Method for expanding a set of asphalt concrete test piece tomographic images based on three-dimensional coordinate rotation

Abstract

This invention proposes a method for expanding the tomographic image set of asphalt concrete specimens based on three-dimensional coordinate rotation. This method employs a bilinear pixel interpolation algorithm or a coordinate-based linear frame interpolation algorithm to correct the morphological proportions of the scanned specimen. Voxel reconstruction technology is used to achieve three-dimensional voxel modeling of the initial tomographic image set. Based on the inverse operation of three-dimensional voxel reconstruction, multi-directional tomographic scanning of the asphalt concrete specimen is simulated in computer memory. This method can fully exploit the spatial correlation of geometric features of the internal material of the asphalt concrete specimen and effectively avoids the high cost required for multiple physical tomographic scans of the same specimen. The resulting multiple new directional tomographic images can be used for subsequent microstructural mechanics simulations and high-accuracy convolutional neural network training.

Description

Technical Field

[0001] This invention relates to the field of digital processing technology for tomographic images, and in particular to a method for expanding a tomographic image set of asphalt concrete specimens based on three-dimensional coordinate rotation. Background Technology

[0002] A tomographic image set is a sequential collection of multiple homologous two-dimensional images that characterize the internal geometry of an object. It is typically obtained by rotating and processing images along the surface contour of the specimen using computed tomography (CT) or magnetic resonance (MR) equipment. Tomographic images of asphalt concrete contain explicit morphological information such as the geometric features, density distribution, and material boundaries of the heterogeneous materials within the specimen. After digital image processing, these images can be used for numerical simulations and convolutional neural network training for the analysis of microstructural features and the study of mechanical behavior.

[0003] The pixel information of asphalt concrete cross-sectional images is represented by 8-bit codes "0-255". Pixels with a value of "0" appear dark, pixels with a value of "255" appear bright, and pixels with values ​​of "1-254" show a linear distribution of features according to this rule. In the two-dimensional image, dark areas formed by multiple "0" value pixels represent voids, dark areas formed by multiple low-value pixels represent asphalt slurry, and bright areas formed by multiple medium- and high-value pixels represent aggregates.

[0004] The origin O is defined as the center of the tomographic image, the image plane is xOy, and the image normal direction is the z-axis. Voxel reconstruction technology is used to reconstruct two-dimensional pixel information from a tomographic image set. Extended along the normal z-axis to form three-dimensional voxel information The technique of generating three-dimensional digital models based on voxel coordinates.

[0005] To ensure that the tomographic images of asphalt concrete specimens are of uniform size, existing scanning techniques generate tomographic image sets sequentially from top to bottom. However, fixed-angle imaging can only construct the correlation of material geometric features of a single cross-section. Furthermore, asphalt concrete tomographic imaging is affected by insufficient scanning equipment precision, uneven image exposure, and differences in material density within the specimen. Consequently, the explicit geometric information such as the material shape and contour contained in the image is often obscured, resulting in the inability to fully develop and utilize the internal material geometric information contained in the asphalt concrete specimen.

[0006] For example, in deep learning-based prediction of the mechanical behavior of asphalt concrete, it is necessary to first perform mechanical simulations based on two-dimensional tomographic images using numerical simulation techniques. Then, the two-dimensional slice images and their corresponding stress-strain contour maps are used to create a training set to train a deep learning convolutional neural network. However, the limited training set created from the tomographic images obtained by unidirectional sequential scanning contains insufficient effective information to meet the data requirements for training high-accuracy convolutional neural networks. To address this issue, current methods often involve linearly expanding the training set by rotating, cropping, or adding image noise to the two-dimensional images, thereby supplementing the training of the neural network to improve its prediction accuracy, but the effect is minimal. Summary of the Invention

[0007] This invention addresses the shortcomings and deficiencies of existing technologies by proposing a method for expanding the tomographic image set of asphalt concrete specimens. This method uses voxel reconstruction technology to create a three-dimensional voxel model of the initial tomographic image set, then performs voxel spatial coordinate transformation based on three-dimensional coordinate rotation, and finally simulates the multi-angle tomographic scanning process of the asphalt concrete specimen in computer memory. This invention can fully exploit the spatial correlation of the geometric features of the internal material of the specimen, achieving nonlinear expansion of the specimen's tomographic image set, and effectively solving the problem of insufficient training data faced by existing convolutional neural networks using asphalt concrete tomographic images as training sets.

