Method and system for evaluating porosity of grouting material consolidated body based on image recognition

Abstract

The invention provides a grouting material consolidated body porosity evaluation method and system based on image recognition, and belongs to the field of grouting material performance evaluation.The grouting material consolidated body porosity evaluation method comprises the steps that a grouting material consolidated body test block is selected, the test block is preprocessed to prepare a test sample, and a three-dimensional model is established; determining slice parameters, performing slice operation on the three-dimensional model, generating a plurality of two-dimensional slice images, and preprocessing the two-dimensional slice images; performing resolution processing on the preprocessed two-dimensional slice image, and constructing a slice image data set; inputting images in a training set of the constructed slice image data set into a convolutional neural network model for training to obtain a trained convolutional neural network model; processing a grouting material consolidation body to be detected to obtain a preprocessed two-dimensional image; and inputting the preprocessed two-dimensional image into the trained convolutional neural network model to obtain an identification result of the pores, and obtaining the porosity of the to-be-detected grouting material consolidated body based on the identification result of the pores.

Description

Technical Field

[0001] The present invention belongs to the field of performance evaluation of grouting materials, and particularly relates to a method and system for evaluating the porosity of a consolidated body of a grouting material based on image recognition. Background Art

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

[0003] In the field of civil engineering, such as foundation reinforcement, tunnel engineering, bridge repair, etc., the grouting technology is a commonly used construction method, which plays a crucial role in improving the stability of engineering structures, enhancing the bearing capacity, and preventing leakage. As the core element of the grouting technology, the working performance of the grouting material directly affects the quality and durability of the project.

[0004] The working performance of the grouting material covers multiple aspects, and the porosity of the consolidated body is a key performance index among them. A lower porosity usually means that the material is denser, and it may have higher strength and better impermeability; while a higher porosity may affect the mechanical properties and durability of the material. Researchers need to accurately evaluate the porosity of grouting materials under different formulations and different preparation processes in order to deeply understand the relationship between the performance and structure of the materials, so as to develop grouting materials with better performance.

[0005] Existing conventional methods for detecting the porosity of materials include the density method, microphotography method, ultrasonic detection, etc. Most of the density method, microphotography method, etc. are destructive detections. In the density method, it is necessary to measure the volume and mass of the material to calculate the porosity, which often requires operations such as cutting or crushing the material, damaging the grouting material, and making it impossible to conduct subsequent performance research or practical application on the same batch of materials. Although non-destructive detection methods such as ultrasonic detection avoid damaging the material, they also have problems such as low detection efficiency and slow speed, and the detection accuracy of tiny pores inside the material is limited. Since the wavelength of ultrasonic waves is relatively long, the reflection and scattering of tiny pores are weak, making it difficult to accurately detect the existence of these pores, and thus there is a certain difficulty in accurately evaluating the porosity.

[0006] In addition, the existing non-destructive detection of the porosity of materials is mostly indirect measurement, which indirectly calculates the porosity by measuring the physical properties related to the pores, rather than directly measuring the absolute value of the porosity. Although there is a method in the existing technology that uses images to characterize the pore-fracture structure, such as "Research on the Pore-Fracture Structure Characterization and Seepage Characteristics of Geotechnical Materials Based on Digital Images", however, this solution is not aimed at grouting materials. Secondly, this technology focuses on the characterization of the microscopic pore structure of geotechnical materials and the prediction of the permeability coefficient, and does not propose a direct prediction method for the porosity of materials, and cannot achieve accurate evaluation of the porosity. Summary of the Invention

[0007] To overcome the deficiencies of the above-mentioned prior art, the present invention provides a method for evaluating the porosity of a grouting material solidified body based on image recognition, which acquires the internal structure information of a grouting material solidified body sample and reconstructs a three-dimensional structure, performs slice analysis on the three-dimensional structure, identifies pores in the three-dimensional structure slices through machine learning image recognition, and estimates the porosity of the structure in a three-dimensional space to achieve the purpose of efficient and non-destructive inspection of the porosity of the grouting material. This method can directly calculate and evaluate the microscopic pores of the material with higher accuracy.

[0008] To achieve the above object, one or more embodiments of the present invention provide the following technical solutions:

[0009] In a first aspect, a method for evaluating the porosity of a grouting material solidified body based on image recognition is disclosed, including:

[0010] Select a test block of the grouting material solidified body, preprocess the test block to prepare a test sample and establish a three-dimensional model;

[0011] Determine the slice parameters, perform slice operations on the three-dimensional model to generate multiple two-dimensional slice images, and preprocess the two-dimensional slice images;

[0012] Construct a network structure, perform resolution processing on the preprocessed two-dimensional slice images, and construct a slice image data set;

[0013] Input the images in the training set of the constructed slice image data set into a convolutional neural network model for training to obtain a trained convolutional neural network model;

[0014] Process the grouting material solidified body to be measured to obtain a preprocessed two-dimensional image;

[0015] Input the preprocessed two-dimensional image into the trained convolutional neural network model to obtain the recognition result of pores, and obtain the porosity of the grouting material solidified body to be measured based on the recognition result of pores.

[0016] As a further technical solution, after preprocessing the test block to prepare a test sample, the sample is rotated and scanned at multiple angles, and two-dimensional projection data of the sample is collected from different angles.

[0017] As a further technical solution, the process of establishing a three-dimensional model is as follows:

[0018] Assume an initial three-dimensional image;

[0019] Project the current initial three-dimensional image onto each acquisition angle to obtain the corresponding two-dimensional projection image;

[0020] Compare the calculated two-dimensional projection image with the actually acquired two-dimensional projection data, and calculate the difference between the two;

[0021] Correct the three-dimensional image according to the difference, and adjust the value of each voxel in the three-dimensional image to minimize the projection difference;

[0022] Repeat the above steps until a certain convergence condition is met to obtain a three-dimensional model.

