A Method and System for Predicting Road Skid Resistance Based on Hyperspectral Image Texture Features

Abstract

This invention discloses a method and system for predicting road surface skid resistance performance based on hyperspectral image texture features. The method includes: acquiring airborne hyperspectral images of the road surface; performing wavelet transform on each band of the hyperspectral image to obtain high-frequency components, and calculating Shannon entropy and energy on the high-frequency components; then using Shannon entropy and energy to construct band scores to filter out spectral feature bands reflecting road surface characteristics; further, filtering texture parameters based on the texture features of the spectral feature bands to form a texture parameter dataset; finally, inputting the texture parameter dataset into a CGS-XGBoost model for training and predicting the skid resistance performance of other unknown road surfaces. This invention expands upon the limitations of existing evaluation indicators in representing road surface texture, predicting skid resistance performance from the perspective of macroscopic image texture, avoiding the drawbacks of traditional methods with small measurement areas, and significantly improving the accuracy and breadth of road surface skid resistance performance prediction.

Description

Technical Field

[0001] This invention relates to the field of road traffic safety monitoring and intelligent maintenance technology, specifically to a method and system for predicting road surface skid resistance performance based on hyperspectral image texture features, which combines airborne hyperspectral image texture analysis and machine learning algorithms. This method is a non-contact prediction method for road surface skid resistance performance and is suitable for efficient assessment and early warning of skid resistance performance in large-scale road networks. Background Technology

[0002] The skid resistance of asphalt pavements is a key factor affecting driving safety, directly impacting vehicle braking efficiency and driving stability. With significant increases in traffic volume, the rise in the proportion of heavy vehicles, and accelerated road aging, the decline in the skid resistance of asphalt pavements has become increasingly prominent, posing a significant road safety hazard. This problem is particularly severe in suburban and remote areas with relatively scarce road maintenance resources. Due to the lack of systematic skid resistance monitoring methods, the gradual degradation of pavement skid resistance is difficult to identify in a timely manner, significantly increasing the probability of traffic accidents. Therefore, achieving large-scale, efficient, and high-precision monitoring and prediction of asphalt pavement skid resistance is of significant theoretical and practical value for improving road maintenance management efficiency and ensuring road traffic safety.

[0003] The measurement of skid resistance of asphalt pavements mainly relies on three traditional methods: pendulum friction coefficient measuring instruments, lateral force coefficient testing vehicles, and vehicle-mounted friction testers. However, these methods all have certain limitations: pendulum friction coefficient measuring instruments use a single-point static measurement method, which is simple to operate but has a limited measurement range, making it difficult to assess the skid resistance performance of large-scale road sections; lateral force coefficient testing vehicles can achieve dynamic continuous measurement and are suitable for high-grade roads, but the equipment purchase and maintenance costs are high, and they are poorly adaptable to small-scale monitoring; vehicle-mounted friction testers rely on dedicated vehicles, and their applicability is limited by road conditions, making it difficult to widely apply to road networks of different grades. Overall, traditional measurement methods cannot simultaneously meet the needs of large-scale coverage, high-precision prediction, and low-cost implementation, especially lacking non-contact solutions based on remote sensing data.

[0004] In recent years, UAV hyperspectral imaging technology has shown broad application prospects in the field of road engineering due to its advantages of high efficiency, wide-area coverage, non-destructive testing characteristics, and low cost. Hyperspectral images reflect the material state and surface structure of asphalt pavement materials through different spectral bands. The texture features generated based on spectral bands can quantify the spatial distribution characteristics of pavement structure. Since pavement structure directly affects skid resistance, combining the texture information analysis of airborne hyperspectral images with measured pavement skid resistance data can achieve rapid, non-contact prediction of asphalt pavement skid resistance. This solves the problem that traditional remote sensing technology cannot directly correlate with pavement mechanical properties, providing a new technical means for pavement maintenance management and traffic safety assessment. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and apply hyperspectral imaging technology in the field of road mechanical performance testing. By combining the texture features of airborne images and the CGS-XGBoost algorithm, it provides a method and system for predicting road skid resistance performance based on the texture features of hyperspectral images. This enables large-scale and rapid acquisition of road skid resistance performance, improves the accuracy of road skid resistance performance assessment, and allows the road skid resistance performance to be presented in a visual manner in the later stages.

[0006] Therefore, the present invention provides the following technical solution:

[0007] On the one hand, the technical solution of the present invention provides a method for predicting the anti-skid performance of road surfaces based on hyperspectral image texture features, comprising the following steps:

[0008] Step 1: Acquire airborne hyperspectral images of the road surface and crop out the road surface area, and measure the anti-skid performance index of the road surface sample points as training labels;

[0009] Step 2: Screening spectral feature bands, which involves performing wavelet transform on each band of the hyperspectral image to obtain high-frequency components, calculating Shannon entropy and energy on the high-frequency components, and then using Shannon entropy and energy to construct band scores to screen out spectral feature bands that reflect road surface characteristics.

