Intelligent Detection Method and System for Compaction Degree of Rock-filled Subgrade Based on Image Processing

Abstract

The present invention discloses an intelligent detection method and system for the compaction degree of a rock-filled subgrade based on image processing, including: obtaining the corresponding relationship between different gradations, dosages and the compaction degree of crushed stone aggregates through experiments; extracting image indexes matching the compaction degree of the rock-filled subgrade; converting the parameters corresponding to the gradation, dosage and image indexes into vectors, and using a fully connected neural network to assign different weights to each parameter in the vectors to form an n×n matrix; using a convolutional neural network to process the matrix to obtain a detection model for the compaction degree of the rock-filled subgrade; obtaining the gradation, dosage information and surface image information of the to-be-detected rock-filled subgrade, and using the detection model for the compaction degree of the rock-filled subgrade to obtain the detection result of the compaction degree of the rock-filled subgrade. The present invention automatically establishes the potential correlation between the indexes and the compaction degree of the rock-filled subgrade by using the neural network algorithm, and can realize simple, non-destructive, accurate and rapid detection of the compaction degree of the rock-filled subgrade.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent detection of the compaction degree of rock-filled subgrades, and particularly to an intelligent detection method and system for the compaction degree of rock-filled subgrades based on image processing. Background Art

[0002] The statements in this part merely provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] A large amount of crushed stone waste is generated during the construction of mountain roads. In order to rationally utilize these wastes and save transportation distance and cost, using crushed stone aggregates to fill subgrades is becoming the mainstream construction method for mountain subgrades. The compaction degree of subgrades is an important control index for highway subgrade construction. Insufficient compaction degree will cause problems with the unqualified compaction quality of subgrades, and then lead to situations such as subgrade settlement and pavement fracture during the subsequent use of the highway, greatly affecting the service life of expressways. Therefore, it is necessary to strictly control the compaction degree of subgrades during the subgrade construction process.

[0004] Traditional methods for controlling the compaction degree of rock-filled subgrades, such as the water injection method, settlement difference method, nuclear density gauge method, etc., not only have complex measurement processes and long time consumption, but also cause certain damage to subgrades. Such methods detect some sample points in the subgrade, and the test results are difficult to comprehensively and objectively reflect the overall compaction degree of the road.

[0005] Certain developments have been made in the non-destructive detection methods for the compaction degree of rock-filled subgrades. Among them, the representative ones are the resistivity method, the transient Rayleigh wave method, the additional mass method, and the roller-integrated compaction monitoring technology (RICM).

[0006] The prior art has proposed resistivity models for unsaturated pure sandstone and cohesionless sand with a wider range of applications. It is of certain significance to apply the resistivity method to achieve rapid and accurate evaluation of compaction degree within a large range. However, different coarse aggregate mixtures, different water contents of aggregates, and different compaction processes will all affect the accuracy of model prediction. The prior art uses Rayleigh waves to detect the compaction degree of existing highway subgrades. This method has been verified in the overhaul projects of expressways and has certain feasibility. However, the algorithm for inverting the compaction degree of subgrades based on Rayleigh waves is not yet mature, and there is a certain deviation between the inversion result and the actual situation. At the same time, the layout work of survey lines and detectors for the Rayleigh wave detection method is quite cumbersome, time-consuming and laborious. The prior art applies the additional mass non-destructive testing method to airport construction. Through twenty groups of test results, an empirical formula for soil stiffness and dry density is established. The test results show that this method has good effects for the density measurement of super-coarse grained soil. Although the additional mass method has good effects, it still needs to be combined with the water injection method in current applications, and calibration tests need to be carried out when facing different coarse grained soils. The integrated compaction degree detection technology of rollers is a new method for real-time compaction degree detection during the compaction process. After being combined with GPS, this technology shows great potential in the field of refined control of construction quality. Current research shows that there are still some uncertain correlations between the intelligent compaction measurement values (ICMV) obtained by this method and the compaction quality; during the detection process, changes in drum types, machine operation settings, and the instability of the machine during actual operation usually affect the accuracy of ICMV; in addition, the patterns of the stiffness of soil on soft layers and hard layers changing with the exciting force are not the same. The stiffness of the soft layer decreases with the increase of the exciting force, while the hard layer is just the opposite; the vibration amplitude, the stiffness of the compaction layer, and the water content have great influences on the change of ICMV.

