Asphalt pavement aggregate gradation prediction method and device based on image processing technology

Abstract

This invention provides a method and apparatus for predicting aggregate gradation of asphalt pavement based on image processing technology, comprising: acquiring a developed image of the side surface of an asphalt pavement core sample; rapidly segmenting the developed image of the side surface of the asphalt pavement core sample based on a deep learning algorithm; interactively extracting aggregate particle information from the core sample; and predicting the gradation of the asphalt pavement core sample. The apparatus allows for convenient on-site prediction of asphalt pavement core sample gradation. Its weight and volume are portable by a single person, making it easy to operate and highly adaptable to complex working environments. The PyImageJ interactive module significantly simplifies on-site gradation prediction, enabling rapid, end-to-end image processing and extraction of aggregate particle information without switching between various image processing software. Furthermore, it addresses the problem that purely visual methods cannot measure fine aggregate gradation.

Description

Technical Field

[0001] This invention relates to the field of civil engineering road technology, and in particular to a method and apparatus for predicting aggregate gradation of asphalt pavement based on image processing technology. Background Technology

[0002] Traditional methods for determining aggregate gradation can be divided into laboratory testing methods and field testing methods. Laboratory testing methods include asphalt mixture extraction and X-ray computed tomography (CT). Asphalt mixture extraction separates the asphalt binder from the aggregate in the asphalt mixture to estimate aggregate gradation; this method is typically used for laboratory sampling and cannot be used for field assessment. X-ray CT captures a series of CT images from core samples in the field, which are then processed using digital image processing techniques to reconstruct a 3D numerical model of the asphalt mixture. Aggregate gradation can be estimated based on the virtual asphalt mixture; however, the reconstruction process makes the entire gradation analysis very time-consuming.

[0003] Outdoor testing methods involve capturing digital information using cameras, laser sensors, and other tools. In this case, the analysis of pavement aggregates is achieved through image segmentation. However, pavement images obtained from cameras are limited to two-dimensional images, lacking the ability to determine texture depth and discern fine spatial characteristics of aggregates. 3D laser scanning technology can capture pavement surface texture at high resolution, achieving millimeter-level accuracy. However, laser sensors are expensive; and processing 3D data requires building complex network models, which typically involve determining a large number of hyperparameters that are empirically based and cannot be directly interpreted. Especially for pavement data that the model has never processed before, the accuracy of predictions cannot be guaranteed. Summary of the Invention

[0004] The purpose of this invention is to provide a solution that can accurately, efficiently, and quickly analyze the gradation of asphalt mixtures in the field, addressing the shortcomings of the aforementioned background technology.

[0005] To achieve the above objectives, this invention provides a method for predicting aggregate gradation of asphalt pavement based on image processing technology, comprising the following steps:

[0006] S1, Obtain the unfolded image of the side surface of the asphalt pavement core sample;

[0007] S2, based on deep learning algorithm, quickly segment the unfolded image of the side surface of the asphalt pavement core sample;

[0008] S3, interactive extraction of core sample aggregate particle information;

[0009] S4, predicting the gradation of core samples for asphalt pavement.

[0010] Furthermore, in S2, the asphalt concrete image segmentation network TU-Net++ is used to segment the aggregate particles in the unfolded image of the side surface of the acquired asphalt pavement core sample, and the segmentation yields a grayscale image. The aggregate particles are marked as white, and other components in the mixture are marked as black.

[0011] The asphalt concrete image segmentation network TU-Net++ is based on the U-Net++ model. The U-Net++ model is trained with a large number of asphalt concrete images to obtain the asphalt concrete image segmentation network TU-Net++.

[0012] Furthermore, S3 integrates ImageJ and Python to build a PyImageJ interactive module, which uses PyImageJ to quickly post-process the grayscale image obtained in S2 and extract the feature parameter information of the core sample particles.