[0010] Step 3: Filter texture parameters, that is, calculate the texture parameters of the hyperspectral image corresponding to each spectral feature band, filter out the texture parameters that are most correlated with anti-skid performance in each spectral feature band, and form a texture parameter dataset;

[0011] Step 4: Construct a road surface skid resistance performance prediction model based on the XGBoost model, using the texture parameter dataset as the model input and training the model using the training labels;

[0012] Step 5: Input the texture parameters of the airborne hyperspectral image of the road surface to be tested into the trained road skid resistance performance prediction model to obtain skid resistance performance index.

[0013] More preferably, in step 2, during the screening of spectral feature bands, for each decomposition layer obtained by wavelet decomposition for each band, the following is performed:

[0014] Firstly, Shannon entropy and energy are calculated on the high-frequency components of the decomposition layer;

[0015] Then, the scores corresponding to the decomposition layers are calculated;

[0016] Next, the scores of all decomposition layers are weighted and summed to obtain the band score;

[0017] Finally, the spectral characteristic bands are selected based on the band scores;

[0018] Among them, the scores corresponding to the decomposition layer E represents energy, and F represents Shannon entropy. It is a minimum value.

[0019] More preferably, the formulas for the Shannon entropy and the energy are as follows:

[0020]

[0021]

[0022]

[0023] In the formula, , It is the first in the high-frequency component coefficients The, the There are several values; N is the total number of high-frequency components; It is the normalized probability.

[0024] More preferably, the process of screening texture parameters in step 3 is to first use the gray-level co-occurrence matrix to calculate the texture parameters of the hyperspectral image of each spectral feature band, and then screen out the texture parameters that are most correlated with anti-skid performance in each spectral feature band.

[0025] The texture parameters of the hyperspectral image are a combination or partial combination of mean, variance, homogeneity, contrast, variability, entropy, energy, and correlation.

[0026] More preferably, the relationship between mean, variance, homogeneity, contrast, dissimilarity, entropy, energy, correlation, and the gray-level co-occurrence matrix is ​​as follows:

[0027]

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035] in, Let be the element values ​​corresponding to gray values ​​i and j in the gray-level co-occurrence matrix, and G be the number of gray levels. The mean, For variance, Homogeneity For contrast, For the sake of difference, For entropy, For energy, For correlation, It is the overall average gray value. It is a local minimum; and These are the expected values ​​in the row and column directions of the gray-level co-occurrence matrix, respectively. and It is the standard deviation of the gray-level co-occurrence matrix.

[0036] More preferably, in step 4, a staged grid search method based on coupling constraints is introduced to optimize the hyperparameter combination of the XGBoost model to obtain the CGS-XGBoost model, thereby constructing a pavement skid resistance performance prediction model based on the CGS-XGBoost model. The hyperparameter combination includes: tree depth max_depth, number of base learners T, and learning rate. ;

[0037] The process of introducing a staged grid search method based on coupling constraints to optimize the hyperparameter combination of the XGBoost model is as follows:

[0038] First, initialize the partial tree depth max_depth, the number of base learners T, and the learning rate using equal-interval step sizes. This leads to the construction of multiple sets of hyperparameter combinations, and the learning rate is defined under a fixed tree depth. There is a coupling relationship with the number of base learners T: , is a constant representing the balance factor between the learning rate and the number of base learners;

[0039] Secondly, based on the GridSearchCV algorithm, cross-validation is used to pre-search different hyperparameter combinations, calculate the error of each hyperparameter combination, and then select the hyperparameter combination with the smallest error to determine the optimal combination for each tree depth max_depth. value;

[0040] Next, based on the discrete tree depth max_depth and the corresponding discrete optimal... The value is obtained by using interpolation methods to discretely determine the optimal value. The value extends to the full tree depth range;

[0041] Then, based on the complete tree depth range and the corresponding optimal Value and learning rate The number of base learners is calculated, and then a set of hyperparameter combinations is constructed.

[0042] Finally, based on the GridSearchCV algorithm, cross-validation is used to filter the hyperparameter combinations in the hyperparameter combination set to determine the optimal hyperparameter combination.

[0043] More preferably, when performing pre-search on different hyperparameter combinations using cross-validation and calculating the error of each hyperparameter combination, five-fold cross-validation and mean squared error are used to select the correct combination. The minimum combination of hyperparameters is given by the following formula:

[0044]

[0045]

[0046] In the formula, These are measured values. It is a predicted value. This represents a set of hyperparameter combinations. Let represent the mean squared error of the k-th fold validation, and n represent the total number of samples in the k-th fold validation set. The circular index represents each sample corresponding to the k-th fold.