[0007] In summary, the traditional non-contact measurement methods have their own limitations, making it difficult to achieve the goal of simple, non-destructive, accurate, and rapid detection of the compaction degree of rock-filled subgrades. Summary of the Invention

[0008] To solve the above problems, the present invention proposes an intelligent detection method and system for the compaction degree of rock-filled subgrades based on image processing. Image processing technology is used to extract surface features related to the aggregate void ratio of the rock-filled subgrade. Taking the aggregate void ratio as the dependent variable, a neural network function is used to establish the relationship between the aggregate surface features and the aggregate void ratio, so that the compaction status of the rock-filled subgrade can be evaluated through a camera.

[0009] In some embodiments, the following technical solutions are adopted:

[0010] An intelligent detection method for the compaction degree of rock-filled subgrades based on image processing, comprising:

[0011] Carry out vibration compaction experiment schemes for the compaction degree of gravel aggregates with multiple different gradations. For each experiment scheme, conduct indoor compaction tests with multiple different dosages; obtain the corresponding relationships between different gradations, dosages, and the compaction degree of gravel aggregates.

[0012] Collect and process the surface images of the compacted gravel aggregates in the experiment, construct an image database for predicting the compaction degree of gravel aggregates, and extract image indexes that match the compaction degree of the rock-filled subgrade.

[0013] Convert the parameters corresponding to the gradation, dosage, and image indexes into vectors, use a fully connected neural network to assign different weights to each parameter in the vector, obtain 1×n vectors with the same number as the number of indexes n, and form an n×n matrix.

[0014] Use a convolutional neural network to process the matrix to obtain a detection model for the compaction degree of the rock-filled subgrade.

[0015] Obtain the gradation, dosage information, and surface image information of the rock-filled subgrade to be measured, and use the detection model for the compaction degree of the rock-filled subgrade to obtain the detection result of the compaction degree of the rock-filled subgrade.

[0016] In some other embodiments, the following technical solution is adopted:

[0017] An intelligent detection system for the compaction degree of a rock-filled subgrade based on image processing, comprising:

[0018] A vibration compaction experiment module for carrying out vibration compaction experiment schemes for the compaction degree of gravel aggregates with multiple different gradations. For each experiment scheme, conduct indoor compaction tests with multiple different dosages; obtain the corresponding relationships between different gradations, dosages, and the compaction degree of gravel aggregates.

[0019] An image processing module for collecting and processing the surface images of the compacted gravel aggregates in the experiment, constructing an image database for predicting the compaction degree of gravel aggregates, and extracting image indexes that match the compaction degree of the rock-filled subgrade.

[0020] A detection model for the compaction degree of the rock-filled subgrade, which is used to convert the parameters corresponding to the gradation, dosage, and image indexes into vectors, use a fully connected neural network to assign different weights to each parameter in the vector, obtain 1×n vectors with the same number as the number of indexes n, and form an n×n matrix; use a convolutional neural network to process the matrix to obtain a detection model for the compaction degree of the rock-filled subgrade.

[0021] A detection module for the compaction degree of the rock-filled subgrade, which is used to obtain the gradation, dosage information, and surface image information of the rock-filled subgrade to be measured, and use the detection model for the compaction degree of the rock-filled subgrade to obtain the detection result of the compaction degree of the rock-filled subgrade.

[0022] Compared with the prior art, the beneficial effects of the present invention are:

[0023] (1) Through indoor tests analyzing the physical factors affecting the compaction degree of rock-filled subgrades, the gradation and dosage are determined as physical indicators closely related to the compaction degree of rock-filled subgrades; by processing the surface images of the compacted rock-filled subgrades, five image indicators reflecting the compaction degree of rock-filled subgrades are determined. Based on the obtained seven indicators, the neural network algorithm is used to automatically establish the potential correlation between the indicators and the compaction degree of rock-filled subgrades, enabling simple, non-destructive, accurate, and rapid detection of the compaction degree of rock-filled subgrades.