[0023] As a further technical solution, determining the slice parameters includes determining the slice thickness and multiple slice directions.

[0024] As a further technical solution, preprocess the two-dimensional slice image, including:

[0025] Adjust the grayscale of the two-dimensional slice image to highlight the difference between the pores and the surrounding materials;

[0026] Enhance the brightness or darkness of the pore area to form a more distinct contrast with the surrounding materials;

[0027] Adopt a denoising algorithm to effectively remove noise, improve the signal-to-noise ratio of the image, and reduce interference factors.

[0028] As a further technical solution, construct a network structure to perform resolution processing on the preprocessed two-dimensional slice image. Among them, the network structure includes a feature extraction layer, a sub-pixel convolution layer, and an upsampling layer;

[0029] Input the preprocessed two-dimensional slice image into the feature extraction layer, which is usually composed of several convolutional layers and is used to extract the features of the image;

[0030] The image after feature extraction enters the sub-pixel convolution layer, which learns the sub-pixel displacement information of the pixels through convolution operations and outputs a high-resolution feature map;

[0031] Convert the high-resolution feature map into the final high-resolution two-dimensional slice image through the upsampling layer.

[0032] As a further technical solution, the convolutional neural network model uses a deformable convolution kernel to improve the convolutional neural network. The deformable convolution kernel consists of a traditional convolution kernel and an offset learning network;

[0033] Calculate the output of the deformable convolution kernel: Assume that the input image is I, the traditional convolution kernel is K, and the offset is Δp. Then the output of the deformable convolution kernel is where p is the pixel position of the output image, q is the sampling point position of the traditional convolution kernel, and R is the sampling area of the traditional convolution kernel;

[0034] During the calculation process, first, the sampling point positions of the traditional convolution kernel are adjusted according to the offset amounts output by the offset learning network, and then the adjusted convolution kernel is convolved with the input image to obtain the output of the deformable convolution kernel.

[0035] As a further technical solution, the porosity of the consolidation body of the grouting material to be measured is obtained based on the recognition result of pores. The specific process is as follows:

[0036] Calculate the proportion of pore area: For each two-dimensional slice image, count the number of pixels in the pore region, and then divide it by the total number of pixels in the image to obtain the proportion of pore area.

[0037] Calculate the porosity by the volume integration method: Multiply the proportion of pore area of each slice image by the slice thickness to obtain the pore volume corresponding to the slice.

[0038] Then sum up the pore volumes of all slices, and divide by the volume of the whole sample to obtain the porosity of the three-dimensional structure.

[0039] In the second aspect, an evaluation system for the porosity of the consolidation body of the grouting material based on image recognition is disclosed, including:

[0040] A three-dimensional model establishment module, configured to: select a test block of the consolidation body of the grouting material, preprocess the test block to prepare a test sample and establish a three-dimensional model;

[0041] A two-dimensional slice image generation module, configured to: determine the slice parameters, perform slicing operations on the three-dimensional model to generate multiple two-dimensional slice images, and preprocess the two-dimensional slice images;

[0042] A slice image data set construction module, configured to: construct a network structure, perform resolution processing on the preprocessed two-dimensional slice images, and construct a slice image data set;

[0043] A convolutional neural network model training module, configured to: input the images in the training set of the constructed slice image data set into the convolutional neural network model for training to obtain a trained convolutional neural network model;

[0044] A porosity evaluation module for the consolidation body of the grouting material to be measured, configured to: process the consolidation body of the grouting material to be measured to obtain a preprocessed two-dimensional image;

[0045] Input the preprocessed two-dimensional image into the trained convolutional neural network model to obtain the recognition result of pores, and obtain the porosity of the consolidation body of the grouting material to be measured based on the recognition result of pores.

[0046] The above one or more technical solutions have the following beneficial effects:

[0047] The porosity evaluation method of grouting materials based on image recognition proposed by the technical solution of the present invention uses non-destructive testing technology, does not require the destruction of specimens, and can save a large amount of specimen resources. It reduces the testing cost and avoids the additional costs caused by specimen destruction. In practical applications, the porosity of grouting materials can be evaluated without affecting the integrity of the engineering structure, providing an economical and efficient solution for engineering quality control. At the same time, it also reduces the impact on the environment, meeting the requirements of sustainable development.

[0048] The porosity evaluation method proposed by the technical solution of the present invention scans the specimens with high recognition accuracy, can provide more accurate images, and provides a reliable basis for machine learning. The super-resolution algorithm efficient sub-pixel convolutional neural network (ESPCN) is used to process the pixels of the images, and high-resolution images can be obtained, which can better present the detailed information of the pores. The deformable convolutional kernel is also used for neural network training, which can make the convolutional neural network more flexibly adapt to pores of different shapes and sizes. Especially for tiny pores with irregular shapes and different sizes, it can better capture their features, thus significantly improving the accuracy of pore recognition. This high-precision porosity evaluation method is crucial for ensuring the quality of grouting materials and the safety of the project, and can provide accurate parameter basis for engineering design and construction.

[0049] Through scanning and deep learning image recognition technology, the detection speed of the porosity of grouting materials in the technical solution of the present invention is greatly improved. The ESPCN super-resolution algorithm has an efficient sub-pixel convolutional structure, can quickly process a large number of two-dimensional slice images, improves the speed and efficiency of data processing, and saves time for constructing a high-quality slice image dataset. The deformable convolutional kernel improves the adaptability and performance of the convolutional neural network by additionally introducing a branch network to learn the offset amount without significantly increasing the computational complexity, making the model training more efficient. For the improved training model structure of the deformable convolutional kernel, please refer to the attached Figure 2 shown.

[0050] Compared with traditional detection methods, it greatly saves time and improves the detection efficiency. It is crucial for large-scale engineering construction and quality inspection, can obtain a large amount of accurate porosity data in a short time, and provide timely basis for engineering decision-making. Quick detection can also reduce the time cost of engineering construction, speed up the project progress, and improve the overall efficiency of the project.