[0013] Furthermore, the specific execution flow of PyImageJ in S3 includes:

[0014] S31, Grayscale Image Binarization: Convert a grayscale image into a black and white image after binarization.

[0015] S32, Hole filling: The Fill holes algorithm is used to fill the cavities of aggregate particles in the binarized image;

[0016] S33, Adhesion Separation: The Adjustable watershed algorithm is used to separate the adhesion of aggregate particles;

[0017] S34, extract particle information and export the area and maximum Freret diameter parameters of each particle.

[0018] Furthermore, S4 specifically includes:

[0019] S41, Predicting the mass ratio between different grades of coarse aggregates based on a morphological model:

[0020] For coarse aggregates with a nominal particle size greater than 2.36 mm, the mass ratio between different grades of coarse aggregates is predicted based on the following formula:

[0021] s p =cFeret(p)

[0022] Where Feret(p) is the Feret diameter of particle p, c is the empirical conversion factor, and s p To convert aggregate particle size, as a basis for nominal particle size;

[0023]

[0024] Where q sFor the mass ratio of aggregates in grade S, the A content in the molecule m ρ represents the cross-sectional area of ​​the aggregate particles in grade S. m Let n be the particle density, n be the number of aggregate particles in that grade, and k be an empirical power exponent, encompassing the power relationship between the cross-sectional area and volume of the aggregate particles. The two-dimensional parameter A is then expressed through k. m Converted to three-dimensional volume parameters; the denominator is all particles A on the side surfaces of all core samples. t k and ρ t Multiply and then sum;

[0025] S42, based on the sequence prediction model, predict the mass ratio between each grade of fine aggregate. For fine aggregate with a nominal particle size of less than 2.36 mm, based on the calculated mass ratio between each grade of coarse aggregate, construct and train the sequence prediction model LSTM to predict the mass ratio between each grade of fine aggregate.

[0026] S43, the gradation of asphalt pavement core samples is obtained by the mass ratio of each aggregate.

[0027] Furthermore, different empirical coefficient ranges are determined for different types of pavement; for SMA-13 ​​pavement, c is 0.71 to 0.76, and k is 1.15 to 1.20; for AC-20 pavement, c is 0.76 to 0.86, and k is 1.15 to 1.25; for AC-25 pavement, c is 0.71 to 0.76, and k is 1.15 to 1.20.

[0028] Furthermore, the structure of the sequence prediction model LSTM constructed in S42 includes:

[0029] Input layer: The feature dimension of each sequence point is 1, indicating that the model accepts a single feature value as input;

[0030] LSTM layer: consists of two LSTM layers, each with 64 hidden state dimensions;

[0031] Fully connected layer: The output of the LSTM layer is input to a fully connected layer with six output nodes, corresponding to the mass ratio of the fine aggregate at the six levels that need to be predicted.

[0032] This invention also provides an asphalt pavement aggregate gradation prediction device based on image processing technology, employing the asphalt pavement aggregate gradation prediction method based on image processing technology as described above. The device includes a housing and a housing cover, the housing cover being hinged to the housing. An operation panel is provided on the upper part of the housing, the operation panel having multiple buttons. A light-shielding cover is provided on the operation panel, hinged to the operation panel and capable of opening relative to the operation panel. The space within the housing directly below the light-shielding cover is designated as an image acquisition area, and an acquisition window is provided on one side of the image acquisition area. A linear scan camera is installed in the acquisition window to capture images of different surfaces of the asphalt pavement core sample. A supplementary light is also installed at the location of the acquisition window. A bottom support plate is installed at the bottom of the image acquisition area, and a rotating platform driven by a motor is installed on the bottom support plate. The rotating platform is rotatably connected to the bottom support plate. An industrial control computer is installed inside the box cover. The display screen of the industrial control computer is located inside the box cover. The industrial control computer integrates a system, which includes a PyImageJ interactive module. The operation panel is associated with the industrial control computer.