[0024] Other features and advantages of additional aspects of the present invention will be partially given in the following description, partially will become apparent from the following description, or will be understood through the practice of this aspect. Brief Description of the Drawings

[0025] Figure 1 It is a physical diagram of the compaction cylinder and the vibrating table in the embodiment of the present invention;

[0026] Figure 2 It is a schematic diagram of the process of processing the image in the embodiment of the present invention;

[0027] Figure 3 It is the search result of the maximum void width in the embodiment of the present invention;

[0028] Figure 4 It is a schematic diagram of the fully connected neural network model in the embodiment of the present invention;

[0029] Figure 5 It is a schematic diagram of the convolutional neural network model in the embodiment of the present invention;

[0030] Figure 6 It is a broken line graph of the void ratio of gravel aggregates with different gradations under different dosages in the embodiment of the present invention;

[0031] Figure 7 It is a broken line graph of the void ratio of gravel aggregates with different dosages under different gradations in the embodiment of the present invention;

[0032] Figure 8 It is a broken line graph of the surface void ratio of gravel aggregates with different gradations under different dosages in the embodiment of the present invention;

[0033] Figure 9 It is a broken line graph of the surface void ratio of gravel aggregates with different dosages under different gradations in the embodiment of the present invention;

[0034] Figure 10 It is a kernel density distribution graph of the void area of the 16th graded gravel aggregate under different dosages in the embodiment of the present invention;

[0035] Figure 11It is the kernel density distribution diagram of the void perimeter of the crushed stone aggregate with the No. 2 gradation in the embodiments of the present invention under different dosages;

[0036] Figure 12 It is the kernel density distribution diagram of the maximum void width of the crushed stone aggregate with the No. 7 gradation in the embodiments of the present invention under different dosages;

[0037] Figure 13 It is the kernel density distribution diagram of the void roundness of the crushed stone aggregate with the No. 9 gradation in the embodiments of the present invention under different dosages;

[0038] Figure 14 It is the result curve of the prediction of the model on the validation set in the embodiments of the present invention. Specific Embodiments

[0039] It should be noted that the following detailed description is exemplary and is intended to provide further illustration of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs.

[0040] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0041] Using image processing techniques to analyze the surface condition of coarse aggregate subgrades and predicting the compaction of rock-filled subgrades based on the void distribution information contained in the image is considered a better option. Currently, digital image processing (DIP) techniques, such as the watershed transform, are used to analyze the characteristics of coarse aggregate mixtures, primarily by analyzing images obtained through X-ray scanning or camera capture. The specific method involves segmenting the aggregates and voids within the coarse aggregate mixture. Based on the aggregate information, appropriate indicators are set to determine the roundness, gradation, movement characteristics, orientation, and structure of the aggregates in the image. Based on the void information, the surface void ratio of the coarse aggregate mixture can be calculated. Image processing techniques can be used to calculate the surface void ratio of crushed stone aggregates. There is a positive correlation between the surface void ratio of an aggregate and its own interstitial ratio. Generally speaking, a higher interstitial ratio of crushed stone aggregates results in a higher surface void ratio. Therefore, using the surface void ratio of crushed stone aggregates to predict the compaction of rock-filled subgrades is theoretically feasible. Cameras can only capture two-dimensional surface information of a rock-fill roadbed, while voids are three-dimensional. Image analysis ignores the depth information of the crushed stone aggregate, making it difficult to predict aggregate interstitial ratio using only surface void ratio information. Given that neural network methods can model the complex relationships between multiple variables, using image processing technology to analyze the surface characteristics of a rock-fill roadbed from multiple angles and utilizing neural network methods to explore the potential relationships between image feature information, physical feature information, and the compaction degree of the rock-fill roadbed is a promising approach to improving prediction accuracy.

[0042] In light of this, the present invention analyzes physical indicators closely related to the compaction of rock-fill roadbeds through indoor testing. Using video cameras, the surface conditions of compacted crushed stone aggregates from the indoor tests were captured. An image database for predicting crushed stone aggregate compaction was constructed. Image processing techniques were then used to process the captured images and extract image indicators closely related to the compaction of the rock-fill roadbed. A neural network model was then used to model the potential relationships between these physical and image indicators and the compaction of the rock-fill roadbed.