[0008] To achieve the above objectives, the technical solution of the present invention is as follows:

[0009] This invention is a method for expanding the tomographic image set of asphalt concrete specimens based on three-dimensional coordinate rotation. It mainly includes three-dimensional voxel reconstruction, voxel coordinate space transformation and simulated tomographic scanning, and all steps can be completed in computer memory through programming.

[0010] The specific steps of this method are as follows:

[0011] Step 1: Prepare asphalt concrete specimens, determine the scanning direction, and use physical tomography to obtain the initial tomographic image set.

[0012] Step 2: Based on the geometric features of the specimen and the pixel information of the initial fault image set, a coordinate-based linear frame interpolation algorithm or a bilinear pixel interpolation algorithm is used to unify the cross-sectional spacing and the side length of the square pixels in the fault image. The cubic voxel reconstruction technology is used to complete the three-dimensional reconstruction of the asphalt concrete specimen, and the initial three-dimensional Cartesian coordinate system is calibrated according to the planar spatial relationship of the fault image.

[0013] Step 3: Based on the voxel coordinate information of the initial voxel model in the three-dimensional Cartesian coordinate system, formulate a model rotation strategy and use three-dimensional coordinate rotation matrix transformation to realize the spatial rotation of the voxel model.

[0014] Step 4: Based on the inverse operation of 3D voxel reconstruction, simulate tomographic scanning of the 3D voxel model along the new direction in computer memory to obtain a set of tomographic images in the new direction.

[0015] Step 5: Repeat the process “Step 2-Step 4” to obtain a set of tomographic images in multiple new directions.

[0016] Furthermore, in step 2, to ensure that the geometric proportions of the model are consistent with those of the actual specimen, an edge length of l is used. 体素 To achieve 3D reconstruction using cubic voxels, the tomographic image must have a square pixel side length of l. 像素 Equal to the cross-sectional spacing Δl; for possible l in the initial tomographic image set obtained from physical tomography. 像素 The three magnitude relationships with Δl are distinguished and handled as follows:

[0017] 1. When l 像素 =Δl:

[0018] Let l 体素 =l 像素 =Δl, which can be directly modeled.

[0019] 2. When l 像素 When >Δl:

[0020] Let l 体素 =Δl, based on the bilinear pixel interpolation algorithm, to expand the pixel density of each tomographic image, so that l 像素 =Δl, and then use the tomographic image set after pixel interpolation to perform three-dimensional voxel modeling.

[0021] 3. When l 像素 When <Δl:

[0022] Let l 体素 =l 像素 A coordinate-based linear frame interpolation algorithm is used to expand the number of images in the tomographic image set, so that Δl = l 像素 Then, the expanded set of tomographic images is used for three-dimensional voxel modeling.

[0023] Preferably, the asphalt concrete specimens include hexahedrons, cylinders, and specimens with other arbitrary shapes and sizes.

[0024] Preferably, the physical tomographic scanning adopts a dense acquisition method, and the size of the cross-section spacing should be matched as closely as possible to the square pixel side length of the tomographic image.

[0025] Preferably, the tomographic image types include RGB color images, grayscale images, and binary images, and the image resolutions include 1507×914, 1920×1080, 3840×2160, and other size ratio types.

[0026] Preferably, each tomographic image contains all pixel information of the complete specimen on the cross section, and redundant pixel positions are filled with pixels with a value of "0".

[0027] Preferably, the new direction includes any direction other than the initial fault acquisition direction.