[0047] Secondly, the present invention also provides a system based on the above prediction method, comprising sequentially connected or interconnected components:

[0048] The hyperspectral image processing module is used to acquire airborne hyperspectral images of the road surface and crop out the road surface area, as well as measure the anti-skid performance index of the road surface sample points as training labels;

[0049] The spectral feature band filtering module is used to filter spectral feature bands. It performs wavelet transform on each band of the hyperspectral image to obtain high-frequency components, calculates Shannon entropy and energy on the high-frequency components, and then uses Shannon entropy and energy to construct band scores to filter out spectral feature bands that reflect road surface characteristics.

[0050] The texture parameter filtering module is used to filter texture parameters, calculate the texture parameters of the hyperspectral image corresponding to each spectral feature band, and filter out the texture parameters that are most correlated with anti-skid performance in each spectral feature band to form a texture parameter dataset.

[0051] The model building and training module is used to build a road surface skid resistance performance prediction model based on the XGBoost model. The texture parameter dataset is used as the model input, and the model is trained using training labels.

[0052] The prediction module is used to input the texture parameters of the airborne hyperspectral image of the road surface to be tested into the trained road skid resistance performance prediction model to obtain skid resistance performance index.

[0053] Thirdly, the present invention also provides a computer device, comprising: one or more processors and a memory storing one or more computer programs;

[0054] The processor invokes a computer program to implement the steps of the method for predicting road surface skid resistance based on hyperspectral image texture features.

[0055] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, which is invoked by a processor to implement:

[0056] The steps of the method for predicting road surface skid resistance performance based on hyperspectral image texture features.

[0057] Beneficial effects

[0058] This invention proposes a method and system for predicting road surface skid resistance performance based on hyperspectral image texture features. The method utilizes hyperspectral images acquired by UAVs to select spectral feature bands reflecting road surface characteristics. Texture parameters are calculated using these selected spectral feature bands, and texture parameters with high correlation to skid resistance performance are selected to form a texture parameter dataset. An XGBoost model is then used for training and prediction. The proposed method significantly expands the application scope of hyperspectral remote sensing technology in road engineering safety monitoring, solves the problem of road network monitoring in remote areas, and reduces costs by 60%. The high-resolution image data acquired by the airborne hyperspectral imaging system can effectively cover a wide-area road network, increasing the measurement area by 100 times compared to traditional methods, overcoming the limitations of traditional ground-based detection methods.

[0059] This invention further optimizes the XGBoost model by introducing a staged grid search method based on coupling constraints to optimize the hyperparameter combination of the XGBoost model, resulting in the CGS-XGBoost model. The parameters are optimized independently for GridSearch CV, ignoring the coupling relationship between parameters, so that the CGS-XGBoost model obtains the optimal hyperparameter combination, further ensuring the prediction performance of the model.

[0060] The technical solution of this invention utilizes the high-frequency components derived from wavelet decomposition, which reveal the texture and detail information of an image. Therefore, calculating Shannon entropy on the high-frequency components can reflect the complexity of the texture; lower entropy indicates a more ordered and concentrated texture. Selecting high-energy high-frequency components allows for the extraction of clearer and more structurally robust textures. Thus, the technical solution of this invention uses Shannon entropy and energy to select wavebands with clear, ordered, and representative textures, further ensuring the accuracy of the entire prediction method. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating the method for predicting road surface skid resistance performance based on airborne hyperspectral image texture features provided in an embodiment of the present invention.

[0062] Figure 2 These are airborne hyperspectral images of three road sections;

[0063] Figure 3 It is a visualization result of the selected single-band image and eight texture feature parameters;

[0064] Figure 4 This is a comparison chart of predicted anti-skid performance and measured anti-skid performance;

[0065] Figure 5 It is a visualization of pendulum value prediction for some late-stage aging road surfaces;

[0066] Figure 6 It is a visualization of pendulum value prediction for some mid-aged road surfaces;

[0067] Figure 7 It is a visualization of pendulum value prediction at the junction of the early and middle stages of aging;

[0068] Figure 8 This is a hardware connection diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. The technical features involved in the various embodiments of the invention described below can be combined with each other as long as they do not conflict with each other.

[0070] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0071] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0072] The present invention proposes a method for predicting the anti-skid performance of road surfaces based on the texture features of hyperspectral images. This method can achieve large-scale, efficient, and non-contact prediction of the anti-skid performance of road surfaces. The present invention will be further described below with reference to embodiments.

[0073] Example 1:

[0074] This embodiment provides a method for predicting the anti-skid performance of road surfaces based on hyperspectral image texture features, including the following steps:

[0075] Step 1: Acquire airborne hyperspectral image data of the road surface and crop out the road area, and select sample points on the road surface and measure the anti-skid performance index of the sample points.