[0043] Example 1

[0044] In one or more embodiments, a method for intelligently detecting the compaction degree of a rock-filled roadbed based on image processing is disclosed, which specifically includes the following steps:

[0045] S1: Conduct multiple vibration compaction test schemes for crushed stone aggregate with different gradations. Conduct indoor vibration compaction tests with different dosages for each experimental scheme. Obtain the corresponding relationship between different gradations, dosages and crushed stone aggregate compaction;

[0046] Specifically, the specific implementation process of this step is as follows:

[0047] S1-1: Design vibration compaction schemes for crushed stone aggregates with different gradations according to the uniform gradation design method. Considering that it can be obtained in advance at the construction site, different vibration application amounts are designed for each scheme.

[0048] S1-2: Before vibration compaction, the preparatory work of the process needs to be completed. Use a ruler with a precision of millimeters to measure the inner height of the compaction cylinder at four equidistant positions near the inner wall respectively, and take the average value. Then use a caliper to measure the inner diameter of the compaction cylinder along two mutually perpendicular inner walls of the compaction cylinder, and take the average value. The results obtained are the inner diameter and inner height of the compaction cylinder, and the volume of the compaction cylinder is calculated according to the calculation formula of the compaction cylinder volume. The compaction cylinder and the vibration table are as Figure 1 shown. The specific calculation method is as follows.

[0049]

[0050] In the formula: V represents the compaction degree of the rockfill aggregate; r represents the radius (cm) of the inner wall of the compaction cylinder; h represents the height (cm) of the inner wall of the compaction cylinder.

[0051] S1-3: Weigh according to the gradation of the aggregate, mix aggregates of different specifications in a woven bag, shake and roll it up and down, left and right, so that the coarse aggregates are fully mixed evenly. Put the fully mixed specimen into the compaction cylinder, then place the compaction cylinder on the compaction table, set the vibration frequency to 50 Hz, and compact for 180 s. Use a vernier caliper to measure the height of the aggregate from the edge of the compaction cylinder at four positions along the inner wall at a selected distance and take the average value. The bulk density of the coarse aggregate mixture is calculated according to the following formula:

[0052]

[0053] In the formula: ρ represents the natural bulk density (g / cm3) of the coarse aggregate mixture measured by the compaction method; m represents the total mass (kg) of the specimen; h1 represents the height (cm) between the top of the compacted aggregate and the edge of the compaction cylinder.

[0054] S1-4: The calculation method of VCA is as follows:

[0055]

[0056] In the formula: ρ b represents the synthetic bulk volume density (g / cm3) of the aggregate in the coarse aggregate mixture; n represents the VCA of the coarse aggregate mixture.

[0057] S1-5: The specific calculation method of the compaction degree of the rockfill subgrade is as follows:

[0058]

[0059] Where: V amin represents the void volume of the coarse aggregate mixture when it reaches the standard maximum dry density under laboratory conditions; n min represents the VCA of the coarse aggregate mixture when it reaches the standard maximum dry density under laboratory conditions; V s represents the volume of the aggregate in the coarse aggregate mixture; V a represents the volume of the voids in the coarse aggregate mixture; ρ d represents the actual dry density of the coarse aggregate mixture after being compacted at the construction site; ρ dmax represents the standard maximum dry density of the coarse aggregate mixture obtained from laboratory tests; K represents the degree of compaction of the coarse aggregate mixture.

[0060] S1-6: Determine the degree of compaction after vibration compaction for each test scheme, and establish the matching relationship between gradation, dosage, and the degree of compaction of crushed stone aggregates.

[0061] S2: Collect and process the surface images of the compacted crushed stone aggregates in the experiment, establish an image database for predicting the degree of compaction of crushed stone aggregates, and extract image indexes that match the degree of compaction of the rock-filled subgrade;

[0062] The specific implementation process of this step is as follows:

[0063] S2-1: Use a camera device to collect the surface condition of the compacted crushed stone aggregates in the indoor experiment, and establish an image database for predicting the degree of compaction of crushed stone aggregates. Take pictures of the surface condition of the rock-filled aggregate in the vibrating cylinder after compaction, and use a flash device for supplementary lighting during the photographing process to keep the camera perpendicular to the surface of the rock-filled aggregate.