[0033] The above-described solution of the present invention has the following beneficial effects:

[0034] The image processing technology-based method and device for predicting aggregate gradation of asphalt pavement provided by this invention can conveniently predict the gradation of asphalt pavement core samples on site. The weight and volume of the device can be carried by a single person, making it easy to operate and highly adaptable to complex working environments.

[0035] The PyImageJ interactive module significantly simplifies the operation for users to perform gradation prediction on-site. It enables rapid, end-to-end image processing and extraction of aggregate particle information without switching between various image processing software, further improving the speed and convenience of on-site operation.

[0036] Based on TU-Net++, it provides accurate image segmentation results. After the image is sent to the system, it is calculated by a trained dedicated neural network, resulting in high segmentation accuracy. The morphological model predicts the three-dimensional volume between coarse aggregates of each grade from the two-dimensional feature information of the particles and calculates the mass ratio. It can obtain the three-dimensional volume of the particles without the need to use laser instruments to obtain the depth information of the particles, thus simplifying the device components. The sequence prediction model can predict the mass ratio of fine aggregates using the mass ratio of coarse aggregates, and finally obtain complete gradation data, filling the gap that pure vision methods cannot measure the gradation of fine aggregates.

[0037] Other beneficial effects of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0038] Figure 1This is a flowchart of the method steps of the present invention;

[0039] Figure 2 This is a schematic diagram illustrating the operating principle of TU-Net++ of the present invention;

[0040] Figure 3 This is a schematic diagram of the PyImageJ interactive module of the present invention;

[0041] Figure 4 This is a schematic diagram of the device structure of the present invention.

[0042] [Explanation of Labels in the Attached Image]

[0043] 1-Box body; 2-Box cover; 3-Operating panel; 4-Light shield; 5-Asphalt pavement core sample; 6-Linear scan camera; 7-Supplemental light; 8-Bottom support plate; 9-Rotating platform; 10-Industrial control computer. Detailed Implementation

[0044] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0045] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0046] It should also be noted that the illustrations provided in the following embodiments are merely schematic representations of the basic concept of this disclosure. The illustrations only show components relevant to this disclosure and are not drawn according to the actual number, shape, and size of components in implementation. In actual implementation, the type, quantity, and proportion of each component can be arbitrarily changed, and the component layout may be more complex. Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0047] Embodiments of the present invention provide a method for predicting aggregate gradation of asphalt pavement based on image processing technology, comprising the following steps:

[0048] S1, acquire the unfolded image of the side surface of the asphalt pavement core sample.

[0049] In this embodiment, the development images of the side surface of asphalt pavement core samples are obtained through on-site sampling to efficiently and quickly analyze the aggregate gradation of asphalt pavement. Specifically, the acquisition of the development images of the side surface of the asphalt pavement core samples is accomplished using a miniaturized portable device.

[0050] S2, based on deep learning algorithms, performs rapid segmentation of the unfolded image of the side surface of the asphalt pavement core sample.

[0051] In this embodiment, the asphalt concrete image segmentation network TU-Net++ is used to segment the aggregate particles in the unfolded image of the side surface of the acquired asphalt pavement core sample, obtaining a grayscale image. The aggregate particles are marked in white, and other components in the mixture are marked in black, such as... Figure 2 As shown.

[0052] It should be noted that the asphalt concrete image segmentation network TU-Net++ is built and trained based on the U-Net++ model. The U-Net++ model, due to its dense short-connection structure and deep supervision, enhances image segmentation performance and is widely used for segmenting medical cell images. Since medical cell images and asphalt concrete images (distinguishing aggregate particles from other components) are similar, this embodiment trains the U-Net++ model with a large number of asphalt concrete images to obtain the asphalt concrete image segmentation network TU-Net++. TU-Net++ is then used to segment the aggregate particles in the acquired core sample side surface unfolded image, resulting in the grayscale image shown in the figure.

[0053] S3, interactive extraction of core sample aggregate particle information.