[0028] The present invention has the following beneficial effects:

[0029] 1. Unlike existing linear expansion methods for asphalt concrete fault image sets, this invention expands the fault image set by performing multi-directional fault scanning on the same specimen. Fault images in different directions contain new knowledge necessary for training high-accuracy neural networks, which were not developed by the original technology, regarding the spatial correlation of the specimen material's geometric features.

[0030] 2. Using computer programming technology to simulate tomographic scanning can fully explore the spatial correlation of the geometric features of the material inside the specimen, and effectively avoid the high cost required for multiple physical tomographic scans of the same specimen.

[0031] 3. All processes and algorithms can be quickly implemented by editing loop programs on a computer. Attached Figure Description

[0032] The technical solution of the present invention will be further described in detail below with reference to specific embodiments:

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a complete flowchart of the present invention.

[0035] Figure 2 This is a schematic diagram of a tomographic scan according to an embodiment of the present invention.

[0036] Figure 3 This is a schematic diagram illustrating the coordinate system calibration of the tomographic image set and the three-dimensional voxel model according to an embodiment of the present invention.

[0037] Figure 4 This is a schematic diagram of the bilinear pixel interpolation algorithm according to an embodiment of the present invention.

[0038] Figure 5 This is a schematic diagram of a coordinate-based linear frame interpolation algorithm according to an embodiment of the present invention.

[0039] Figure 6 This is a schematic diagram of the spatial rotation of a three-dimensional voxel model according to an embodiment of the present invention.

[0040] Figure 7 This is a schematic diagram of three-dimensional voxel reconstruction and simulated tomography in an embodiment of the present invention. Detailed Implementation

[0041] The technical solution of this patent will be further described below with reference to specific embodiments:

[0042] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] A method for expanding the tomographic image set of asphalt concrete specimens based on three-dimensional coordinate rotation, such as... Figure 1 This includes the following steps:

[0044] Step 1, Tomographic scanning of asphalt concrete specimens: Reference Figure 2 In accordance with the requirements of the "Test Procedures for Asphalt and Asphalt Mixtures in Highway Engineering" and the "Technical Specifications for Construction of Asphalt Pavement on Highways," parameters such as aggregate gradation, void ratio, and asphalt-aggregate ratio were determined. Cylindrical asphalt concrete specimens with a height h = 75 mm and a diameter d = 65 mm were prepared using a rotary compactor. CT tomography was employed to acquire n = 750 tomographic images along the height of the specimen. The cross-sectional spacing Δl = 0.1 mm, and the effective resolution of the tomographic images was 650 × 650, i.e., the side length l of a square pixel was [missing value]. 像素 =d / 650 = 0.1mm.

[0045] Step 2, 3D voxel reconstruction and coordinate system calibration: Reference Figure 3 Based on the voxel reconstruction method, using edge length l 体素 A 0.1 mm cube voxel was used to reconstruct the three-dimensional structure of an asphalt concrete specimen, and a three-dimensional Cartesian coordinate system was established with the specimen center as the origin. It is important to note that, to ensure the geometric proportions of the model are consistent with the actual specimen, this invention uses an edge length of l. 体素 To achieve 3D reconstruction using cubic voxels, the side length of the square pixels in the tomographic image must be l. 像素 It is equal to the cross-sectional spacing Δl. This refers to the l that may appear in the initial tomographic image set obtained from physical tomographic scanning. 像素 Regarding the three possible relationships between Δl and its magnitude, the present invention proposes the following solutions.

[0046] 1. When l 像素 =Δl:

[0047] Let l 体素 =l 像素 =Δl, which can be directly modeled.

[0048] 2. When l 像素 When >Δl:

[0049] Let l 体素 =Δl, based on the bilinear pixel interpolation algorithm, to expand the pixel density of each tomographic image, so that l 像素 =Δl, and then use the pixel-interpolated tomographic image set for 3D voxel modeling, refer to Figure 4 The specific algorithm is as follows:

[0050] Assuming a 2×2 block in the original tomographic image is interpolated to (q+1)×(q+1) pixels, the total number of pixels in this image is k times the number of pixels in the original tomographic image. 2 If the multiple is equal, then the following relationship holds:

[0051]

[0052]

[0053]

[0054] Where, η n Let the pixel value be the grayscale value of the nth slice in the initial slice image set. Let be the grayscale value of the new pixel in the nth slice after pixel interpolation, i be the x-axis pixel coordinate of the corresponding known pixel, j be the y-axis pixel coordinate of the known pixel, u be the x-axis coordinate of the new pixel relative to the four surrounding known pixels, and v be the y-axis coordinate of the new pixel relative to the four surrounding known pixels.