[0064] S2-2: The specific image processing flow is as Figure 2 shown. First, it is necessary to cut the collected images to distinguish the crushed stone aggregate information from the background information. The specific method is to frame the pixel area occupied by the crushed stone aggregate in the picture, draw the minimum circumscribed rectangle of the corresponding area, and cut the image with this as the boundary. Set the pixel values of the pixel point areas unrelated to the crushed stone aggregate in the cut image to 0. At this time, the remaining area in the image is the area where the aggregate is located. Convert the color image into a grayscale image, and use the Otsu method to perform binary processing on the grayscale image. In the processed picture, the area with a pixel value of 0 represents voids, and the area with a pixel value of 255 represents the crushed stone aggregate information. Based on the image processing, five image indexes are proposed, namely the surface void ratio, the area of each void, the perimeter of each void, the maximum width of each void, and the roundness of each void.

[0065] S2-3: Use the four-connected seed filling algorithm to find the pixel area corresponding to each void, and fill the voids with different colors. At this time, the area of each void in the image can be calculated. The number of pixels in the pixel area corresponding to each void found by the four-connected seed filling algorithm represents the area of this void. However, the number of pixels in the same void area will change with the change of the image resolution. To ensure that the areas of the same voids in images with different resolutions have the same size and are also conducive to the analysis in this paper, the calculation method of the void area is adjusted as follows.

[0066]

[0067] In the formula: S represents the number of pixels occupied by a certain void in the picture; w represents the pixel width of the picture, that is, how many pixels there are in each row of the picture; h represents the pixel length of the picture, that is, how many pixels there are in each column of the picture; S v represents the area used for a certain void in this paper.

[0068] S2-4: After calculating the area of each void in the image, the surface void ratio of the image can be calculated.

[0069]

[0070] In the formula: S a represents the total sum of the void areas in the picture; S s represents the total sum of the aggregate areas in the picture; n sur represents the surface void ratio of the crushed aggregate analyzed in the picture.

[0071] S2-5: Use the contour extraction algorithm to find the contour of each void in the image, and calculate the perimeter of the void according to the contour. To ensure that the same void has the same perimeter at different resolutions, the method for calculating the perimeter in this embodiment is adjusted as follows:

[0072]

[0073] In the formula: C represents the number of pixels of the isocontour of a certain void in the picture; C v represents the perimeter used for a certain void in this paper.

[0074] S2-6: The maximum width of each void is a new index proposed in the present invention to measure the void width. After the contour extraction algorithm finds the contour of the void, iterate all the pixels surrounded by the contour, calculate the maximum inscribed circle that each pixel can draw, and compare the radii of the inscribed circles drawn by each pixel. The specific results of the maximum width search by the algorithm are as Figure 3 shown.

[0075] S2-7: Roundness is a common index for describing the shape of geometric bodies, and this index reflects the shape angle and texture of the aggregate. In this embodiment, roundness is used to characterize the shape of the voids. Generally speaking, voids with a larger roundness are more likely to be in the crushed stone subgrade with a lower degree of compaction. The specific calculation method is as follows.

[0076]

[0077] In the formula: R represents the roundness of the void.

[0078] S3: Convert the parameters corresponding to the gradation, dosage, and image indexes into vectors, and use a fully connected neural network to assign different weights to each parameter in the vector to obtain a 1×n vector with the same number as the number n of indexes, forming an n×n matrix;

[0079] Specifically, each physical index and image index analyzed in steps S1 and S2 that is closely related to the compaction degree of the rock-filled subgrade will correspond to a series of parameters. For example, the cumulative percentage of sieve residue of 10 sieve holes is usually required to represent the crushed stone gradation. Convert the series of parameters corresponding to each index into a 1×m vector. The influence degrees of the m parameters in the vector on the compaction degree of the rock-filled subgrade are not exactly the same. Use a fully connected neural network to analyze the vector, assign different weights to each parameter in the vector, increase the influence weight of the parameters closely related to the compaction degree of the rock-filled subgrade in the vector, and weaken the influence weight of the parameters relatively less related to the compaction degree of the rock-filled subgrade in the vector. When outputting the result, convert the vector into a 1×n vector, where n is consistent with the number of finally adopted indexes. The specific structure of the model is as Figure 4 shown, and the formula is as follows:

[0080] h = ELU(w3ELU(w2ELU(w1x + b1)+b2)+b3) (9)

[0081] In the formula: ELU represents the ELU activation function; x = (x1, x2,..., x m ) T represents the input data of the model; h = (h1, h2,..., h n ) T represents the output data of the model; w1, w2, and w3 represent the model weight parameters; b1, b2, and b3 represent the model bias parameters. Both the input and output data are in the form of vectors. For example, the input data can be the parameter data corresponding to indexes such as mass, gradation, and surface void ratio; the output data is the vector converted based on the input data.

