Method and device for evaluating dispersibility of steel fibers in UHPC (Ultra High Performance Concrete) based on deep learning

Abstract

The invention provides a deep learning-based UHPC steel fiber dispersibility evaluation method and device, and the method comprises the steps: carrying out the pixel-level semantic segmentation of a UHPC steel fiber image based on a deep learning framework, achieving the quick and accurate recognition of fibers, carrying out the noise reduction of the image before the calculation of a fiber dispersion coefficient, and achieving the evaluation of the dispersibility of a UHPC steel fiber. Therefore, the calculation accuracy is further improved, and a relatively accurate calculation result is still provided for a low-quality image.

Description

Technical Field

[0001] The present application belongs to the field of concrete materials and image recognition technology, and in particular, relates to a method and device for evaluating the dispersion of steel fibers in UHPC based on deep learning. Background Art

[0002] Ultra-high performance concrete (UHPC) is an innovative cement-based composite material that is widely used in engineering structures due to its excellent workability, mechanical properties and durability. High amounts of steel fibers greatly improve the mechanical properties of UHPC, especially the tensile and bending properties. When UHPC is subjected to tension, the steel fibers act as a bridge in the UHPC to transfer stress. The bridging effect is significantly affected by the distribution and orientation of the fibers in the UHPC. How to accurately and conveniently obtain the fiber distribution characteristics has become a prerequisite for solving such problems.

[0003] At present, there are many methods for analyzing the dispersion of steel fibers in ultra-high strength concrete, such as X-CT tomography, sieving, washing, weighing and image processing. Among them, although X-CT tomography and three-dimensional reconstruction can accurately reflect the dispersion of fibers, they are too expensive to be widely used; although the sieving, washing and weighing method is simple, it has a large deviation from the actual fiber distribution and cannot obtain the fiber orientation angle parameters; although the image analysis method is low-cost and the results obtained are accurate, it is highly subjective and complicated to operate. The shortcomings of the image analysis method are specifically reflected in the high quality requirements for sample photos when acquiring images; in the image processing process, the selection of thresholds is highly subjective, resulting in differences in results among different operators; in addition, image analysis usually relies on a large database, and building a large database usually requires a lot of manual labor.

[0004] Therefore, it is necessary to propose a new technical solution to overcome the above technical problems and realize the rapid and accurate evaluation of the dispersion of steel fibers in UHPC. Summary of the invention

[0005] A method and device for evaluating the dispersion of steel fibers in UHPC based on deep learning is used to solve the above technical problems. Based on the deep learning framework, pixel-level semantic segmentation of UHPC steel fiber images is performed to achieve fast and accurate fiber recognition, and the image is denoised before calculating the fiber dispersion coefficient, thereby further improving the accuracy of the calculation, so that it still has a relatively accurate calculation result for low-quality images.

[0006] In a first aspect, an embodiment of the present application provides a method for evaluating the dispersion of steel fibers in UHPC based on deep learning, comprising the following steps:

[0007] S1. Obtain a UHPC image dataset, preprocess each image in the image dataset, and construct a training set and a test set based on the preprocessed image dataset;

[0008] S2. constructing a U-Net semantic segmentation model, training the model using a training set, and testing the model using a test set to obtain a UHPC fiber recognition model;

[0009] S3, inputting the UHPC image to be evaluated into the UHPC fiber recognition model to obtain a fiber distribution image; wherein the fiber distribution image is a binary image, and white pixels represent the recognized fibers;

[0010] S4. Remove the noise in the fiber distribution image and calculate the fiber dispersion coefficient.

[0011] The step S1 of preprocessing each image in the image data set specifically includes:

[0012] Resizing and graying each image in the image data set, and cutting it into a predetermined number of sub-images; wherein each sub-image has the same size;

[0013] Rotate each sub-image after cutting.

[0014] The U-Net semantic segmentation model in step S2 includes an Encoder structure and a Decoder structure. The Encoder structure is used to extract semantic feature information, and the Decoder structure is used to perform segmentation prediction.

[0015] The encoder structure includes 4 layers of downsampling, the decoder structure includes 4 layers of upsampling, and Skip Connection is used to perform skip layer connection, and the feature maps in the downsampling are fused in the upsampling.

