Construction joint surface evaluation system, construction joint surface evaluation method, and construction joint surface evaluation program

Abstract

To provide a construction joint surface evaluation system, a construction joint surface evaluation method, and a construction joint surface evaluation program capable of evaluating a construction joint surface more accurately than before.SOLUTION: A placing joint surface evaluation system 1 for evaluating a placing joint surface of concrete includes an image acquisition unit 10 configured to acquire an image of the placing joint surface, and a determination unit 20 configured to determine a surface roughness of the placing joint surface from the image by using a learning model, wherein the learning model is learned by learning data in which at least one of a color tone of the image, weather at a time of photographing, and presence or absence of water on a surface, and the surface roughness of the placing joint surface are associated with the image as a label.SELECTED DRAWING: Figure 1A

Description

[Technical Field]

[0001] The present invention relates to a construction joint surface evaluation system, a construction joint surface evaluation method, and a construction joint surface evaluation program. [Background technology]

[0002] When pouring concrete in the construction of concrete dams, box culverts, etc., it is necessary to remove laitance and loose aggregate from the surface of the joints using a high-pressure washer or similar equipment, and to roughen the surface. If the joints are not properly treated, the integrity of the concrete structure will be compromised, leading to water leakage and structural defects. Therefore, it is necessary to treat the joints while checking their condition. Currently, the joints in concrete dams are checked visually using reference photographs of joint alignment as an indicator.

[0003] A related technique is described in Patent Document 1. The technique described in Patent Document 1 judges the quality of the construction joint surface treatment based on an image of the construction joint surface. The photographed image is smoothed to generate two smoothed images, the first and second, and the aggregate portion is determined by the difference between them. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2023-064262 Summary of the Invention [Problem to be solved by the invention]

[0005] The technology described in Patent Document 1 calculates pixel differences, and is therefore strongly affected by weather and shadows. As a result, the technology described in Patent Document 1 has the problem of poor accuracy depending on the environment in which the construction joint surface is photographed. In other words, this technology does not take into account factors that affect the determination of the construction joint surface (for example, the weather at the time of photography), resulting in low accuracy in the determination.

[0006] From this perspective, the present invention provides a joint surface evaluation system, a joint surface evaluation method, and a joint surface evaluation program that can evaluate joint surfaces more accurately than conventional systems. [Means for solving the problem]

[0007] The construction joint surface evaluation system according to the present invention is a system for evaluating a concrete construction joint surface. The construction joint surface evaluation system includes an image acquisition unit that acquires an image of the construction joint surface, and a determination unit that determines the surface roughness of the construction joint surface from the image using a learning model. The learning model is trained using learning data that is linked to the image as labels, including at least one of the color of the image, the weather at the time of shooting, and the presence or absence of water on the surface, and the surface roughness of the joint surface.

[0008] The learning model is trained using learning data in which the color of the image, the weather at the time of shooting, the presence or absence of water on the surface, and the surface roughness of the joint surface are linked to the image as labels, for example.

[0009] The construction joint surface evaluation system of the present invention learns the relationship between the image of the construction joint surface and the surface roughness of the construction joint surface, including factors that affect the judgment of the construction joint surface (for example, the color of the image, the weather at the time of shooting, the presence or absence of water on the surface of the construction joint, etc.). Therefore, factors that affect the judgment of the construction joint surface are taken into consideration, and the construction joint surface can be evaluated more accurately than before.

[0010] The method for evaluating a concrete joint surface according to the present invention includes an image acquisition step of causing an image acquisition unit to acquire an image of the concrete joint surface, and a determination step of determining the surface roughness of the concrete joint surface from the image using a learning model. The learning model is trained using learning data that is linked to the image as labels, including at least one of the color of the image, the weather at the time of shooting, and the presence or absence of water on the surface, and the surface roughness of the joint surface.

[0011] The construction joint surface evaluation method according to the present invention learns the relationship between the image of the construction joint surface and the surface roughness of the construction joint surface, including factors that affect the judgment of the construction joint surface (for example, the color of the image, the weather at the time of shooting, the presence or absence of water on the surface of the construction joint, etc.). Therefore, factors that affect the judgment of the construction joint surface are taken into consideration, and the construction joint surface can be evaluated more accurately than before.