[0082] S4: Use a convolutional neural network to process the matrix to obtain a detection model for the compaction degree of the rock-filled subgrade;

[0083] Specifically, after the processing in the previous step, n indicators can obtain n 1×n vectors, which are combined into an n×n matrix. The influencing degrees of n different indicators on the aggregate void ratio of the coarse aggregate mixture are different. A convolutional neural network is used to process the matrix, aiming to assign different importance to different indicators, amplify the influence of important parameters, and convert the matrix into a single numerical output through multiple convolutions. The model is as Figure 5 shown, and the specific calculation formula is as follows:

[0084] y = θ3 * ELU(θ2 * ELU(θ1 * H + b4) + b5) + b6 (10)

[0085] In the formula: H = (h1, h2,... h7), representing the input data; θ1, θ2, and θ3 represent the weight parameters of the convolutional layer; b4, b5, and b6 represent the bias parameters, and y represents the output result of the model, that is, the detection result of the compaction degree of the rock-filled subgrade.

[0086] S5: Obtain the gradation, dosage information, and surface image information of the rock-filled subgrade to be measured, and use the rock-filled subgrade compaction degree detection model to obtain the rock-filled subgrade compaction degree detection result.

[0087] As a specific example, this embodiment designs 18 gradation schemes of basalt crushed aggregates, and indoor vibration compaction tests are carried out for five different dosages of 2 kg, 4 kg, 6 kg, 8 kg, and 10 kg for each test scheme. The specific gradation information of the crushed aggregates is shown in Table 1. The aggregate void ratios of the crushed aggregates with different gradations under different dosages are shown in Table 2.

[0088] Table 1 Cumulative percentage of sieve residue of basalt coarse aggregate mixtures of different grades

[0089]

[0090] Table 2 Aggregate void ratios of basalt coarse aggregate mixtures with different gradations under different masses

[0091]

[0092] Figure 6 The change of the void ratio of the crushed aggregates with different gradations under different dosages is plotted as a line chart. It can be clearly seen from the chart that the void ratio of the crushed aggregates gradually increases with the increase of the dosage. Figure 7 The change of the void ratio of the crushed aggregates under different dosages with different gradations is plotted as a line chart. It can be seen from the chart that the void ratio of the aggregates changes with different gradations, indicating that the gradation has a certain influence on the void ratio of the aggregates.

[0093] Collect the surface images of the vibrated and compacted crushed aggregates, perform digital image processing, and calculate the surface void ratio, the area, perimeter, maximum width, and roundness of each void of the crushed aggregates in the images. Plot the changes in the surface void ratio of the crushed aggregates with different gradations and different dosages as a line graph as shown in Figure 8 Figure . It can be found that the surface void ratio of the crushed aggregates obtained from the images increases with the increase in dosage, which is basically consistent with the variation relationship between the void ratio and dosage determined by the tests. Plot the changes in the surface void ratio of the crushed aggregates with different dosages and different gradations as a line graph as shown in Figure 9 Figure . The surface void ratio of the aggregates changes with the change in the aggregate gradation, indicating that there is a correlation between the aggregate gradation information and the surface void ratio of the aggregates. Figures 10 - 12 Figures and are the kernel density distribution graphs of the area, perimeter, and maximum width of the crushed aggregates under different dosages respectively. The peak value of the kernel density curve gradually decreases with the increase in dosage. As can be seen from Figure 6 Figure , the VCA of the crushed aggregates continuously increases with the increase in dosage, which is consistent with the change trend of the kernel density peak value. Therefore, it can be shown that there is a consistency between the void area, perimeter, and maximum width distribution of the crushed aggregates under different dosages and the VCA of the crushed aggregates. Figure 13 Figure is the kernel density distribution graph of the void roundness of the crushed aggregates under different dosages. The roundness curve has two peak values. The peak value on the left side of the image gradually decreases with the increase in dosage, and the peak value on the right side of the image gradually increases with the increase in dosage, indicating that as the void ratio of the crushed aggregates increases, the proportion of voids with a larger roundness in the image gradually increases.