[0016] Wherein, removing the noise in the fiber distribution image in step S4 specifically includes:

[0017] S411, acquiring each connected domain composed of white pixels in the fiber distribution image;

[0018] S412, calculating the area of ​​each connected domain;

[0019] S413, identifying a connected domain whose area is smaller than a preset threshold, and obtaining a small area;

[0020] S414: Set the value of each pixel in each small area to 0.

[0021] The calculation of the fiber dispersion coefficient in step S4 specifically includes:

[0022] S421, dividing the fiber distribution image into a plurality of units;

[0023] S422, counting the number of fibers in each unit;

[0024] S423. Calculate the fiber dispersion coefficient α using the following formula:

[0025]

[0026] Among them, n is the number of divided units, x i is the number of fibers in the ith unit, and x0 is the average number of fibers.

[0027] In a second aspect, an embodiment of the present application provides a device for evaluating the dispersion of steel fibers in UHPC based on deep learning, the device comprising:

[0028] A data set acquisition module, used to acquire a UHPC image data set, preprocess each image in the image data set, and construct a training set and a test set based on the preprocessed image data set;

[0029] A model building module is used to build a U-Net semantic segmentation model, train the model using a training set, and test the model using a test set to obtain a UHPC fiber recognition model;

[0030] A fiber identification module is used to input the UHPC image to be evaluated into the UHPC fiber identification model to obtain a fiber distribution image; wherein the fiber distribution image is a binary image, and white pixels represent the identified fibers;

[0031] The coefficient calculation module is used to remove noise in the fiber distribution image and calculate the fiber dispersion coefficient.

[0032] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0033] The present invention provides a method and device for evaluating the dispersion of steel fibers in UHPC based on deep learning. Based on the deep learning framework, pixel-level semantic segmentation of UHPC steel fiber images is performed to achieve fast and accurate fiber recognition, and denoising is performed on the image before calculating the fiber dispersion coefficient, thereby further improving the accuracy of the calculation, so that it still has a relatively accurate calculation result for low-quality images. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0035] Figure 1is a flow chart of a method for evaluating the dispersion of steel fibers in UHPC based on deep learning provided in an embodiment of the present application;

[0036] Figure 2 It is a structural diagram of the U-Net semantic segmentation model provided in an embodiment of the present application;

[0037] Figure 3 is a schematic diagram of removing noise from a fiber distribution image using a removal method provided in an embodiment of the present application;

[0038] Figure 4 is a schematic diagram of a steel fiber dispersion evaluation device in UHPC based on deep learning provided in an embodiment of the present application;

[0039] Figure 5 is a comparison chart of the denoising effect of the denoising method provided in the embodiment of the present application and other denoising methods;

[0040] Figure 6 It is a comparison chart of the carbon fiber prediction results of the analysis method provided in the embodiment of the present application and other analysis methods. DETAILED DESCRIPTION

[0041] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0042] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0043] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0044] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0045] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", "fourth", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0046] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0047] First embodiment

[0048] Figure 1 A flow chart of a method for evaluating the dispersion of steel fibers in UHPC based on deep learning provided in Example 1 of the present application is shown. Figure 1 As shown, the steel fiber dispersion evaluation method in UHPC based on deep learning specifically includes:

[0049] S1. Obtain a UHPC image dataset, preprocess each image in the image dataset, and construct a training set and a test set based on the preprocessed image dataset;

[0050] The preprocessing of each image in the image data set specifically includes:

[0051] Resizing and graying each image in the image data set, and cutting it into a predetermined number of sub-images; wherein each sub-image has the same size;

[0052] Each cut sub-image is rotated to enhance the robustness of the image data.

[0053] The step of constructing the training set and the test set includes: constructing the training set and the test set based on each of the cut sub-images.

[0054] Among them, the samples in the training set need to be labeled, such as using labelme for labeling.

[0055] Before the image is preprocessed, the following steps are also included:

[0056] Screening the images in the UHPC image data set, selecting valid images as sample initial images, and further preprocessing each image in the image data set includes preprocessing the selected sample initial images.

[0057] The selected sample initial image is an image with a flat surface, uniform fill light, and no obvious light spots, reflections or exposure.