[0076] like Figure 2 As shown, this embodiment selects three road sections with different aging levels, namely, early, middle, and late aging stages. The hyperspectral image is obtained by a UAV with 164 wavelength bands ranging from 350 to 1002 nm. The road surface area is cropped from the acquired hyperspectral image, and obvious markings on the road surface are selected as image control points. The ground point coordinates are obtained using Real-Time Kinematic (RTK) for geometric correction of the hyperspectral image. The verification process is feasible with existing technology and therefore will not be described in detail. In some embodiments, the actual pendulum value of the marked sample points on the road surface is measured using a pendulum meter as an indicator of anti-skid performance; in other embodiments, it is also feasible to use other indicators in the art to represent anti-skid performance.

[0077] Step 2: Screening spectral feature bands. Perform wavelet transform on each band of the hyperspectral image to obtain high-frequency components, and calculate Shannon entropy and energy on the high-frequency components. Then, use Shannon entropy and energy to construct band scores to screen out spectral feature bands that reflect road surface characteristics.

[0078] In some embodiments, wavelet multi-level decomposition is employed, and for each decomposition layer obtained by wavelet decomposition for each band, the following steps are performed: First, the Shannon entropy and energy are calculated on the high-frequency components of the decomposition layer; then, the score corresponding to the decomposition layer is calculated; next, the scores of all decomposition layers are weighted and summed as the band score; finally, the spectral feature bands are selected based on the band scores.

[0079] In other embodiments, wavelet single-layer decomposition also falls within the scope of protection of this invention, the difference being the precision.

[0080] This embodiment uses the Daubechies 4th order wavelet (DB4) to perform wavelet decomposition on each band of the road surface airborne hyperspectral image. The specific process is as follows:

[0081] Step 21: Use Daubechies 4th order wavelet (DB4) to decompose each band of the road surface airborne hyperspectral image, perform high-pass and low-pass filtering on the row direction of the image and downsample it.

[0082] The results after processing in the row direction are then subjected to high-pass and low-pass filtering and downsampling to obtain four components: one low-frequency component (LL) and three high-frequency components (including vertical detail component (LH), horizontal detail component (HL), and diagonal detail component (HH)).

[0083] In this embodiment, a single wavelet decomposition yields three high-frequency components. These high-frequency components represent the texture and detail information of the image. Calculating Shannon entropy on these high-frequency components reflects the complexity of the texture; lower entropy indicates a more ordered and concentrated texture. Selecting high-energy high-frequency components allows for the extraction of clearer and more structurally robust textures. Therefore, Shannon entropy and energy can be used to select wavebands with clear, ordered, and representative textures. Thus, the formulas for calculating energy (E) and Shannon entropy (F) are:

[0084] (1)

[0085] (2)

[0086] (3)

[0087] in, It is the first in the high-frequency component coefficients There are several values; N is the total number of high-frequency components; It is the normalized probability; It is a very small value to prevent division by zero.

[0088] Then calculate the score for each decomposition layer using the following formula:

[0089] (4)

[0090] For example, the scores calculated from two layers of wavelet decomposition are weighted and summed in a 6:4 ratio to form a comprehensive score. The optimal bands with prominent texture features and low noise are selected, namely band 129 (wavelength: 862 nm), band 133 (wavelength: 874 nm), and band 133 (wavelength: 874 nm). The union of the optimal spectral feature bands selected from the three road sections yields two feature bands, namely band 129 (wavelength: 862 nm) and band 133 (wavelength: 874 nm). It should be understood that the weights of each decomposition layer can be adaptively adjusted according to the prediction accuracy during the weighted summation process.

[0091] In some embodiments, one or more bands with the highest scores are selected as the optimal bands and used as spectral feature bands. In other feasible embodiments, the method of selecting bands based on their scores, while meeting accuracy requirements, also falls within the protection scope of this invention, such as using bands with scores exceeding a set threshold as spectral feature bands.

[0092] Step 3: Filter texture parameters. Calculate the texture parameters of the hyperspectral image corresponding to each spectral feature band, and filter out the texture parameters that are most correlated with anti-skid performance in each spectral feature band to form a texture parameter dataset.

[0093] In this embodiment, the gray-level co-occurrence matrix is ​​preferably used to extract the hyperspectral image texture features of each spectral feature band, calculate texture parameters, and select the texture parameters that are most correlated with anti-skid performance to form a texture parameter dataset. The specific steps are as follows:

[0094] Step 31: For the selected hyperspectral imagery reflecting road surface characteristics, perform geometric correction on the spectral feature bands, calculate the gray-level co-occurrence matrix of the hyperspectral imagery for each spectral feature band, and further calculate texture parameters to quantify texture features (including mean, variance, homogeneity, contrast, dissimilarity, entropy, energy, and correlation), generating a texture feature map. Statistically analyze 10 samples centered on the sample points. Average texture parameters for a 10-pixel area.