[0094] After indoor tests and image analysis, seven indicators closely related to the compaction degree of the rock-filled subgrade can be obtained, including the dosage of the crushed aggregates, the gradation of the crushed aggregates, the surface void ratio of the image, the void area distribution of the image, the void perimeter distribution of the image, the maximum width distribution of the image voids, and the roundness distribution of the image voids.

[0095] A total of 173 indoor tests were carried out. The surface characteristics of the crushed aggregates collected from the 173 indoor tests and the compaction conditions determined by the crushed aggregate experiments were constructed into an image dataset for rapid detection of the compaction degree of the crushed aggregates. The dataset was divided into a training set and a validation set according to 8:2. The MSE was used as the loss function of the neural network model, and 100 rounds of training were carried out. The following indicators were used to evaluate the prediction accuracy of the model. Assume y = y1,...,y n represents the actual results in the real world, represents the predicted value given by the model, and Ω represents the index of the observed samples. The evaluation indicators are defined as follows:

[0096] Root Mean Square Error (RMSE)

[0097]

[0098] Mean Absolute Error (MAE)

[0099]

[0100] Mean Absolute Percentage Error (MAPE)

[0101]

[0102] The prediction results of the model are as Figure 14 shown. The mean absolute percentage error of the model prediction is 3.11%; the mean absolute error is 1.77; the root mean square error is 1.76. It shows that the method of this embodiment has high prediction accuracy.

[0103] Embodiment 2

[0104] In one or more embodiments, an intelligent detection system for the compaction degree of a rock-filled subgrade based on image processing is disclosed, including:

[0105] A vibration compaction experiment module, configured to perform vibration compaction experiment schemes for the compaction degree of crushed stone aggregates with multiple different gradations. For each experiment scheme, multiple indoor compaction tests with different dosages are carried out; the corresponding relationships between different gradations, dosages and the compaction degree of crushed stone aggregates are obtained;

[0106] An image processing module, configured to collect and process the surface images of the crushed stone aggregates after compaction in the experiment, construct an image database for predicting the compaction degree of the crushed stone aggregates, and extract image indexes matching the compaction degree of the rock-filled subgrade;

[0107] A rock-filled subgrade compaction degree detection model, configured to convert the parameters corresponding to the gradation, dosage and image indexes into vectors, assign different weights to each parameter in the vectors using a fully connected neural network, obtain vectors of 1×n with the same number as the number of indexes n, and form an n×n matrix; use a convolutional neural network to process the matrix to obtain a rock-filled subgrade compaction degree detection model;

[0108] A rock-filled subgrade compaction degree detection module, configured to obtain the gradation, dosage information and surface image information of the rock-filled subgrade to be measured, and use the rock-filled subgrade compaction degree detection model to obtain the detection result of the rock-filled subgrade compaction degree.

[0109] It should be noted that the specific implementation manners of the above modules are the same as those in Embodiment 1 and will not be elaborated herein.

[0110] Although the specific implementation manners of the present invention have been described above in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications or deformations that can be made without creative efforts on the basis of the technical solutions of the present invention are still within the protection scope of the present invention.

Claims

1. An intelligent detection method for the compaction degree of rock-filled subgrade based on image processing, characterized in that Including: Conduct vibration compaction experimental schemes for the compaction degree of gravel aggregates with multiple different gradations, and conduct indoor compaction tests with multiple different dosages for each experimental scheme; Obtain the corresponding relationships among different gradations, dosages, and the compaction degree of gravel aggregates; Collect and process the surface images of the compacted gravel aggregates in the experiment, construct an image database for predicting the compaction degree of gravel aggregates, and extract image indicators that match the compaction degree of the rock-fill subgrade; The specific image indicators are: cut the collected images to distinguish the gravel aggregate information from the background information; propose the surface void ratio, the area of each void, the perimeter of each void, the maximum width of each void, and the roundness of each void as image indicators respectively; Among them, the surface porosity Specifically: ; The area of each gap Specifically: ; Perimeter of each gap Specifically: ; The maximum width of each void is specifically: find the contour of the void through the contour extraction algorithm, iterate all the pixels enclosed by the contour, calculate the maximum inscribed circle that each pixel can draw, and select the maximum diameter among all the inscribed circles as the maximum width of the void; Roundness of each gap Specifically: ; In the formula, represents the total area of voids in the image; represents the total area of aggregates in the image; represents the number of pixel points occupied by a certain void in the image, represents the pixel width of the image, represents the pixel length of the image; represents the number of pixel points of the contour line of a certain void in the image; Convert the parameters corresponding to gradation, dosage, and image metrics into vectors, and use a fully connected neural network to assign different weights to each parameter in the vector to obtain a vector with the same number of metrics as the same number of to form a matrix; Use a convolutional neural network to process the matrix to obtain a rock-fill subgrade compaction degree detection model; Obtain the gradation, dosage information, and surface image information of the rock-fill subgrade to be measured, and use the rock-fill subgrade compaction degree detection model to obtain the rock-fill subgrade compaction degree detection result.