[0055] 3. When l 像素 When <Δl:

[0056] Let l 体素 =l 像素 A coordinate-based linear frame interpolation algorithm is used to expand the number of images in the tomographic image set, so that Δl = l 像素 Then, using the expanded set of tomographic images, three-dimensional voxel modeling was performed, referencing... Figure 5 The specific algorithm is as follows:

[0057] It should be noted that, unlike the time axis calibration of coordinate-based linear frame interpolation algorithms, the algorithm of this invention uses the z-axis coordinate of the initial Cartesian coordinate system for calibration in order to select the reference object for the pixel features of the inserted frame.

[0058] h=Δl×n=Δl aim ×n aim

[0059] Δl aim =l 像素

[0060]

[0061] When the z-axis coordinate of the i-th frame to be inserted is equal to the z-axis coordinate of the j-th frame in the original tomographic image set:

[0062] ζ i (x,y)=φ j (x,y)

[0063] When the z-axis coordinate of the i-th frame to be inserted is located between the j-th and (j+1)-th frames in the original tomographic image set:

[0064]

[0065] Where n is the total number of sections in the initial fault image set, Δl is the section spacing in the initial fault image set, and n aim Δl represents the total number of cross sections in the expanded tomographic image set. aim ζ represents the cross-sectional spacing of the expanded fault image set. i For the i-th tomographic image in the expanded tomographic image set, Let z be the j-th image in the initial tomographic image set. k This represents the z-axis coordinate of the tomographic image with sequence number k.

[0066] Step 3: Develop a model rotation strategy, using a 3D coordinate rotation matrix transformation to achieve voxel model space rotation: For example... Figure 6 As shown, in this embodiment, the coordinate axes are fixed, the model spins around the origin O, the simulated tomographic scanning angle is fixed in the positive z-axis direction, the rotation sequence is set to first rotate around the x-axis and then around the y-axis, the right-hand rule direction is defined as positive when the model spins (or the left-hand rule is defined as positive when the coordinate system spins), the rotation increment Δθ = 20°, and the number of rotations N = (180° / Δθ). 2 =81 times.

[0067] It should be noted that the voxel coordinates obtained by rotating the model by Δθ according to the right-hand rule are consistent with the results of rotating the coordinate system by Δθ according to the left-hand rule. This is for ease of description. Figure 6 A schematic diagram of the coordinate system's rotation was drawn.

[0068] Based on the three-dimensional voxel model constructed from the initial tomographic image set, when the model is fixed, after rotating the coordinate system by β° and β'° respectively in the order of x first and then y according to the left-hand rule, the coordinate transformation of voxel A can be expressed by the following formula. Extending this rule to the voxel coordinates of the entire model, the coordinate transformation results of the voxel points of the entire model can be obtained:

[0069]

[0070]

[0071] Among them, M yx The corresponding spatial coordinate rotation matrix is ​​[x, y, z], where the rotation order is x first, then y. T Let [x', y', z'] be the 3D coordinates of each voxel in the initial 3D voxel model. T To determine the three-dimensional coordinates of the corresponding voxel after spatial rotation.

[0072] Step 4, Simulated Tomographic Scan: Based on the inverse operation of 3D voxel reconstruction, simulated tomographic scans are performed on the model for each rotation with the z-axis as the cross-sectional normal, to obtain a new set of tomographic images, such as... Figure 7 As shown.

[0073] Step 5: Repeat steps 2-4 until the model rotation strategy is completed, obtaining a total of 81 fault images in different directions, thus achieving nonlinear augmentation of the fault image data. Based on the geometric information contained in the fault images in the new directions, mechanical behavior simulations are conducted. The fault images and their simulation results are used as the training set to input into the convolutional neural network, improving the prediction accuracy of the convolutional neural network.