[0095] The formulas for calculating the gray-level co-occurrence matrix and texture features are as follows:

[0096] (5)

[0097] (6)

[0098] (7)

[0099] (8)

[0100] (9)

[0101] (10)

[0102] (11)

[0103] (12)

[0104] (13)

[0105] (14)

[0106] in, Let represent the element values ​​corresponding to gray values ​​i and j in the gray-level co-occurrence matrix, M be the number of rows of pixels in the image, and N be the number of columns of pixels in the image. Image location The gray value at that location, where G is the gray level. It is a given direction vector. It is the normalized gray-level co-occurrence matrix; The mean, For variance, Homogeneity For contrast, For the sake of difference, For entropy, For energy, For correlation, It is the overall average gray value. It is a minimum value (to prevent the occurrence of) ); and These are the expected values ​​in the row and column directions of the gray-level co-occurrence matrix, respectively. and It is the standard deviation of the gray-level co-occurrence matrix.

[0107] It should be understood that the preferred texture parameters in this embodiment include mean, variance, homogeneity, contrast, variability, entropy, energy, and correlation; in other feasible embodiments, the texture parameters may also include combinations of other parameters or partial combinations of the above parameters. This invention does not impose specific limitations in this regard.

[0108] Step 32: For each spectral feature band, calculate the correlation between the texture parameters of the spectral feature band and the anti-slip performance of the sample points, select the texture parameters whose correlation meets the preset requirements, and then combine the texture parameters selected from all spectral feature bands to form a texture parameter dataset.

[0109] For example, in the above example, for each selected spectral feature band, namely band 129 (wavelength: 862 nm) and band 133 (wavelength: 874 nm), the correlation between the eight texture parameters of each spectral feature band and the anti-slip performance of the sample points is calculated. Texture parameters with a correlation greater than 0.7 are selected. Five texture parameters (including mean, variance, homogeneity, contrast, and variability) are selected for band 129, and seven texture parameters (including mean, variance, homogeneity, contrast, variability, entropy, and energy) are selected for band 133. Finally, a total of 13 texture parameters are selected to form a texture parameter dataset.

[0110] Step 4: Construct a road surface skid resistance performance prediction model based on the XGBoost model, using the texture parameter dataset as the model input and training the model using training labels.

[0111] In some embodiments, the optimization of steps 2 and 3 above improves the selection of spectral feature bands and texture parameters, thereby making the feature parameters of the input XGBoost model unique and significantly improving the model prediction accuracy. Even when using a common XGBoost model, it can meet the requirements of the technical solution of this invention.

[0112] In this embodiment, in order to improve the overall effect, a staged grid search method based on coupling constraints is preferred to optimize the hyperparameter combination of the XGBoost model to obtain the CGS-XGBoost model (Coupled-GridSearchCV-XGBoost). Then, a road surface skid resistance performance prediction model based on the CGS-XGBoost model is constructed. The airborne hyperspectral texture parameter dataset is input into the road surface skid resistance performance prediction model for model training to obtain the trained prediction model.

[0113] In this embodiment, the process of optimizing the XGBoost pavement skid resistance prediction model using a staged grid search method based on coupling constraints is as follows:

[0114] Step 41: Divide the texture parameter dataset (obtained from the hyperspectral image of the road surface where the sample points are located through step 32) to obtain the model training set and the validation set. The model training set is used for training the CGS-XGBoost model, and the model validation set is used to verify the correctness of the CGS-XGBoost model. For example, 80% is used for training the model and 20% is used as the test set for testing.

[0115] Step 42: Initialize the partial tree depth (max_depth), the number of base learners T, and the learning rate using equal-interval step sizes. For example, first set equally spaced tree depths ([1, 4, 7, 10]) and learning rates ([0.1, 0.3, 0.5, 0.7, 0.9]), and initially set the number of base learners to 100, increasing to 1500 in increments of 200, constructing multiple sets of hyperparameter combinations. With a fixed tree depth, the learning rate... There is a coupling relationship with the number of base learners T:

[0116] (15)

[0117] in, is a constant representing the balance factor between the learning rate and the number of base learners.

[0118] Based on the GridSearchCV method, five-fold cross-validation is used to pre-search different hyperparameter combinations, calculate the mean squared error (MSE) of each combination, and select the hyperparameter combination with the smallest CV_Err to determine the optimal combination at that tree depth. The error calculation formula for the 50% cross-validation is as follows:

[0119] (16)

[0120] (17)

[0121] in, These are measured values. It is a predicted value. This represents a set of hyperparameter combinations. Let represent the mean squared error of the k-th fold validation, and n represent the total number of samples in the k-th fold validation set. The circular index represents each sample corresponding to the k-th fold.

[0122] In the examples, the optimal tree depths (max_depth) are 1, 4, 7, and 10. The values ​​are 7, 3, 3, and 3 respectively.

[0123] Step 43: Use a one-dimensional linear interpolation method to convert the discrete optimal... Values ​​are extended to the full tree depth range for optimal results. The value is continuously distributed at different tree depths, and the known depth is... Values ​​are interpolated to the full tree depth range (1 to 10), for example, a tree depth of 4 corresponds to... The value will be assigned to the adjacent tree depths of 3 and 5.