[0058] S2. constructing a U-Net semantic segmentation model, training the model using a training set, and testing the model using a test set to obtain a UHPC fiber recognition model;

[0059] Among them, U-Net is a semantic segmentation model, and its output is the probability that each pixel in the image belongs to each category.

[0060] See the structure diagram of the U-Net semantic segmentation model for details. Figure 2 , which includes an Encoder structure and a Decoder structure. The Encoder structure extracts semantic feature information, and the Decoder structure performs segmentation prediction.

[0061] The encoder structure includes 4 layers of downsampling, the decoder structure includes 4 layers of upsampling, and Skip Connection is used to perform skip layer connection, and the feature maps in the downsampling are fused in the upsampling.

[0062] S3. Inputting the UHPC image to be evaluated into the UHPC fiber recognition model to obtain a fiber distribution image; wherein the fiber distribution image is a binary image, and white pixels represent the recognized fibers.

[0063] Specifically, after the UHPC image to be evaluated is input into the UHPC fiber recognition model, the fiber distribution image output by the model is a binary image, in which black pixels represent the background, i.e., the concrete matrix, and white pixels represent the position and morphology of the fibers.

[0064] S4. Remove the noise in the fiber distribution image and calculate the fiber dispersion coefficient.

[0065] Among them, there is still a certain amount of noise in the binary image output by the UHPC fiber recognition model, that is, some small-area background pixels are easily mistakenly identified as fibers. Therefore, in order to remove the small-area background area mistakenly identified as fibers, the present application further denoises the fiber distribution image to obtain accurate fiber distribution, and then accurately calculate the fiber dispersion coefficient.

[0066] The step of removing the noise in the fiber distribution image specifically involves using a removal method to remove the noise in the fiber distribution image. Figure 3 A schematic diagram of the method is given, in which each square on the right side of the figure represents a pixel, 0 represents that the pixel is a background pixel, and 1 represents that the pixel is a foreground pixel.

[0067] The removal method specifically includes:

[0068] S411, obtaining each connected domain composed of white pixels in the fiber distribution image; Figure 3 Each area formed by the white pixels in ;

[0069] Specifically, the cv2.connectedComponents() function of OpenCV is used to find each connected domain in the image.

[0070] S412, calculating the area of ​​each connected domain;

[0071] S413, identifying the connected domain whose area is smaller than a preset threshold, and obtaining a small area; Figure 3 The area encircled by the grey circle;

[0072] S414: Set the value of each pixel in each small area to 0.

[0073] Specifically, the np.where() function of OpenCV is used to find a small area according to a preset threshold, and the value of each pixel in the small area is set to 0.

[0074] The calculation of fiber dispersion coefficient specifically includes:

[0075] S421, dividing the fiber distribution image into a plurality of units;

[0076] Specifically, the fiber distribution map may be divided into equal-area units to obtain a plurality of units, such as units of 4*4 pixels.

[0077] S422, counting the number of fibers in each unit;

[0078] Specifically, a math module is imported to count the number of roots of each unit in the fiber distribution image.

[0079] The math module is a built-in library in Python, which is used to implement mathematical operations on floating-point numbers.

[0080] S423. Calculate the fiber dispersion coefficient α using the following formula:

[0081]

[0082] Among them, n is the number of divided units, x i is the number of fibers in the ith unit, x0 is the average number of fibers, and the calculation formula of x0 is as follows:

[0083]

[0084] Second embodiment

[0085] Figure 4 A schematic diagram of a steel fiber dispersion evaluation device in UHPC based on deep learning provided in Example 2 of the present application is shown, Figure 4 As shown, the steel fiber dispersion evaluation device 4 in UHPC based on deep learning specifically includes:

[0086] A data set acquisition module 41 is used to acquire a UHPC image data set, preprocess each image in the image data set, and construct a training set and a test set based on the preprocessed image data set;

[0087] A model building module 42 is used to build a U-Net semantic segmentation model, train the model using a training set, and test the model using a test set to obtain a UHPC fiber recognition model;

[0088] The fiber identification module 43 is used to input the UHPC image to be evaluated into the UHPC fiber identification model to obtain a fiber distribution image; wherein the fiber distribution image is a binary image, and white pixels represent the identified fibers;

[0089] The coefficient calculation module 44 is used to remove noise in the fiber distribution image and calculate the fiber dispersion coefficient.

