certain indentation area corresponds to a material hardness prediction value of 50 MPa, indicating that the road surface material is relatively strong.

[0040] Specifically, in step S43, if the optimized network parameters meet the preset convergence conditions, the current feature map in the initial feature map set is associated with the material property label to generate a mapping relationship; If the optimized network parameters do not meet the preset convergence conditions, the forward propagation and the backward propagation are repeated to adjust the network parameters until the convergence conditions are met and the mapping relationship is obtained.

[0041] The mean squared error loss function measures the difference between the feature map and the actual material hardness value, for example, if the predicted material hardness value deviates from the actual material hardness value by 0.05. Backpropagation adjusts the network parameters until the loss converges, thus constructing the fourth prediction model.

[0042] It is worth noting that if Figure 6 As shown, step S5 includes the following sub-steps: S51: Obtaining the material hardness prediction value output by the fourth prediction model and the traffic flow intensity in the load data to obtain a pavement life prediction data set; performing standardization processing on the pavement life prediction data set to obtain a standardized data set; S52: extracting features from the standardized data set using a random forest algorithm to obtain a feature vector set; S53: Calculating the importance of each feature based on the feature vector set, and determining the weight of a weighted fusion mechanism based on the importance of each feature; processing the feature vector set using the weighted fusion mechanism to obtain a set of pavement life prediction values; S54: Selecting a median from the pavement life prediction value set to determine a final pavement life prediction value.

[0043] For example, in the pavement life prediction data set, the material hardness prediction value reflects the pavement's compressive strength, and the traffic flow intensity in the load data is counted by the road monitoring system, such as an average daily number of 5,000 vehicles. In one possible implementation, the standardization process aims to unify the characteristic scales of different dimensions. The numerical ranges of hardness and flow characteristics vary greatly, such as hardness values of 40 to 60 and flow values of 1,000 to 10,000. Standardization uses the Z-score method to convert each feature into a distribution with a mean of 0 and a standard deviation of 1. For example, the hardness value of a certain road section of 50MPa is standardized to 0.5, and the flow of 8,000 vehicles is standardized to 0.8. This processing facilitates subsequent algorithm analysis.

[0044] Specifically, the random forest algorithm is used for feature extraction, constructing multiple decision trees to evaluate the relationships between features. In one embodiment, for 1,000 road section samples, each tree analyzes the impact of material hardness and traffic flow characteristics on lifespan, generating a set of feature vectors. For example, the feature vector for a particular road section is [0.5, 0.8], corresponding to material hardness and traffic flow, respectively. This method can capture nonlinear relationships between features.

[0045] It should be noted that feature importance calculation is based on the output of the random forest, reflecting the contribution of each feature to lifespan prediction. For example, if the importance of material hardness is 0.6 and the importance of traffic flow is 0.4, weights are determined accordingly, such as a weight of 0.6 for material hardness and 0.4 for traffic flow. This weighting method emphasizes the influence of key features. Preferably, a weighted fusion mechanism integrates feature vectors to generate a set of pavement lifespan prediction values. Specifically, set the initial lifespan to X and calculate the corresponding pavement lifespan prediction value S = X + 0.6A - 0.4B for each road section, where A is the material hardness prediction value and B is the traffic flow intensity in the load data. For example, a weighted calculation of the feature vectors for a certain road section yields a pavement lifespan prediction value of 15 years. The pavement lifespan prediction value set contains pavement lifespan prediction values for 1,000 road sections, such as 12 to 18 years. Selecting the median of these 1,000 pavement lifespan prediction values as the final pavement lifespan prediction value effectively avoids interference from extreme values. For example, the median of 16 years is used as the final lifespan prediction for a road section.

[0046] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. It is apparent to those skilled in the art that various changes, modifications, substitutions, and variations to these embodiments may be made without departing from the principles and spirit of the present invention, and these changes and modifications still fall within the scope of protection of the present invention.

Claims

1. A method for predicting the service life of a pavement based on the distribution of road construction indentations, characterized in that: The following steps are involved: S1: Acquire three-dimensional point cloud data of a road surface, separate the indentation area from the three-dimensional point cloud data and generate a first data set; S2: Obtain image data from the first dataset and use an edge detection mechanism to generate a second feature set containing indentation boundary data; S3: Calculate the point cloud height difference according to the height data of the first data set to obtain a third feature set containing depth feature data; S4: Inputting the second feature set including indentation boundary data, the third feature set including depth feature data, and the measured value of material hardness corresponding to the same historical time into a convolutional neural network for training, establishing a mapping relationship between indentation features and material hardness, and generating a fourth prediction model, wherein the indentation features include indentation boundary data and depth feature data; inputting the current second feature set including indentation boundary information and the third feature set including depth features into the fourth prediction model, and outputting a predicted value of material hardness; S5: Calculating a pavement life prediction value based on the material hardness prediction value output by the fourth prediction model and combining it with the road load data.