[0012] The construction joint surface evaluation program according to the present invention is for causing a computer to execute the construction joint surface evaluation method described above. [Effects of the Invention]

[0013] According to the present invention, the joint surface can be evaluated more accurately than conventional methods. [Brief explanation of the drawings]

[0014] [Figure 1A] 1 is an example of a configuration diagram of a construction joint surface evaluation system according to a first embodiment of the present invention. [Figure 1B] 1 is an example of a configuration diagram of a construction joint surface evaluation system according to a first embodiment of the present invention. [Figure 2] FIG. 1 is a diagram for explaining a learning model. [Figure 3] This is an example of classification of surface roughness. [Figure 4] 10A and 10B are examples of color classification, where (a) shows an example of the luminance distribution and gray level histogram of a white image, and (b) shows an example of the luminance distribution and gray level histogram of a brown image. [Figure 5] FIG. 10 is a diagram illustrating an example of a workflow for creating learning data. [Figure 6] FIG. 10 is a diagram showing the relationship between categories, labels, and classes of images of construction joints. [Figure 7] FIG. 10 is a diagram illustrating an example of a workflow for learning a learning model. [Figure 8] FIG. 10 is a diagram showing an example of a workflow for determining a joint surface. [Figure 9A] 10 is an example of an acquired image. [Figure 9B]10 is an example of a divided image. [Figure 9C] This is an image of the combined image. [Figure 9D] These are the results of a verification test to confirm the evaluation accuracy. DETAILED DESCRIPTION OF THE INVENTION

[0015] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. Each drawing is merely a schematic illustration to allow a sufficient understanding of the present invention. Therefore, the present invention is not limited to the illustrated examples. In each drawing, common or similar components are designated by the same reference numerals, and redundant explanations thereof may be omitted.

[0016] (Configuration of construction joint surface evaluation system) The configuration of a construction joint surface evaluation system 1 according to an embodiment will be described with reference to Figures 1A and 1B. Figures 1A and 1B are examples of configuration diagrams of a construction joint surface evaluation system 1 according to an embodiment. The construction joint surface evaluation system 1A shown in Figure 1A and the construction joint surface evaluation system 1B shown in Figure 1B have different hardware configurations.

[0017] The construction joint surface evaluation system 1A shown in FIG. 1A is a system that evaluates a construction joint surface based on an image of the concrete construction joint surface. The construction joint surface evaluation system 1A includes an image acquisition unit 10 and a determination unit 20. In the construction joint surface evaluation system 1A, a tablet terminal 2 includes the image acquisition unit 10 and the determination unit 20. A terminal such as a smartphone may be used instead of the tablet terminal 2. The tablet terminal 2 and the smartphone are examples of a computer.

[0018] The image acquisition unit 10 is a means for acquiring an image of the concrete joint surface. The image acquisition unit 10 is, for example, a digital camera, and is capable of capturing color images (digital images having RGB values).

[0019] The determination unit 20 is a means for determining the treatment state of a concrete joint surface based on an image of the joint surface. The determination unit 20 is realized by executing a program using, for example, a CPU (Central Processing Unit). The determination unit 20 determines the treatment state of the joint surface using artificial intelligence technology (AI technology). The program that realizes the determination unit 20 is an application program that utilizes artificial intelligence technology. The determination unit 20 has a learning model, and during learning, it learns the relationship between the image of the joint surface and the treatment state of the joint surface. Then, the determination unit 20 uses the learned learning model to determine the treatment state of the joint surface from the image of the joint surface.

[0020] The construction joint surface evaluation system 1B shown in FIG. 1B is a system that evaluates a construction joint surface based on an image of the concrete construction joint surface. The construction joint surface evaluation system 1B includes an image acquisition unit 10 and a determination unit 20. In the construction joint surface evaluation system 1B, a tablet terminal 2 includes the image acquisition unit 10, and a personal computer 3 includes the determination unit 20. The personal computer 3 is installed, for example, in a site office, and is capable of wireless communication with the tablet terminal 2. The personal computer 3 is an example of a computer. Note that the following description will be given assuming the construction joint surface evaluation system 1A shown in FIG. 1A.

[0021] (Explanation of the learning model) The learning model assumed in this embodiment will be described with reference to Fig. 2. Fig. 2 is a diagram for explaining the learning model. A training model is a model that can perform image classification tasks. An example of a training model is YOLOv8, but it is not limited to this. An image classification task is to classify objects in an image into predefined categories. When classifying image data into predefined categories, a large amount of labeled training data with categories labeled is prepared during training. Then, feature conversion is performed on the training data using a feature extractor, and a classifier is created through training. During inference, the feature extractor and classifier are used to determine which category the input image belongs to.