[0051] Advantages of additional aspects of the present invention will be partly given in the following description, partly will become obvious from the following description, or be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings forming a part of the specification are used to provide a further understanding of the present invention. The schematic embodiments and descriptions thereof of the present invention are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0053] Figure 1 It is a flowchart of the method according to the embodiment of the present invention.

[0054] Figure 2 It is a structural diagram of the training model after the deformable convolution kernel is improved. Detailed implementation manners

[0055] It should be noted that the following detailed descriptions are all exemplary and are intended to provide further explanations of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0056] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary embodiments according to the present invention.

[0057] In the case of no conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0058] Glossary of terms:

[0059] Efficient Sub-Pixel Convolutional Neural Network, abbreviated as ESPCN, Chinese: Efficient Sub-Pixel Convolutional Neural Network.

[0060] Embodiment 1

[0061] Refer to the attached Figure 1 As shown, this embodiment discloses a method for evaluating the porosity of the grouting material consolidation body based on image recognition, including:

[0062] Step 1: Select a representative consolidation body specimen of the grouting material, preprocess the sample to prepare a PXCT test specimen, and establish a three-dimensional model.

[0063] Specifically, select a part from the grouting material consolidation body that can represent the overall material characteristics as the sample. Comprehensive judgment can be made according to factors such as different parts of the consolidation body and different construction conditions to ensure that the selected sample is typical. Cut and process the selected sample to make its size and shape meet the requirements of the PXCT equipment to prepare for accurate data acquisition.

[0064] Use the PXCT equipment to emit X-rays, rotate and scan the sample at multiple angles, and collect the X-ray projection data of the sample from different angles to comprehensively obtain the internal information of the sample.

[0065] The three-dimensional reconstruction is performed on the collected two-dimensional projection data using an iterative reconstruction algorithm. A three-dimensional model that can display the internal microstructure is obtained, providing a detailed structural basis for subsequent analysis.

[0066] The iterative reconstruction algorithm is an algorithm for reconstructing three-dimensional images by continuously iteratively optimizing. The specific implementation method is as follows:

[0067] Initialize the three-dimensional image: First, assume an initial three-dimensional image, which can usually be a uniform low-resolution image.

[0068] Projection calculation: Project the current three-dimensional image onto each acquisition angle to obtain the corresponding two-dimensional projection image.

[0069] Comparison and correction: Compare the calculated two-dimensional projection image with the actually collected two-dimensional projection data, and calculate the difference between the two. Then, correct the three-dimensional image according to the difference. Usually, an optimization algorithm (such as the gradient descent method, etc.) is used to adjust the value of each voxel in the three-dimensional image to minimize the projection difference.

[0070] Repeat iteration: Repeat the above steps until a certain convergence condition is met, such as the projection difference is less than a certain threshold or a certain number of iterations is reached.

[0071] Algorithm formula and description: Let the collected two-dimensional projection data be P i (i represents different acquisition angles), and the current three-dimensional image be V. Then the projection calculation can be expressed as P i = Projection(V), where Projection represents the projection function. The correction process can be expressed as where α is the step size parameter, is the gradient of the objective function E. The objective function is usually defined as the difference between the actual projection data and the calculated projection data, that is

[0072] In this step, the PXCT has high scanning accuracy and can obtain the fine structure inside the consolidated body of the grouting material. The three-dimensional model established after PXCT scanning can display the connection relationship and spatial layout between the multi-layer structures inside the consolidated body, laying a good foundation for subsequent pore identification. It can solve the problems of low accuracy of traditional testing methods and unclear exploration of internal structures.

[0073] Step 2: Determine the slice parameters, perform slicing operations on the three-dimensional model to generate multiple two-dimensional slice images, and preprocess the two-dimensional slice images.

[0074] Specifically, determine the slice thickness and multiple slice directions. Determine the slice thickness according to specific analysis requirements and the size of the sample. A smaller thickness needs to be selected to ensure that the internal microstructure of the sample can be detailedly presented. Select multiple different angles for the slice directions, such as along the three coordinate axes of X, Y, and Z and directions at a certain angle to the coordinate axes. While ensuring the calculation efficiency, present the internal microstructure of the sample as detailedly as possible. Select multiple slice directions to observe the sample from different perspectives, avoiding missing important pore information due to the limitation of a single direction. Ensure in-depth analysis of the internal structure of the sample from different angles. Each generated two-dimensional slice image must show the fine structure at a determined position inside the sample.

[0075] Perform slicing operations on the three-dimensional model. According to the determined slice parameters, slice the model obtained by three-dimensional reconstruction. Professional image processing software or algorithms can be used to implement the slicing operation. For example, for a model represented by a three-dimensional array, extract two-dimensional slice images according to specific slice directions and thicknesses, and generate multiple two-dimensional slice images to observe the internal structure of the sample from different angles.

[0076] Preprocess the two-dimensional slice images, mainly including:

[0077] Gray-scale adjustment: Adjust the gray scale of the two-dimensional slice images to highlight the differences between pores and surrounding materials. This can be achieved by adjusting the gray-scale histogram of the image. For example, adjust the gray-scale values of the pore regions to higher or lower values to make the difference between their gray-scale values and those of the surrounding materials larger. Make the pores more obvious for subsequent pore identification.

[0078] Algorithm formula and explanation: Let the gray-scale value of the original image be I(x, y), and the adjusted gray-scale value be I′(x, y). Gray-scale adjustment can be achieved through the linear transformation formula I′(x, y) = aI(x, y) + b, where a and b are adjustment parameters and need to be selected according to specific situations.