[0090] Third embodiment

[0091] Based on the method and device described in the first and second embodiments above, the dispersion of carbon fiber in UHPC images is evaluated, and the denoising method based on the removal method of the present application and other common denoising methods are compared and analyzed for their denoising effects. The comparison results are as follows: Figure 5 shown.

[0092] It can be seen that the denoising effect of the removal method based on the present application is significantly better than that of several other common denoising methods, including the denoising method based on deep learning (Deep learning), the denoising method based on mean filter (Mean Filter), the denoising method based on median filter (Median Filter), and the denoising method based on Gaussian filter (Gaussian Filter), regardless of whether it is a high-contrast image or a low-contrast image.

[0093] Fourth embodiment

[0094] Based on the method and device described in the above-mentioned embodiment 1 and embodiment 2, the carbon fiber dispersion in the UHPC image is evaluated, and a correlation analysis is performed between the carbon fiber dispersion coefficient evaluation result based on deep learning in this application and the actual value of the dispersion coefficient calculated after manual processing. The analysis results are as follows: Figure 6 As shown in the figure, for specimens with various different proportions, the dispersion coefficient evaluation results based on deep learning provided in this application are relatively close to the actual values, and the calculation accuracy is relatively high.

[0095] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0096] In the embodiments provided in the present application, it should be understood that the disclosed devices / network equipment and methods can be implemented in other ways. For example, the device / network equipment embodiments described above are merely schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0097] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0098] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for evaluating the dispersion of steel fibers in UHPC based on deep learning, characterized in that: The following steps are involved: S1. Obtain a UHPC image dataset, preprocess each image in the image dataset, and construct a training set and a test set based on the preprocessed image dataset; S2. constructing a U-Net semantic segmentation model, training the model using a training set, and testing the model using a test set to obtain a UHPC fiber recognition model; S3, inputting the UHPC image to be evaluated into the UHPC fiber recognition model to obtain a fiber distribution image; wherein the fiber distribution image is a binary image, and white pixels represent the recognized fibers; S4. Remove the noise in the fiber distribution image and calculate the fiber dispersion coefficient.

2. The method according to claim 1, characterized in that The step S1 pre-processes each image in the image data set, specifically including: Resizing and graying each image in the image data set, and cutting it into a predetermined number of sub-images; wherein each sub-image has the same size; Rotate each sub-image after cutting.

3. The method according to claim 1, characterized in that The U-Net semantic segmentation model in step S2 includes an Encoder structure and a Decoder structure. The Encoder structure is used to extract semantic feature information, and the Decoder structure is used to perform segmentation prediction.

4. The method according to claim 3, characterized in that The Encoder structure includes 4 layers of downsampling, and the Decoder structure includes 4 layers of upsampling. Skip Connection is used for skip connection, and feature maps in downsampling are fused in upsampling.

5. The method according to claim 1, characterized in that The step S4 of removing noise from the fiber distribution image specifically includes: S411, acquiring each connected domain composed of white pixels in the fiber distribution image; S412, calculating the area of ​​each connected domain; S413, identifying a connected domain whose area is smaller than a preset threshold, and obtaining a small area; S414: Set the value of each pixel in each small area to 0.

6. The method according to claim 5, characterized in that The fiber dispersion coefficient is calculated in step S4, specifically including: S421, dividing the fiber distribution image into a plurality of units; S422, counting the number of fibers in each unit; S423. Calculate the fiber dispersion coefficient α using the following formula: Among them, n is the number of divided units, x i is the number of fibers in the ith unit, and x0 is the average number of fibers.

7. A device for evaluating the dispersion of steel fibers in UHPC based on deep learning, characterized in that: The device comprises: A data set acquisition module, used to acquire a UHPC image data set, preprocess each image in the image data set, and construct a training set and a test set based on the preprocessed image data set; A model building module is used to build a U-Net semantic segmentation model, train the model using a training set, and test the model using a test set to obtain a UHPC fiber recognition model; A fiber identification module is used to input the UHPC image to be evaluated into the UHPC fiber identification model to obtain a fiber distribution image; wherein the fiber distribution image is a binary image, and white pixels represent the identified fibers; The coefficient calculation module is used to remove noise in the fiber distribution image and calculate the fiber dispersion coefficient.

Images