[0054] With the continuous updates to Python software libraries over the years, they can meet increasingly diverse image processing needs. Meanwhile, the Java-based open-source software platform ImageJ is also constantly evolving. To experience the best of both Python and ImageJ simultaneously and avoid switching between different interfaces, a PyImageJ interactive module was built in this step, fundamentally integrating the ImageJ and Python software ecosystems. PyImageJ strongly supports data interoperability between Python and ImageJ, enabling users to create workflows that incorporate elements of both Python and ImageJ. Its principle diagram is shown below. Figure 3 As shown.

[0055] In this embodiment, the grayscale image obtained in S2 is quickly post-processed using PyImageJ, and the feature parameter information of the core sample particles is extracted. The specific operation flow is as follows:

[0056] S31, Grayscale Image Binarization: The grayscale image is binarized and converted into a black and white image to facilitate the algorithm's identification and determination of the aggregate's outline, thereby extracting the particle's feature information.

[0057] S32, Hole filling: The Fill holes algorithm of PyImageJ is used to fill the holes of aggregate particles in the binarized image.

[0058] It should be noted that in the binarized image, some aggregate particles contain cracks and voids, which are mostly caused by mud or uneven cutting on the particle surface. The binarization process treats this part as a non-particle component, which will affect the extraction of particle feature information, especially when extracting particle area information. Therefore, it is necessary to fill the voids.

[0059] S33, Adhesion Segmentation: The adjustable watershed algorithm of PyImageJ is used to segment the adhesion of aggregate particles.

[0060] It should be noted that interlocking contact exists between aggregate particles in asphalt concrete, forming localized adhesion, which significantly impacts image-based statistical analysis of aggregate particles. Adhesive particles merge into one, resulting in deviations in particle size and shape, especially between two coarse particles. To avoid the algorithm treating multiple particles as a single particle when extracting particle information, it is necessary to separate the adhered particles. However, binarization and filling operations do not reduce adhesion; therefore, particle adhesion segmentation is required.

[0061] S34. After image post-processing, use PyImageJ to extract particle information.

[0062] PyImageJ can export various particle feature information. In this embodiment, only the area and maximum Feret diameter of each particle need to be exported for subsequent gradation calculations.

[0063] S4, predicting the gradation of core samples for asphalt pavement.

[0064] In this step, based on the extracted particle feature information, the gradation of asphalt pavement core samples is predicted using PyImageJ. Since the TU-Net++ model cannot accurately segment finer particles in the image, the prediction of the mix proportions is as follows: First, a morphological model is proposed to predict the mass ratio between coarse particles with a nominal particle size greater than 2.36 mm. Then, a sequence prediction model is constructed to predict the mass ratio between fine aggregates with a nominal particle size less than 2.36 mm. Finally, complete gradation data, i.e., the gradation data of all particles from 0.075 mm to the largest size, can be obtained through simple conversion.

[0065] This step specifically includes the following sub-steps:

[0066] S41, Predicting the mass ratio between different grades of coarse aggregates based on a morphological model:

[0067] For coarse aggregates with a nominal particle size greater than 2.36 mm, the mass ratio between different grades of coarse aggregates is predicted based on the following formula:

[0068] s p =cFeret(p)

[0069] Where Feret(p) is the Feret diameter of particle p, c is the empirical conversion factor, and s p This value is used as the basis for converting aggregate particle size;

[0070]

[0071] Where q s This refers to the mass ratio of aggregates in grade S, specifically, when S takes the range of 2.36mm-4.75mm, 4.75mm-9.5mm, ..., 19mm-26.5mm, 26.5mm-31.5mm, the percentage of all aggregate particles in grade S to the total mass, and the A content in the molecule. m ρ is the cross-sectional area of ​​the aggregate particles in this grade. m Let n be the particle density, n be the number of aggregate particles in that grade, and k be an empirical power exponent, encompassing the power relationship between the cross-sectional area and volume of the aggregate particles. The two-dimensional parameter A is then expressed through k. m Converted to three-dimensional volume parameters; the denominator is all particles A on the side surfaces of all core samples. t k and ρt Multiply and then sum.