[0079] Contrast enhancement: Enhance the brightness or darkness of the pore regions to form a more distinct contrast with the surrounding materials. Contrast enhancement algorithms such as histogram equalization and local contrast enhancement can be used. For example, histogram equalization adjusts the gray-scale histogram of the image to make the gray-scale distribution of the image more uniform, thereby enhancing the contrast. Make the two-dimensional image slices easier to identify and analyze the pore regions.

[0080] Algorithm formula and explanation: For histogram equalization, let the gray-scale histogram of the original image be H(I), and the cumulative distribution function be Then the equalized gray-scale value is I′ = round((C(I) - C min ) / (C max - C min))), where C min and C max are the minimum and maximum values of the cumulative distribution function respectively, and L is the total number of gray levels.

[0081] Denoising: Appropriate denoising algorithms are adopted to effectively remove noise, improve the signal-to-noise ratio of the image, and reduce interference factors. Different denoising algorithms can be selected according to the characteristics of the noise and the properties of the image. The algorithms that can be used include: mean filtering, median filtering, Gaussian filtering, etc. Mean filtering replaces the value of each pixel point in the image with the average value of its neighboring pixel points; median filtering replaces the value of each pixel point in the image with the median value of its neighboring pixel points; Gaussian filtering realizes denoising by performing Gaussian convolution on the image.

[0082] Algorithm formula and explanation: For mean filtering, assume that the neighborhood of a pixel point (x, y) in the image is N(x, y), then the filtered pixel value is where |N(x, y)| is the total number of neighboring pixel points.

[0083] Step 3: Collect a large number of preprocessed two-dimensional slice images, use the super-resolution algorithm efficient sub-pixel convolutional neural network for resolution processing, and construct a slice image dataset.

[0084] Specifically, using the super-resolution algorithm efficient sub-pixel convolutional neural network for resolution processing includes:

[0085] Construct an ESPCN network structure. ESPCN mainly consists of a feature extraction layer, a sub-pixel convolutional layer, and an upsampling layer.

[0086] Input the collected preprocessed two-dimensional slice images into the feature extraction layer, which is usually composed of several convolutional layers and is used to extract the features of the image.

[0087] Algorithm formula and explanation: Assume that the input of the first convolutional layer is an image with a size of H×W×C in where H and W are the height and width of the image respectively, and C in is the number of input channels. The output of this convolutional layer is H×W×C out where C out is the number of output channels, and the convolution operation can be expressed as where X is the input image, W is the convolution kernel, and Y is the output image.

[0088] The image after feature extraction enters the sub-pixel convolutional layer, which learns the sub-pixel displacement information of the pixels through convolution operations.

[0089] It should be noted that compared with traditional convolutional neural networks, the efficient sub-pixel convolutional neural network: in terms of the upsampling method, it uses a sub-pixel convolutional layer and PixelShuffle operation instead of the traditional interpolation method, which can better restore details; it has higher computational efficiency, with a small amount of sub-pixel convolution calculations, which can reduce resource consumption; it has stronger feature representation ability and can use local features to learn complex mappings to restore high-frequency details; it has better integration ability and can be easily integrated into existing structures with good scalability; the network structure design is more efficient, achieving better super-resolution results with fewer layers and parameters, reducing complexity and training difficulty and improving generalization ability.

[0090] Algorithm formula and description: Assume that the input of the sub-pixel convolutional layer is a feature map with a size of H×W×C in and the output is a high-resolution feature map of rH×rW×C out , where r is the magnification factor. The output of the sub-pixel convolutional layer can be expressed as I SR = pixelshuffle(W sp * F(I LR ) + b sp ), where I SR is the low-resolution sliced image, pixelshuffle is the sub-pixel arrangement operation that rearranges the pixels in the low-resolution feature map into the form of a high-resolution image. W sp and b sp are the weights and biases of the sub-pixel convolutional layer respectively.

[0091] Finally, the high-resolution feature map is converted into the final high-resolution two-dimensional sliced image through the upsampling layer, enabling the two-dimensional sliced image to better present the detailed information of the pores.

[0092] Construct a sliced image dataset using the high-resolution image, specifically including:

[0093] Image sorting and annotation: Sort and classify the high-resolution two-dimensional sliced images processed by ESPCN. According to different pore characteristics, such as pore size, shape, distribution density, etc., annotate the images.

[0094] Data augmentation: In order to increase the diversity of the dataset and improve the generalization ability of the model, data augmentation operations can be performed on the training set. Common data augmentation methods include random rotation, flipping, scaling, cropping, adding noise, etc.

[0095] Finally, a sliced image dataset is successfully constructed.

[0096] Step 4: Use the convolutional neural network (CNN) algorithm to train a deep learning model, improve the deformable convolutional kernel, carry out pore identification, and calculate the porosity of the test sample at the three-dimensional level.

[0097] Specifically, a model is trained using the Convolutional Neural Network (CNN) algorithm. The specific steps are as follows:

[0098] Construct a convolutional neural network structure: Classical convolutional neural network architectures such as VGGNet and ResNet can be adopted and appropriately adjusted according to the task of pore identification. The network usually includes convolutional layers, pooling layers, fully connected layers, etc.

[0099] Input the images in the training set of the constructed slice image dataset into the convolutional neural network for training: During the training process, the weights and biases of the network are continuously adjusted through the backpropagation algorithm, making the output of the network gradually approach the true pore annotations. The core of the backpropagation algorithm is to calculate the gradient of the loss function with respect to the network parameters and update the parameters according to the gradient descent method. Assuming the loss function is L and the network parameters are θ, the parameter update formula is where α is the learning rate.

[0100] Algorithm formula and explanation: The training process of the convolutional neural network involves a large number of matrix operations and optimization algorithms. Taking a simple two-layer convolutional neural network as an example, let the input image be X, the first-layer convolutional kernel be W 1 , and the bias be b 1 . Let the activation function be f, then the output of the first layer is Z 1 = f(W 1 * X + b 1 ), where * represents the convolution operation. The output of the second layer can be calculated similarly. The loss function can adopt the cross-entropy loss function or the mean squared error loss function, etc. For example, for a binary classification problem, the cross-entropy loss function can be expressed as where N is the total number of samples, y i is the true label, is the label predicted by the model.