[0074] The above description is merely a description of preferred embodiments of this application and is not intended to limit the scope of this application in any way. Any changes or modifications made by those skilled in the art based on the above-disclosed technical content should be considered as equivalent and valid embodiments and fall within the scope of protection of the technical solution of this application.

Claims

1. A method for expanding the set of fault images of an asphalt concrete test specimen based on three-dimensional coordinate rotation, all steps can be realized by computer development cycle program; characterized in that, The specific steps are as follows: Step 1: Prepare asphalt concrete specimens, determine the scanning direction, and use physical tomography to obtain the initial tomographic image set; Step 2: Based on the geometric features of the specimen and the pixel information of the initial fault image set, a coordinate-based linear frame interpolation algorithm or a bilinear pixel interpolation algorithm is used to unify the cross-sectional spacing and the side length of the square pixels in the fault image. The cubic voxel reconstruction technology is used to complete the three-dimensional reconstruction of the asphalt concrete specimen, and the initial three-dimensional Cartesian coordinate system is calibrated according to the spatial relationship of the fault image in the plane. Step 3: Based on the voxel coordinate information of the initial voxel model in the three-dimensional Cartesian coordinate system, formulate a model rotation strategy and use three-dimensional coordinate rotation matrix transformation to realize the spatial rotation of the voxel model; Step 4: Based on the inverse operation of 3D voxel reconstruction, simulate tomographic scanning of the 3D voxel model along the new direction in computer memory to obtain a set of tomographic images in the new direction; Step 5: Repeat the process "Steps 2-4" to obtain a set of tomographic images in multiple new directions.

2. The method of claim 1, wherein the method is based on three-dimensional coordinate rotation. The asphalt concrete specimens include hexahedrons and cylinders.

3. The method of claim 1, wherein the method is based on three-dimensional coordinate rotation. The physical tomographic scanning uses a dense acquisition method and requires that the spacing between cross sections match the side length of the square pixels in the tomographic image.

4. The method for expanding the tomographic image set of asphalt concrete specimens based on three-dimensional coordinate rotation as described in claim 1, characterized in that, The tomographic image types include RGB color images, grayscale images, and binary images, with image resolutions including 1507×914, 1920×1080, and 3840×2160.

5. The method for expanding the tomographic image set of asphalt concrete specimens based on three-dimensional coordinate rotation as described in claim 1, characterized in that, Each tomographic image contains all pixel information of the complete specimen on that section, with extra pixel positions filled with "0" value pixels.

6. The method for expanding the tomographic image set of asphalt concrete specimens based on three-dimensional coordinate rotation as described in claim 1, characterized in that, The new direction includes any direction other than the initial fault acquisition direction.

7. The method for expanding the tomographic image set of asphalt concrete specimens based on three-dimensional coordinate rotation as described in claim 1, characterized in that, In step 2, to ensure that the geometric proportions of the model are consistent with those of the actual specimen, an edge length of [missing information] is used. l 体素 To achieve 3D reconstruction using cubic voxels, the required square pixel side length of the tomographic image is... l 像素 Spacing between sections Δl Equal; for the initial tomographic images obtained from physical tomography scans that appear l 像素 and Δl The three types of size relationships are distinguished and processed as follows: when l 像素 = Δl hour: make l 体素 = l 像素 = Δl You can directly model it; when l 像素 > Δl hour: make l 体素 = Δl Based on the bilinear pixel interpolation algorithm, the pixel density of each tomographic image is increased, making... l 像素 = Δl Then, the tomographic image set after pixel interpolation is used to perform three-dimensional voxel modeling; when l 像素 < Δl hour: make l 体素 = l 像素 A coordinate-based linear frame interpolation algorithm is used to expand the number of images in the tomographic image set, making... Δl = l 像素 Then, the expanded set of tomographic images is used for three-dimensional voxel modeling.

Images