[0072] It should be noted that different empirical coefficient ranges are determined for different types of pavement. For SMA-13 ​​pavement, c is 0.71 to 0.76, and k is 1.15 to 1.20; for AC-20 pavement, c is 0.76 to 0.86, and k is 1.15 to 1.25; for AC-25 pavement, c is 0.71 to 0.76, and k is 1.15 to 1.20.

[0073] S42, based on the sequence prediction model, predicts the mass ratio between each grade of fine aggregate.

[0074] The sequence prediction model can learn data patterns from a large amount of sequence data, thus predicting missing data based on partial data. In this embodiment, for fine aggregates with a nominal particle size of less than 2.36 mm, a sequence prediction model LSTM is constructed and trained based on the calculated mass ratio between each grade of coarse aggregate to predict the mass ratio between each grade of fine aggregate. Finally, the mass ratio of each grade of aggregate is obtained to obtain the asphalt pavement core sample gradation, that is, to obtain the asphalt pavement aggregate gradation.

[0075] The structure of the constructed sequence prediction model LSTM includes:

[0076] Input layer: The feature dimension of each sequence point is 1, indicating that the model accepts a single feature value as input;

[0077] LSTM layer: consists of two LSTM layers, each with 64 hidden state dimensions;

[0078] Fully connected layer: The output of the LSTM layer is input to a fully connected layer, which has six output nodes, corresponding to the mass ratio of the fine aggregate at the six levels that need to be predicted.

[0079] Therefore, in this embodiment, the sequence prediction model LSTM can reasonably and reliably predict the mass ratio of fine aggregates in six grades.

[0080] Field tests using the method provided in this embodiment showed that for AC-20, AC-25, and SMA-13 ​​pavement material types, the predicted aggregate gradation was consistent with the gradation curves measured by the traditional extraction method. The predicted curves did not exceed the upper and lower limits of the gradation, proving the reliability and practicality of the method.

[0081] Based on the same inventive concept, this embodiment also provides an asphalt pavement aggregate gradation prediction device based on image processing technology, such as... Figure 4As shown, the device includes a housing 1 and a cover 2. The cover 2 is hinged to the housing 1 and can be opened relative to the housing 1. An operation panel 3 is also located on the upper part of the housing 1. The operation panel 3 has multiple buttons for power on / off, parameter setting, start / pause, and other operations. A light-shielding cover 4 is attached to the operation panel 3, also hinged to it, allowing the light-shielding cover 4 to be opened relative to the operation panel 3. This allows the asphalt pavement core sample 5 to be placed inside the housing 1, and then the light-shielding cover 4 to be closed to block light, facilitating accurate image acquisition of the asphalt pavement core sample 5.

[0082] The box 1 directly below the light-shielding cover 4 is the image acquisition area. It is isolated from and light-shielded from other parts of the box 1 by a partition. An acquisition window is set on one side of the image acquisition area, and a line scan camera 6 is installed in the acquisition window to acquire images of the side surface of the placed asphalt pavement core sample 5. At the same time, a supplementary light 7 is also set at the location of the acquisition window to control the lighting conditions of the line scan camera 6 during acquisition.

[0083] Since the aforementioned method requires the acquisition of images of all side surfaces of the asphalt pavement core sample 5, a bottom support plate 8 is provided at the bottom of the image acquisition area in this embodiment. A rotating platform 9 driven by a motor is provided on the bottom support plate 8. The rotating platform 9 is rotatably connected to the bottom support plate 8 and can rotate slowly under the drive of the motor (with reducer), so that each surface of the asphalt pavement core sample 5 placed on the rotating platform 9 will pass through the acquisition window and be imaged by the line scan camera 6.