[0101] Use the validation set to monitor the training process. When the loss on the validation set no longer decreases, stop the training and save the network model at this time.

[0102] In view of the shape complexity of the pore images of the grouting material solidified body, to improve the accuracy of pore identification, a deformable convolutional kernel is used to improve the convolutional neural network. The specific steps are as follows:

[0103] Introduce a deformable convolutional kernel: On the basis of the traditional convolutional neural network, a deformable convolutional kernel is introduced. The deformable convolutional kernel consists of a traditional convolutional kernel and an offset learning network.

[0104] Calculate the output of the deformable convolutional kernel: Assuming the input image is I, the traditional convolutional kernel is K, and the offset is Δp, then the output of the deformable convolutional kernel is Where p is the pixel position of the output image, q is the sampling point position of the traditional convolution kernel, and R is the sampling area of the traditional convolution kernel. During the calculation process, first, the sampling point position of the traditional convolution kernel is adjusted according to the offset amount output by the offset learning network, and then the adjusted convolution kernel is convolved with the input image to obtain the output of the deformable convolution kernel.

[0105] Training the offset learning network: The offset learning network is trained simultaneously with the convolutional neural network. Its input is the same as that of the convolutional neural network, and the output is the offset amount of each sampling point. During the training process, the parameters of the convolutional neural network and the parameters of the offset learning network are updated simultaneously through the backpropagation algorithm, so that the deformable convolution kernel can better adapt to pores of different shapes and sizes. The loss function can adopt the same loss function as the traditional convolutional neural network, such as the cross-entropy loss function or the mean square error loss function, etc. Considering the particularity of the deformable convolution kernel, some regularization terms can be added to constrain the range and variation of the offset amount. For example, L1 or L2 regularization can be used to limit the magnitude of the offset amount to prevent the network from becoming unstable due to excessive offset.

[0106] Carry out pore identification, and the specific steps include:

[0107] The two-dimensional image preprocessed in Step 2 is processed using the efficient sub-pixel convolutional neural network (ESPCN) of the super-resolution algorithm, and then input into the trained convolutional neural network. The network output is the identification result of the pores, that is, the probability that each pixel point in the image belongs to the pores. Finally, the pore region is automatically identified and labeled to quickly and accurately extract the pore characteristic parameters.

[0108] According to the set threshold, the pixel points with a probability greater than the threshold are labeled as pores to obtain a binary image of the pore region, where the white region represents pores and the black region represents non-pores.

[0109] When calculating the porosity of the sample subsequently, the number of pixels occupied by the pores is divided by the total number of pixels of the slice to obtain the porosity of the slice. Calculate the porosity of each slice, and finally, the porosity of the entire sample can be obtained using the volume integration method.

[0110] Calculate the porosity of the test sample: Calculate the proportion of the pore area in each slice image, and then use the volume integration method to calculate the porosity of the three-dimensional structure. The specific steps include:

[0111] Calculate the proportion of the pore area: For each two-dimensional slice image, the identification result is the probability that each pixel point belongs to the pores. According to the identification result, the number of pixels in the pore region is counted, and then divided by the total number of pixels in the image to obtain the proportion of the pore area.

[0112] Calculation of porosity by volume integration method: Multiply the proportion of pore area in each slice image by the slice thickness to obtain the pore volume corresponding to the slice. Then sum the pore volumes of all slices and divide by the volume of the entire sample to obtain the porosity of the three-dimensional structure.

[0113] Algorithm formula and description: Let the proportion of pore area in the i-th slice image be p i , the slice thickness be t, and the total volume of the sample be V. Then the porosity of the three-dimensional structure where n is the total number of slices.

[0114] In this embodiment, during the training process of the neural network, an additional branch network is introduced to learn the offset of the convolution kernel. Based on the traditional convolution kernel, the sampling point positions of each convolution kernel are allowed to be adaptively adjusted according to the learned offset. This can make the convolution kernel more flexible in adapting to objects of different shapes and sizes in the image. For targets such as tiny pores with irregular shapes and different sizes, the deformable convolution kernel can better capture their features.

[0115] The two-dimensional slice images after preprocessing such as gray-scale adjustment, contrast enhancement, and denoising also need to be processed using the super-resolution algorithm Efficient Sub-Pixel Convolutional Neural Network (ESPCN).

[0116] A more specific Example 1: A method for evaluating the porosity of the consolidated body of grouting material based on image recognition, including:

[0117] Step 1: Sample selection and preprocessing, PXCT scanning and three-dimensional model establishment.

[0118] Sample selection and preprocessing include:

[0119] Select samples from the consolidated body of grouting material in a certain water conservancy project. Cut the selected samples into cubes with a side length of about 2 cm to meet the detection size requirements of the PXCT equipment.

[0120] PXCT Scanning and Three-Dimensional Reconstruction

[0121] Use the PXCT equipment to scan the samples, set the X-ray emission intensity at a moderate level, and the scanning resolution at 0.1 mm. Perform multi-angle rotation scanning on each sample, with a rotation angle interval of 15°, and a total of 24 angles of X-ray projection data are collected.

[0122] Three-dimensional reconstruction is performed using an iterative reconstruction algorithm. The three-dimensional image is initialized as a uniform low-resolution image with voxel values all being 120 (gray value range is 0 - 255). In the projection calculation, a projection function based on ray tracing is adopted. In the comparison and correction process, the gradient descent method is used with the step size parameter set to 0.01. After 120 iterations, the projection difference is less than the set threshold of 0.05, and a three-dimensional reconstruction model is obtained.

[0123] Step 2: Slicing of the three-dimensional model and image preprocessing.