[0124] Step 44: After completing the interpolation, based on the complete tree depth range and the corresponding optimal... Value and learning rate The number of base learners is calculated, and a set of hyperparameter combinations is constructed. This set of combinations is then evaluated again using GridSearchCV to determine the optimal combination of hyperparameters for the model.

[0125] For example, combining tree depths of 1–10 with learning rates of 0.01–0.10 (step size of 0.01), according to The number of base learners is calculated based on the learning rate and the value, and the final set of hyperparameter combinations is constructed. Finally, the hyperparameter combination selected using this method is: learning rate of 0.06, tree depth of 1, and number of base learners of 117.

[0126] Step 45: Based on the training set, the CGS-XGBoost model is trained using the optimal hyperparameter combination to obtain the final road surface skid resistance prediction model. The coefficient of determination for the training set of this experimental model is 0.9571, the root mean square error for the test set is 3.1769 and the coefficient of determination is 0.8652, and the root mean square error for the validation set is 2.6979 and the coefficient of determination is 0.9009.

[0127] Step 5: Input the airborne road surface texture features to be tested into the trained prediction model, and output the road surface skid resistance performance prediction result. That is, output the predicted pendulum value of the prediction point.

[0128] The comparison chart of the predicted pendulum value and the measured pendulum value at the prediction point in this implementation case is shown below. Figure 4 As shown, Figure 5 , Figure 6 These are visualizations of pendulum value predictions for road surfaces in the late and middle stages of aging. Figure 7 This is a visualization of pendulum value predictions for road surfaces in the early and middle stages of aging. From... Figure 5-7 The results show that the model can predict road surfaces at different aging stages, and the more severe the aging, the weaker the anti-skid performance of the road surface and the lower the predicted pendulum value.

[0129] Example 2

[0130] This embodiment provides a system based on the road surface skid resistance prediction method, which includes a hyperspectral image processing module, a spectral feature band screening module, a texture parameter screening module, a model building and training module, and a prediction module that are connected in sequence or to each other.

[0131] The hyperspectral image processing module is used to acquire airborne hyperspectral images of the road surface, crop out the road surface area, and measure the anti-skid performance indicators of road surface sample points as training labels. Hyperspectral image data, such as that acquired by a drone, can be used as training labels.

[0132] The spectral feature band filtering module is used to filter spectral feature bands. It performs wavelet transform on each band of the hyperspectral image to obtain high-frequency components, calculates Shannon entropy and energy on the high-frequency components, and then uses Shannon entropy and energy to construct band scores to filter out spectral feature bands that reflect road surface characteristics.

[0133] The texture parameter filtering module is used to filter texture parameters, calculate the texture parameters of the hyperspectral image corresponding to each spectral feature band, and filter out the texture parameters that are most correlated with anti-skid performance in each spectral feature band to form a texture parameter dataset.

[0134] The model building and training module is used to build a road surface skid resistance performance prediction model based on the XGBoost model. The texture parameter dataset is used as the model input, and the model is trained using training labels.

[0135] The prediction module is used to input the texture parameters of the airborne hyperspectral image of the road surface to be tested into the trained road skid resistance performance prediction model to obtain skid resistance performance index.

[0136] For details on the implementation process of each module, please refer to the methods described above; they will not be repeated here. It should be understood that the above division of functional modules is merely a logical functional division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the integrated units described above can be implemented in hardware or as software functional units.

[0137] Example 3

[0138] This invention provides a computer device, including: one or more processors and a memory storing one or more computer programs;

[0139] The processor invokes a computer program to implement the steps of the method for predicting road surface skid resistance based on hyperspectral image texture features.

[0140] In some embodiments, the following is specifically performed:

[0141] Step 1: Acquire airborne hyperspectral images of the road surface and crop out the road surface area, and measure the anti-skid performance index of the road surface sample points as training labels;

[0142] Step 2: Screening spectral feature bands. Perform wavelet transform on each band of the hyperspectral image to obtain high-frequency components, and calculate Shannon entropy and energy on the high-frequency components. Then, use Shannon entropy and energy to construct band scores to screen out spectral feature bands that reflect road surface characteristics.

[0143] Step 3: Filter texture parameters, calculate the texture parameters of the hyperspectral image corresponding to each spectral feature band, filter out the texture parameters that are most correlated with anti-skid performance in each spectral feature band, and form a texture parameter dataset;

[0144] Step 4: Construct a road surface skid resistance performance prediction model based on the XGBoost model, using the texture parameter dataset as the model input and training the model using the training labels;

[0145] Step 5: Input the texture parameters of the airborne hyperspectral image of the road surface to be tested into the trained road skid resistance performance prediction model to obtain skid resistance performance index.

[0146] For details on the implementation of each step, please refer to the aforementioned embodiment of the method for predicting road surface skid resistance based on hyperspectral image texture features.

[0147] In some embodiments, such as Figure 8 As shown, the electronic components of the electronic terminal include:

[0148] The processor 1600 can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor 1600 is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.