[0084] In this embodiment, an industrial control computer 10 is embedded inside the cover 2. The display screen of the industrial control computer 10 is located inside the cover 2 so that the display screen can be easily viewed after the cover 2 is opened. The industrial control computer 10 integrates a system, including the aforementioned PyImageJ interactive module. The operation panel 3 is also associated with the industrial control computer 10, and the system can be operated directly through the operation panel 3.

[0085] Therefore, the asphalt pavement aggregate gradation prediction device based on image processing technology provided in this embodiment has good portability and can easily predict the gradation of asphalt pavement core samples on site, thereby directly predicting the aggregate gradation of asphalt pavement. The weight and volume of the device can be carried by a single person, making it easy to operate and highly adaptable to complex working environments.

[0086] Meanwhile, this embodiment integrates the PyImageJ interactive module within the system, which significantly simplifies the operation for users when performing gradation prediction on-site. It enables rapid, end-to-end image processing and extraction of aggregate particle information without switching between various image processing software, further improving the speed and convenience of on-site operation.

[0087] In this system, TU-Net++ provides accurate image segmentation. After the image is sent to the system, it is processed by a trained dedicated neural network, resulting in high segmentation accuracy. The morphological model predicts the three-dimensional volume between coarse aggregates of each grade from the two-dimensional feature information of the particles and calculates the mass ratio. It can obtain the three-dimensional volume of particles without the need for laser instruments to obtain the depth information of the particles, thus simplifying the device components. The sequence prediction model can predict the mass ratio of fine aggregates using the mass ratio of coarse aggregates, and finally obtain complete gradation data, filling the gap that pure visual methods cannot measure the gradation of fine aggregates.

[0088] In summary, the complete gradation prediction process of this solution is simple and convenient, with low learning time costs. Operators can quickly master the entire operation process in a short period of time, making it user-friendly and competitive in the market.

[0089] Based on the same inventive concept, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for predicting the aggregate gradation of asphalt pavement based on image processing technology.

[0090] This computer-readable medium includes, but is not limited to, any type of disk (including floppy disks, hard disks, optical disks, CD-ROMs, and magneto-optical disks), ROM, RAM, EPROM (Erasable Programmable Read-Only Memory), EEPROM, flash memory, magnetic cards, or optical cards. In other words, this computer-readable medium includes any medium by which a device (such as the aforementioned industrial control computer) stores or transmits information in a readable form.

[0091] Based on the same inventive concept, this embodiment also provides an industrial control computer, including at least one processor and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which, when executed, enables the at least one processor to perform the aforementioned image processing-based asphalt pavement aggregate gradation prediction method.

[0092] The computer-readable storage medium and industrial control computer provided in this embodiment have the same inventive concept and the same beneficial effects as the aforementioned method, and will not be described again here.

[0093] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0094] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for predicting aggregate gradation in asphalt pavement based on image processing technology, characterized in that, Includes the following steps: S1, Obtain the unfolded image of the side surface of the asphalt pavement core sample; S2, based on deep learning algorithm, quickly segment the unfolded image of the side surface of the asphalt pavement core sample; S3, interactive extraction of core sample aggregate particle information; S4, predicting the gradation of asphalt pavement core samples; specifically including: S41, Predicting the mass ratio between different grades of coarse aggregates based on a morphological model: For coarse aggregates with a nominal particle size greater than 2.36 mm, the mass ratio between different grades of coarse aggregates is predicted based on the following formula: yes p =cFerret(p) Where Feret(p) is the Feret diameter of particle p, c is the empirical conversion factor, and s p To convert aggregate particle size, as a basis for nominal particle size; Where q s For the mass ratio of aggregates in grade S, the A content in the molecule m ρ represents the cross-sectional area of ​​the aggregate particles in grade S. m Let n be the particle density, n be the number of aggregate particles in that grade, and k be an empirical power exponent, encompassing the power relationship between the cross-sectional area and volume of the aggregate particles. The two-dimensional parameter A is then expressed through k. m Converted to three-dimensional volume parameters; the denominator is all particles A on the side surfaces of all core samples. t k and ρ t Multiply and then sum; S42, based on the sequence prediction model, predict the mass ratio between each grade of fine aggregate. For fine aggregate with a nominal particle size of less than 2.36 mm, based on the calculated mass ratio between each grade of coarse aggregate, construct and train the sequence prediction model LSTM to predict the mass ratio between each grade of fine aggregate. S43, the gradation of asphalt pavement core samples is obtained by the mass ratio of each aggregate.