[0124] Determination and operation of slicing parameters, including:

[0125] Determine the slice thickness to be 0.4 mm, and select the slice directions as along the X, Y, and Z coordinate axes and the directions at 45° angles to the coordinate axes. Slicing operation is performed on the three-dimensional reconstruction model according to these parameters, and a total of 28 two-dimensional slice images are generated.

[0126] Preprocessing of two-dimensional slice images, including:

[0127] Gray level adjustment: Gray level adjustment is performed using a linear transformation formula, with parameters a = 1.5 and b = -20 set. The gray level values of the pore regions are adjusted to higher values to make the difference in gray level values between them and the surrounding materials more obvious.

[0128] Contrast enhancement: The histogram equalization algorithm is used. After calculating the gray level histogram of the original image, it is equalized according to the formula, enhancing the contrast between the pore regions and the surrounding materials.

[0129] Denosing: The mean filtering algorithm is adopted. For each pixel part in the image, the value of the pixel is replaced with the average value of its 3×3 neighborhood pixel points, effectively removing the noise in the image.

[0130] Step 3: Collect two-dimensional slice images, process them with a high-resolution algorithm, and construct a slice image dataset.

[0131] Super-resolution processing: 600 two-dimensional slice images with known pore characteristics are collected. These images come from previous research projects on similar grouting materials and partial preliminary test results of this experiment. An ESPCN network structure is constructed. The feature extraction layer consists of 3 convolutional layers. The input of the first convolutional layer is an image with a size of 64×64×1 (height and width are 64 pixels, and the number of input channels is 1), and the output is a feature map of 64×64×32. The input of the sub-pixel convolutional layer is a feature map of 64×64×32, and the output is a high-resolution feature map of 128×128×32 (magnification factor is 2).

[0132] The collected low-resolution two-dimensional slice images are input into the ESPCN network for processing. In the feature extraction layer, the convolution kernel size is 3×3, and feature extraction is performed according to the convolution formula. The sub-pixel convolution layer learns the sub-pixel displacement information of pixels through special convolution operations and outputs a high-resolution feature map according to the formula. Finally, the high-resolution feature map is converted into the final high-resolution two-dimensional slice image through the upsampling layer.

[0133] Dataset construction, including:

[0134] The high-resolution two-dimensional slice images processed by ESPCN are sorted and classified. They are divided into two categories: large pores and small pores according to the pore size, and labeled. Data augmentation operations are performed on the training set, including random rotation by 90°, 180°, and 270°, horizontal and vertical flipping, and adding a small amount of Gaussian noise (mean is 0, standard deviation is 0.01). Finally, a dataset containing 480 training images, 120 validation images, and 120 test images is constructed.

[0135] Step 4: Training the deep learning model with the Convolutional Neural Network (CNN) algorithm and improving the deformable convolution kernel, pore identification, and calculating the porosity of the test sample.

[0136] Model training, the specific process is as follows:

[0137] Construct a convolutional neural network structure based on ResNet-18, make appropriate adjustments to it, remove the last few fully connected layers, and add a convolutional layer for feature extraction. The images in the training set of the constructed slice image dataset are input into the convolutional neural network for training. The cross-entropy loss function is adopted, and the learning rate is set to 0.001. During the training process, the weights and biases of the network are continuously adjusted through the backpropagation algorithm. After 200 epochs of training, when the loss on the validation set no longer decreases, stop training and save the network model.

[0138] Improve the convolutional neural network with a deformable convolution kernel. Introduce a deformable convolution kernel, which consists of a traditional 3×3 convolution kernel and an offset learning network. The offset learning network is trained simultaneously with the convolutional neural network. The input is the same as that of the convolutional neural network, and the output is the offset amount of each sampling point. During the training process, the cross-entropy loss function is adopted, and an L2 regularization term (weight is 0.001) is added to constrain the range and variation of the offset amount. The parameters of the convolutional neural network and the offset learning network are updated simultaneously through the backpropagation algorithm.

[0139] Pore identification and porosity calculation

[0140] Input the pre - processed second - stage image into the trained convolutional neural network. The network output is the recognition result of pores, that is, the probability that each pixel in the image belongs to pores. Set the threshold to 0.5, label the pixels with a probability greater than 0.5 as pores, and obtain the binary image of the pore region.

[0141] Calculate the proportion of pore area in each slice image. For each two - dimensional slice image, count the number of pixels in the pore region, and then divide it by the total number of pixels in the image. Use the volume integration method to calculate the porosity of the three - dimensional structure. After calculation, the porosity of the three - dimensional structure of the grouting material solidified body is 3.8%.

[0142] A more specific Example 2: A method for evaluating the porosity of a grouting material solidified body based on image recognition, including:

[0143] Step 1: Sample selection and pre - processing, PXCT scanning and three - dimensional model establishment

[0144] The sample is selected from the grouting material solidified body of a certain building foundation grouting project. Cut the selected sample into a cube with a side length of about 3 cm to meet the detection size requirements of the PXCT equipment.

[0145] PXCT Scanning and Three - Dimensional Reconstruction

[0146] Use the PXCT equipment to scan the sample, set the X - ray emission intensity at a relatively high level, and the scanning resolution at 0.08 mm. Conduct multi - angle rotational scanning for each sample, with a rotational angle interval of 10°, and collect X - ray projection data at 36 angles in total.

[0147] Perform three - dimensional reconstruction using the iterative reconstruction algorithm. Initialize the three - dimensional image as a uniform low - resolution image with voxel values all being 100 (gray value range is 0 - 255). In the simulation calculation, use the projection function based on ray tracing. In the comparison and correction process, use the gradient descent method with a step - size parameter set to 0.008. After 150 iterations, the projection difference is less than the set threshold of 0.03, and a three - dimensional reconstruction model is obtained.