[0149] The memory 1700 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1700 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1700, and the processor 1600 calls and executes the algorithm program of the road surface anti-skid performance prediction method based on hyperspectral image texture features of the embodiments of this invention.

[0150] The input / output interface 1800 is used to implement information input and output.

[0151] The communication interface 1900 is used to enable communication and interaction between this device and other devices. Communication can be achieved via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0152] Bus 2000 transmits information between various components of the device, such as processor 1600, memory 1700, input / output interface 1800, and communication interface 1900.

[0153] The processor 1600, memory 1700, input / output interface 1800 and communication interface 1900 are connected to each other within the device via bus 2000.

[0154] Example 4

[0155] This invention also provides a computer-readable storage medium storing a computer program that is invoked by a processor to implement:

[0156] The steps of the method for predicting road surface skid resistance performance based on hyperspectral image texture features.

[0157] In some embodiments, the following is specifically performed:

[0158] Step 1: Acquire airborne hyperspectral images of the road surface and crop out the road surface area, and measure the anti-skid performance index of the road surface sample points as training labels;

[0159] Step 2: Screening spectral feature bands. Perform wavelet transform on each band of the hyperspectral image to obtain high-frequency components, and calculate Shannon entropy and energy on the high-frequency components. Then, use Shannon entropy and energy to construct band scores to screen out spectral feature bands that reflect road surface characteristics.

[0160] Step 3: Filter texture parameters, calculate the texture parameters of the hyperspectral image corresponding to each spectral feature band, filter out the texture parameters that are most correlated with anti-skid performance in each spectral feature band, and form a texture parameter dataset;

[0161] Step 4: Construct a road surface skid resistance performance prediction model based on the XGBoost model, using the texture parameter dataset as the model input and training the model using the training labels;

[0162] Step 5: Input the texture parameters of the airborne hyperspectral image of the road surface to be tested into the trained road skid resistance performance prediction model to obtain skid resistance performance index.

[0163] For details on the implementation of each step, please refer to the aforementioned embodiment of the method for predicting road surface skid resistance based on hyperspectral image texture features.

[0164] The readable storage medium is a computer-readable storage medium, which can be an internal storage unit of the hardware and software device described in any of the foregoing embodiments, such as the hard drive or memory of the controller. The readable storage medium can also be an external storage device of the controller, such as a plug-in hard drive, Smart MediaCard (SMC), Secure Digital (SD) card, or Flash Card equipped on the controller. Further, the readable storage medium can include both internal storage units and external storage devices of the controller. The readable storage medium is used to store the computer program and other programs and data required by the controller. The readable storage medium can also be used to temporarily store data that has been output or will be output.

[0165] Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned readable storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0166] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This application refers to flowchart illustrations and / or instructions executed by a processor of a method, apparatus (system), and computer program product according to embodiments of this application to create means for implementing the functions specified in one or more flowchart illustrations and / or one or more block diagrams. These computer program instructions may also be stored in a computer-readable storage medium capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowchart illustrations and / or one or more block diagrams. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more blocks of a block diagram.

[0167] It should be emphasized that the examples described in this invention are illustrative rather than limiting. Therefore, this invention is not limited to the examples described in the specific embodiments. Any other embodiments derived by those skilled in the art based on the technical solutions of this invention, without departing from the spirit and scope of this invention, whether modifications or substitutions, are also within the protection scope of this invention.