2. The asphalt pavement aggregate gradation prediction method based on image processing technology according to claim 1, characterized in that, In S2, the asphalt concrete image segmentation network TU-Net++ is used to segment the aggregate particles in the unfolded image of the side surface of the acquired asphalt pavement core sample. The segmentation results in a grayscale image, with the aggregate particles marked as white and other components in the mixture marked as black. The asphalt concrete image segmentation network TU-Net++ is based on the U-Net++ model. The U-Net++ model is trained with a large number of asphalt concrete images to obtain the asphalt concrete image segmentation network TU-Net++.

3. The asphalt pavement aggregate gradation prediction method based on image processing technology according to claim 2, characterized in that, S3 integrates ImageJ and Python to build a PyImageJ interactive module, which uses PyImageJ to quickly post-process the grayscale image obtained in S2 and extract the feature parameter information of the core sample particles.

4. The asphalt pavement aggregate gradation prediction method based on image processing technology according to claim 3, characterized in that, The specific execution flow of PyImageJ in S3 includes: S31, Grayscale Image Binarization: Convert a grayscale image into a black and white image after binarization. S32, Hole filling: The Fill holes algorithm is used to fill the cavities of aggregate particles in the binarized image; S33, Adhesion Separation: The Adjustable watershed algorithm is used to separate the adhesion of aggregate particles; S34, extract particle information and export the area and maximum Freret diameter parameters of each particle.

5. The asphalt pavement aggregate gradation prediction method based on image processing technology according to claim 1, characterized in that, For different types of pavement, different ranges of empirical coefficients are determined; for SMA-13 ​​pavement, c is 0.71 to 0.76, and k is 1.15 to 1.20; for AC-20 pavement, c is 0.76 to 0.86, and k is 1.15 to 1.25; for AC-25 pavement, c is 0.71 to 0.76, and k is 1.15 to 1.

20.

6. The method for predicting aggregate gradation of asphalt pavement based on image processing technology according to claim 1, characterized in that, The structure of the sequence prediction model LSTM constructed in S42 includes: Input layer: The feature dimension of each sequence point is 1, indicating that the model accepts a single feature value as input; LSTM layer: consists of two LSTM layers, each with 64 hidden state dimensions; Fully connected layer: The output of the LSTM layer is input to a fully connected layer with six output nodes, corresponding to the mass ratio of the fine aggregate at the six levels that need to be predicted.

7. An asphalt pavement aggregate gradation prediction device based on image processing technology, employing the asphalt pavement aggregate gradation prediction method based on image processing technology as described in any one of claims 3-6, characterized in that, The device includes a housing and a lid, the lid being hinged to the housing. An operation panel with multiple buttons is located on the upper part of the housing. A light-shielding cover, also hinged to the operation panel, is installed on the operation panel and can be opened relative to it. The space directly below the light-shielding cover is designated as an image acquisition area. An acquisition window is located on one side of the image acquisition area, housing a line scan camera for image acquisition of the side surface of an asphalt pavement core sample. A supplementary light is also located at the acquisition window. A bottom support plate is located at the bottom of the image acquisition area, and a rotating platform driven by a motor is mounted on the bottom support plate, rotatably connected to it. An industrial control computer is housed inside the lid, with its display screen located inside the lid. The industrial control computer integrates a system including a PyImageJ interactive module, and the operation panel is associated with the industrial control computer.

Images