[0148] Step 2: Three - Dimensional Model Slicing and Image Pre - processing

[0149] Determination and Operation of Slicing Parameters

[0150] Determine the slice thickness to be 0.3 mm, and select the slice directions along the X, Y, and Z coordinate axes and the directions at 30° angles to the coordinate axes. Perform slicing operations on the three - dimensional reconstruction model according to these parameters, and a total of 36 two - dimensional slice images are generated.

[0151] Pre - processing of Two - Dimensional Slice Images

[0152] Gray-scale adjustment: Linear transformation formula is used for gray-scale adjustment, with parameters a = 2 and b = -40 set. The gray-scale values of the pore regions are adjusted to higher values to make the difference in gray-scale values from the surrounding materials more obvious.

[0153] Contrast enhancement: The histogram equalization algorithm is used. After calculating the gray-scale histogram of the original image, it is equalized according to the formula, enhancing the contrast between the pore regions and the surrounding materials.

[0154] Denoising: The median filtering algorithm is adopted. For each pixel point in the image, its value is replaced with the median value of the 5×5 neighborhood pixel points, effectively removing the noise in the image.

[0155] Step 3: Collect two-dimensional slice images, process them with a high-resolution algorithm, and construct a slice image dataset

[0156] Super-resolution processing

[0157] 800 two-dimensional slice images with known pore characteristics were collected. These images are from different research projects on building foundation grouting materials and some preliminary test results of this experiment. An ESPCN network structure was constructed. The feature extraction layer consists of 4 convolutional layers. The input of the first convolutional layer is an image with a size of 80×80×1 (height and width are 80 pixels, and the number of input channels is 1), and the output is a feature map of 80×80×48. The input of the sub-pixel convolutional layer is a feature map of 80×80×48, and the output is a high-resolution feature map of 160×160×48 (magnification factor is 2).

[0158] The collected low-resolution two-dimensional slice images are input into the ESPCN network for processing. In the feature extraction layer, the convolutional kernel size is 3×3, and feature extraction is performed according to the convolutional formula. The sub-pixel convolutional layer learns the sub-pixel displacement information of the pixels through special convolutional operations and outputs a high-resolution feature map according to the formula. Finally, the high-resolution feature map is converted into the final high-resolution two-dimensional slice image through the upsampling layer.

[0159] Dataset construction

[0160] The high-resolution two-dimensional slice images processed by the ESPCN are sorted, classified, and labeled. Data augmentation operations are performed on the training set, including random rotations of 90°, 180°, and 270°, horizontal and vertical flips, and adding a small amount of Gaussian noise (mean is 0, standard deviation is 0.02). Finally, a dataset containing 640 training images, 160 validation images, and 160 test images is constructed.

[0161] Step 4: Train a deep learning model with the Convolutional Neural Network (CNN) algorithm and improve it with deformable convolutional kernels, identify pores, and calculate the porosity of the test samples.

[0162] Model training

[0163] A convolutional neural network structure based on VGGNet was constructed and appropriately adjusted. The last few fully connected layers were removed, and two convolutional layers were added for feature extraction. The images in the training set of the constructed slice image dataset were input into the convolutional neural network for training. The cross-entropy loss function was adopted, and the learning rate was set to 0.0008. During the training process, the weights and biases of the network were continuously adjusted through the backpropagation algorithm. After 300 epochs of training, when the loss on the validation set no longer decreased, the training was stopped and the network model was saved.

[0164] The convolutional neural network was improved by using deformable convolutional kernels. Deformable convolutional kernels were introduced, which consist of a traditional 3×3 convolutional kernel and an offset learning network. The offset learning network was trained simultaneously with the convolutional neural network. The input was the same as that of the convolutional neural network, and the output was the offset of each sampling point. During the training process, the cross-entropy loss function was adopted, and an L2 regularization term (weight 0.001) was added to constrain the range and variation of the offset. The parameters of the convolutional neural network and the offset learning network were updated simultaneously through the backpropagation algorithm.

[0165] Pore identification and porosity calculation

[0166] The 36 images preprocessed in the second step were input into the trained convolutional neural network. The network output was the identification result of pores, that is, the probability of each pixel point in the image belonging to pores. A threshold of 0.55 was set, and the pixel points with a probability greater than 0.55 were labeled as pores, obtaining a binary image of the pore region.

[0167] The proportion of the pore area in each slice image was calculated. For each two-dimensional slice image, the number of pixels in the pore region was counted and then divided by the total number of pixels in the image. The three-dimensional structure porosity was calculated using the volume integration method. After calculation, the porosity of the three-dimensional structure of the grouting material consolidation body was 4.2%.

[0168] Example two

[0169] The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the above method are implemented.

[0170] Example three

[0171] The purpose of this embodiment is to provide a computer-readable storage medium.

[0172] A computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the steps of the above method are executed.

[0173] Example 4

[0174] The purpose of this embodiment is to provide a grouting material solidified body porosity evaluation system based on image recognition, including:

[0175] A three-dimensional model establishment module, configured to: select a grouting material solidified body specimen, preprocess the specimen to prepare a test sample and establish a three-dimensional model;

[0176] A two-dimensional slice image generation module, configured to: determine slice parameters, perform slicing operations on the three-dimensional model to generate multiple two-dimensional slice images, and preprocess the two-dimensional slice images;

[0177] A slice image data set construction module, configured to: construct a network structure, perform resolution processing on the preprocessed two-dimensional slice images, and construct a slice image data set;

[0178] A convolutional neural network model training module, configured to: input the images in the training set of the constructed slice image data set into the convolutional neural network model for training to obtain a trained convolutional neural network model;

[0179] A porosity evaluation module for the grouting material solidified body to be measured, configured to: process the grouting material solidified body to be measured to obtain a preprocessed two-dimensional image;

[0180] Input the preprocessed two-dimensional image into the trained convolutional neural network model to obtain the recognition result of pores, and obtain the porosity of the grouting material solidified body to be measured based on the recognition result of pores.