2. The intelligent detection method for the compaction degree of a rock-filled subgrade based on image processing according to claim 1, wherein, Conduct indoor compaction tests with multiple different dosages, specifically: Based on the inner diameter and inner height of the compaction cylinder, calculate the volume of the compaction cylinder according to the calculation formula of the compaction cylinder volume; Based on the bulk density of the coarse aggregate mixture, calculate the void content of coarse aggregates (VCA) of the coarse aggregate mixture, and then obtain the compaction degree of the rock-fill subgrade; Measure the compaction degree after vibration compaction for each test to obtain the relationship among gradation, dosage, and the compaction degree of gravel aggregates.

3. The intelligent detection method for the compaction degree of a rock-filled subgrade based on image processing according to claim 2, characterized in that, Based on the bulk density of the coarse aggregate mixture, calculate the void content of coarse aggregates (VCA) of the coarse aggregate mixture, specifically: Among them, represents the natural bulk density of the coarse aggregate mixture determined by the vibrating compaction method, represents the combined bulk density of the aggregates in the coarse aggregate mixture.

4. The intelligent detection method for the compaction degree of a rock-filled subgrade based on image processing according to claim 2, wherein, Obtain the compaction degree of the rock-fill subgrade, specifically: Among them, represents the void content of coarse aggregates VCA of the coarse aggregate mixture, represents the void content of coarse aggregates VCA of the coarse aggregate mixture when it reaches the standard maximum dry density under laboratory conditions.

5. The intelligent detection method for the compaction degree of a rock-filled subgrade based on image processing according to claim 1, characterized in that, Use a fully connected neural network to assign different weights to each parameter in the vector, specifically: Among them, represents the output data of the model; represents the input data of the model; , , and represent the model weight parameters; and represent the bias parameters of the model; represents the ELU activation function.

6. The intelligent detection method for the compaction degree of a rock-filled subgrade based on image processing according to claim 1, characterized in that Obtain a rock-fill subgrade compaction degree detection model, specifically: Among them, , represents the input data; , , and represent the weight parameters of the convolutional layer; and represent the bias parameters, represents the output result of the model, that is, the compaction degree of the rock-filled subgrade.

7. An intelligent detection system for the compaction degree of rock-filled subgrade based on image processing, which is used to execute the intelligent detection method for the compaction degree of rock-filled subgrade based on image processing according to any one of claims 1-6, characterized in that, Including: A vibration compaction experiment module for conducting vibration compaction experimental schemes for the compaction degree of gravel aggregates with multiple different gradations, and conducting indoor compaction tests with multiple different dosages for each experimental scheme; obtaining the corresponding relationships among different gradations, dosages, and the compaction degree of gravel aggregates; An image processing module for collecting and processing the surface images of the compacted gravel aggregates in the experiment, constructing an image database for predicting the compaction degree of gravel aggregates, and extracting image indicators that match the compaction degree of the rock-fill subgrade; The compaction degree detection model for rock-filled subgrade is used to convert the parameters corresponding to gradation, dosage and image indexes into vectors, and a fully connected neural network is used to assign different weights to each parameter in the vector to obtain the same number of vectors as the number of indexes, forming a matrix; a convolutional neural network is used to process the matrix to obtain the compaction degree detection model for rock-filled subgrade; the same number of vectors, forming a matrix; the convolutional neural network is used to process the matrix to obtain the compaction degree detection model for rock-filled subgrade; A rock-fill subgrade compaction degree detection module for obtaining the gradation, dosage information, and surface image information of the rock-fill subgrade to be measured, and using the rock-fill subgrade compaction degree detection model to obtain the rock-fill subgrade compaction degree detection result.

Images