2. A method for predicting pavement service life based on road construction indentation distribution according to claim 1, characterized in that: The step S1 includes the following sub-steps: S11: using laser scanning technology to obtain three-dimensional point cloud data containing spatial information of the road surface to obtain a first point cloud; S12: Multiple indentation regions are separated from the first point cloud by setting a growth threshold in the region growing segmentation algorithm, each indentation region corresponds to a sub-dataset containing indentation features, and a first data set is obtained, wherein the growth threshold includes the normal vector angle of the point cloud and the distance between two adjacent point clouds.

3. A method for predicting pavement service life based on road construction indentation distribution according to claim 2, characterized in that: After step S12, step S13 is further included, and step S13 includes: The number of points in each sub-dataset is compared with the preset threshold. If the number of points in a sub-dataset is lower than the preset threshold, the missing points in the sub-dataset are supplemented by the bilinear interpolation algorithm to generate an optimized sub-dataset.

4. A method for predicting pavement service life based on road construction indentation distribution according to claim 3, characterized in that: The step S2 includes the following sub-steps: S21: Acquire image data of each sub-dataset from the first data set; S22: converting the image data into a grayscale image using the cv2.cvtColor function in the OpenCV library to obtain a first grayscale image set; S23: for each grayscale image in the first grayscale image set, edge detection is performed using the cv2.Canny function in the OpenCV library, with a high threshold and a low threshold preset as a first threshold and a second threshold, respectively, to extract indentation boundary data including an indentation boundary position; S24: Integrate the indentation boundary data into the corresponding sub-datasets, and combine all the sub-datasets to obtain a second feature set.

5. A method for predicting pavement service life based on road construction indentation distribution according to claim 4, characterized in that: The step S3 includes the following sub-steps: S31: Obtaining point cloud height data of each sub-dataset in the first data set, and calculating the height difference of adjacent point clouds for each sub-dataset to obtain a sub-dataset height difference set; S32: extracting the maximum value, minimum value and median from the height difference set of the sub-dataset, and integrating them into the corresponding sub-dataset to obtain a sub-dataset containing depth feature data; S33: Combining the sub-datasets containing the depth feature data into a third feature set.

6. A method for predicting pavement service life based on road construction indentation distribution according to claim 5, characterized in that: In step S32, the sub-dataset containing the depth feature data is classified using the K-means algorithm to determine the depth feature data category of the sub-dataset and obtain a classification feature set.

7. A method for predicting pavement service life based on road construction indentation distribution according to claim 6, characterized in that: The step S4 includes the following sub-steps: S41: merging the indentation boundary data in the second feature set and the depth feature data in the third feature set into a feature input matrix; S42: forward propagating the feature input matrix through a convolutional neural network, using a ReLU activation function and an Adam optimizer to calculate feature maps of the convolution layer and the pooling layer to obtain an initial feature map set; S43: using a mean square error loss function to calculate the error between the initial feature mapping set and the actual value of the material hardness, adjusting the network parameters through back propagation to obtain optimized network parameters; and obtaining a mapping relationship based on the optimized network parameters and the convergence condition; S44: Constructing a fourth prediction model based on the mapping relationship, and calculating the material hardness prediction value by inputting new indentation boundary data and depth feature data.

8. The method for predicting pavement service life based on road construction indentation distribution according to claim 7, characterized in that: In the step S43, if the optimized network parameters meet the preset convergence conditions, the current feature map in the initial feature map set is associated with the material property label to generate a mapping relationship; If the optimized network parameters do not meet the preset convergence conditions, the forward propagation and the backward propagation are repeated to adjust the network parameters until the convergence conditions are met and the mapping relationship is obtained.

9. A method for predicting pavement service life based on road construction indentation distribution according to claim 8, characterized in that: The step S5 comprises the following sub-steps: S51: Obtaining the material hardness prediction value output by the fourth prediction model and the traffic flow intensity in the load data to obtain a pavement life prediction data set; performing standardization processing on the pavement life prediction data set to obtain a standardized data set; S52: extracting features from the standardized data set using a random forest algorithm to obtain a feature vector set; S53: Calculating the importance of each feature based on the feature vector set, and determining the weight of a weighted fusion mechanism based on the importance of each feature; processing the feature vector set using the weighted fusion mechanism to obtain a set of pavement life prediction values; S54: Selecting a median from the pavement life prediction value set to determine a final pavement life prediction value.

Images