[0181] Example 5

[0182] The purpose of this embodiment is to provide a computer program product containing instructions, which, when running on a computer, enables the computer to execute the methods and functions involved in any one of the above embodiments.

[0183] Each step involved in the device of the above embodiments corresponds to the first method embodiment. For specific implementation manners, reference may be made to the relevant description part of the first embodiment. The term "computer-readable storage medium" should be understood to include a single medium or multiple media containing one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0184] Those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computer device. Optionally, they can be implemented by program codes executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0185] 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 solution of the present invention are still within the protection scope of the present invention.

Claims

1. A method for evaluating the porosity of a grouting material consolidation body based on image recognition, characterized in that: include: Select the consolidation body test block of grouting material, pre-treat the test block to prepare the test sample and establish the three-dimensional model; Determine slice parameters, perform slice operations on the three-dimensional model, generate multiple two-dimensional slice images, and pre-process the two-dimensional slice images; Construct a network structure, perform resolution processing on the preprocessed two-dimensional slice images, and construct a slice image dataset; Input the images in the training set of the constructed slice image data set into the convolutional neural network model for training, so as to obtain a trained convolutional neural network model; The solidified body of the grouting material to be tested is processed to obtain a pre-processed two-dimensional image; The preprocessed two-dimensional image is input into the trained convolutional neural network model to obtain the pore recognition result, and the porosity of the consolidated body of the grouting material to be tested is obtained based on the pore recognition result.

2. The method for evaluating the porosity of a grouting material consolidation body based on image recognition according to claim 1, characterized in that: After the test block is pre-processed to prepare the test sample, the sample is subjected to multi-angle rotation scanning to collect the two-dimensional projection data of the sample from different angles.

3. The method for evaluating the porosity of a grouting material consolidation body based on image recognition according to claim 2, characterized in that: The process of building a 3D model is as follows: Assume an initial 3D image; Project the current initial three-dimensional image to each acquisition angle to obtain the corresponding two-dimensional projection image; The calculated two-dimensional projection image is compared with the actually collected two-dimensional projection data, and the difference between the two is calculated; Correcting the three-dimensional image according to the difference, adjusting the value of each voxel in the three-dimensional image so as to minimize the projection difference; Repeat the above steps until certain convergence conditions are met and a three-dimensional model is obtained.

4. The method for evaluating the porosity of a grouting material consolidation body based on image recognition according to claim 1, characterized in that: Preprocess the 2D slice images, including: Grayscale the 2D slice images to highlight the difference between the pores and the surrounding material; Enhance the brightness or darkness of the porous area to create a sharper contrast with the surrounding material; The denoising algorithm is used to effectively remove noise, improve the signal-to-noise ratio of the image, and reduce interference factors.

5. The method for evaluating the porosity of a grouting material consolidation body based on image recognition according to claim 1, characterized in that: Constructing a network structure to perform resolution processing on the preprocessed two-dimensional slice image, wherein the network structure includes a feature extraction layer, a sub-pixel convolution layer and an upsampling layer; The preprocessed two-dimensional slice image is input into the feature extraction layer, which is usually composed of several convolutional layers to extract the features of the image; The image after feature extraction enters the sub-pixel convolution layer, which learns the sub-pixel displacement information of pixels through convolution operations and outputs a high-resolution feature map; The high-resolution feature map is converted into the final high-resolution 2D slice image through an upsampling layer.

6. The method for evaluating the porosity of a grouting material consolidation body based on image recognition according to claim 1, characterized in that: The neural network model uses a deformable convolution kernel to improve the convolutional neural network. The deformable convolution kernel consists of a traditional convolution kernel and an offset learning network. Calculate the output of the deformable convolution kernel: Assume that the input image is , the traditional convolution kernel is , the offset is , then the output of the deformable convolution kernel is ,in is the pixel position of the output image, is the sampling point position of the traditional convolution kernel, is the sampling area of ​​the traditional convolution kernel; During the calculation process, the sampling point position of the traditional convolution kernel is first adjusted according to the offset output by the offset learning network, and then the adjusted convolution kernel is convolved with the input image to obtain the output of the deformable convolution kernel.

7. The method for evaluating the porosity of a grouting material consolidation body based on image recognition according to claim 1, characterized in that: Based on the pore identification results, the porosity of the consolidation body of the grouting material to be tested is obtained. The specific process is as follows: Calculate the pore area ratio: For each 2D slice image, count the number of pixels in the pore area and then divide it by the total number of pixels in the image to get the pore area ratio. ; The porosity is calculated by volume integration method: the pore area ratio of each slice image is multiplied by the slice thickness to obtain the pore volume corresponding to the slice; The pore volumes of all slices were then summed and divided by the volume of the entire sample to obtain the three-dimensional structural porosity.

8. A grouting material consolidation body porosity assessment system based on image recognition, characterized in that: include: The three-dimensional model building module is configured to: select a test block of a grouting material consolidation body, pre-treat the test block to prepare a test sample and build a three-dimensional model; The two-dimensional slice image generation module is configured to: determine slice parameters, perform slice operations on the three-dimensional model, generate multiple two-dimensional slice images, and pre-process the two-dimensional slice images; The slice image data set construction module is configured to: construct a network structure, perform resolution processing on the preprocessed two-dimensional slice image, and construct a slice image data set; The convolutional neural network model training module is configured to: input the images in the training set of the constructed slice image data set into the convolutional neural network model for training, so as to obtain a trained convolutional neural network model; The porosity evaluation module of the grouting material consolidation body to be tested is configured to: process the grouting material consolidation body to be tested to obtain a pre-processed two-dimensional image; The preprocessed two-dimensional image is input into the trained convolutional neural network model to obtain the pore recognition result, and the porosity of the consolidated body of the grouting material to be tested is obtained based on the pore recognition result.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method described in any one of claims 1 to 7 are implemented.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 7 are performed.

Images