Claims

1. A method for predicting road surface skid resistance performance based on hyperspectral image texture features, characterized in that: Includes the following steps: Step 1: Acquire airborne hyperspectral images of the road surface and crop out the road surface area, and measure the anti-skid performance index of the road surface sample points as training labels; Step 2: Screening spectral feature bands, which involves performing wavelet transform on each band of the hyperspectral image to obtain high-frequency components, calculating Shannon entropy and energy on the high-frequency components, and then using Shannon entropy and energy to construct band scores to screen out spectral feature bands that reflect road surface characteristics. Specifically, for each decomposition layer obtained by wavelet decomposition for each band, the following steps are performed: First, calculate the Shannon entropy and energy on the high-frequency components of the decomposition layer; then, calculate the score corresponding to the decomposition layer; next, perform a weighted summation of the scores of all decomposition layers to obtain the band score; finally, use the band score to filter spectral feature bands. Among them, the scores corresponding to the decomposition layer E represents energy, and F represents Shannon entropy. It is a local minimum; Step 3: Filter texture parameters, that is, calculate the texture parameters of the hyperspectral image corresponding to each spectral feature band, filter out the texture parameters that are most correlated with anti-skid performance in each spectral feature band, and form a texture parameter dataset; Step 4: Construct a road surface skid resistance performance prediction model based on the XGBoost model. The texture parameter dataset is used as the model input, and the model is trained using training labels. A staged grid search method based on coupling constraints is introduced to optimize the hyperparameter combination of the XGBoost model, resulting in the CGS-XGBoost model. This leads to the construction of the road surface skid resistance performance prediction model based on the CGS-XGBoost model. The hyperparameter combination includes: tree depth (max_depth), number of base learners (T), and learning rate. The details are as follows: First, initialize the partial tree depth max_depth, the number of base learners T, and the learning rate using equal-interval step sizes. This leads to the construction of multiple sets of hyperparameter combinations, and the learning rate is defined under a fixed tree depth. There is a coupling relationship with the number of base learners T: , is a constant representing the balance factor between the learning rate and the number of base learners; Secondly, based on the GridSearchCV algorithm, cross-validation is used to pre-search different hyperparameter combinations, calculate the error of each hyperparameter combination, and then select the hyperparameter combination with the smallest error to determine the optimal combination for each tree depth max_depth. value; Next, based on the discrete tree depth max_depth and the corresponding discrete optimal... The value is obtained by using interpolation methods to discretely determine the optimal value. The value extends to the full tree depth range; Then, based on the complete tree depth range and the corresponding optimal Value and learning rate The number of base learners is calculated, and then a set of hyperparameter combinations is constructed. Finally, based on the GridSearchCV algorithm, cross-validation is used to filter the hyperparameter combinations in the hyperparameter combination set to determine the optimal hyperparameter combination; Step 5: Input the texture parameters of the airborne hyperspectral image of the road surface to be tested into the trained road skid resistance performance prediction model to obtain skid resistance performance index.

2. The method according to claim 1, characterized in that: The formulas for the Shannon entropy and the energy are as follows: ； ； ； In the formula, , It is the first in the high-frequency component coefficients The, the There are several values; N is the total number of high-frequency components; It is the normalized probability.

3. The method according to claim 1, characterized in that: In step 3, the process of selecting texture parameters involves first calculating the texture parameters of the hyperspectral image for each spectral feature band using the gray-level co-occurrence matrix, and then selecting the texture parameters that are most correlated with anti-skid performance in each spectral feature band. The texture parameters of the hyperspectral image are a combination or partial combination of mean, variance, homogeneity, contrast, variability, entropy, energy, and correlation.

4. The method according to claim 3, characterized in that: The relationships between mean, variance, homogeneity, contrast, dissimilarity, entropy, energy, correlation, and the gray-level co-occurrence matrix are as follows: ； ； ； ； ； ； ； ； in, Let be the element values ​​corresponding to gray values ​​i and j in the gray-level co-occurrence matrix, and G be the number of gray levels. The mean, For variance, Homogeneity For contrast, For the sake of difference, For entropy, For energy, For correlation, It is the overall average gray value. It is a local minimum; and These are the expected values ​​in the row and column directions of the gray-level co-occurrence matrix, respectively. and It is the standard deviation of the gray-level co-occurrence matrix.

5. The method according to claim 1, characterized in that: The process involves pre-searching different hyperparameter combinations using cross-validation, calculating the error of each hyperparameter combination, and then employing five-fold cross-validation and mean squared error to select the appropriate combination. The minimum combination of hyperparameters is given by the following formula: ； ； In the formula, These are measured values. It is a predicted value. This represents a set of hyperparameter combinations. Let represent the mean squared error of the k-th fold validation, and n represent the total number of samples in the k-th fold validation set. The circular index represents each sample corresponding to the k-th fold.

6. A system based on the method of any one of claims 1-5, characterized in that: Including sequential or interconnected connections: The hyperspectral image processing module is used to acquire airborne hyperspectral images of the road surface and crop out the road surface area, as well as measure the anti-skid performance index of the road surface sample points as training labels; The spectral feature band filtering module is used to filter spectral feature bands. It performs wavelet transform on each band of the hyperspectral image to obtain high-frequency components, calculates Shannon entropy and energy on the high-frequency components, and then uses Shannon entropy and energy to construct band scores to filter out spectral feature bands that reflect road surface characteristics. The texture parameter filtering module is used to filter texture parameters, calculate the texture parameters of the hyperspectral image corresponding to each spectral feature band, and filter out the texture parameters that are most correlated with anti-skid performance in each spectral feature band to form a texture parameter dataset. The model building and training module is used to build a road surface skid resistance performance prediction model based on the XGBoost model. The texture parameter dataset is used as the model input, and the model is trained using training labels. The prediction module is used to input the texture parameters of the airborne hyperspectral image of the road surface to be tested into the trained road skid resistance performance prediction model to obtain skid resistance performance index.

7. A computer device, characterized in that: include: One or more processors; A memory that stores one or more computer programs; The processor invokes a computer program to achieve the following: The steps of the method for predicting road surface skid resistance performance based on hyperspectral image texture features as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that: The computer program is stored and is invoked by the processor to implement: The steps of the method for predicting road surface skid resistance performance based on hyperspectral image texture features as described in any one of claims 1-5.

Images