[0022] (Building training data) Images of concrete joints labeled for each category are prepared as training data (teaching data). A category consists of (1) the surface roughness of the joint, and (2) at least one of the following: the color of the image, the weather at the time of the photo, and the presence or absence of water on the surface of the joint. In other words, the training data consists of images of concrete joints, each of which is associated with a label that includes at least one of the following: (1) the surface roughness of the joint, and (2) the color of the image, the weather at the time of the photo, and the presence or absence of water on the surface of the joint. A label is a classification name assigned to a group of images divided into categories.

[0023] Four surface roughness categories are defined: shallow, medium, deep, and untreated. The surface roughness categories of "shallow, medium, and deep" are determined based on the results of quantitative evaluation of the centerline average roughness Ra, measured using the light-section method to measure the unevenness of the joint surface. "Untreated" refers to a condition in which the laitance layer has not been removed. Figure 3 summarizes the relationship between the evaluation value Ra obtained using the light-section method and example images of the joint surface. Figure 3 shows an example of surface roughness classification. In Figure 3, an evaluation value Ra of less than 0.2 mm using the light-section method is defined as "shallow," an evaluation value Ra of 0.2 mm or more but less than 0.5 mm is defined as "medium," and an evaluation value Ra of 0.5 mm or more is defined as "deep."

[0024] Two color shades are defined for the categories: brown and white. Color shades are quantitatively determined based on the brightness information of an image of the joint surface. For example, an RGB color image is converted to a grayscale image, and a histogram is used to determine whether the color is brown or white. Figure 4 shows an example of color shade classification. (a) shows an example of the brightness distribution and histogram of a white image, and (b) shows an example of the brightness distribution and histogram of a brown image. After converting the image to a grayscale image using the formula (R+G+B) / 3 and checking the histogram (256 levels), the average histogram values ​​were approximately 160 for white and approximately 120 for brown. Therefore, a histogram average of 140 or greater is determined as white, and one below 140 is determined as brown.

[0025] The weather categories are defined as "sunny," "cloudy," and "rainy." For example, the sky is visually checked when photographing the joint surface, and the weather is determined based on the results.

[0026] The presence or absence of water as a category is defined as "water present" and "water absent." For example, the presence or absence of water on the surface of the photographed joint is visually confirmed, and the presence or absence of water is determined based on the results. "Water present" refers to, for example, when there is a puddle on the surface of the joint, and "water absent" refers to, for example, when there is no puddle on the surface of the joint. When the surface of the joint is wet, it is classified as "water absent."

[0027] The workflow for creating learning data will be described with reference to Fig. 5. Fig. 5 is a diagram showing an example of the workflow for creating learning data. In step S11, the image of the construction joint surface is photographed using the image acquisition unit 10, and an image of the construction joint surface is acquired. Next, in step S12, the acquired image is divided into images of any size, and divided images are created. Note that the division process of the acquired image is not an essential step and can be omitted.

[0028] Next, in steps S13 to S16, labels are assigned to the classified images. Note that the order of labeling is not limited to that shown in Fig. 5. In this embodiment, four pieces of information are assigned as labels to the classified images: (a) the surface roughness of the joint surface, (b) the color of the image, (c) the weather at the time of photographing, and (d) the presence or absence of water on the surface of the joint surface. In other words, the categories in this embodiment are composed of the above-mentioned (a) to (d).

[0029] In step S13, the divided images are classified based on the surface roughness, and labeled images are created by adding the classification results (shallow, medium, deep, or unprocessed) as labels. In addition, in step S14, the labeled images (divided images) created in step S13 are classified based on color, and labeled images are created by further adding the classification results (either brown or white) as labels.

[0030] In step S15, the labeled images (divided images) created in step S14 are classified based on weather, and labeled images are created by further adding the classification result (either sunny, cloudy, or rainy) as a label. In addition, in step S16, the labeled images (divided images) created in step S15 are classified based on the presence or absence of water, and a labeled image is created to which the classification result (either water present or water absent) is further attached as a label.

[0031] Next, in step S17, labeled images with four pieces of information, namely (a) the surface roughness of the joint surface, (b) the color of the image, (c) the weather at the time of photographing, and (d) the presence or absence of water on the surface of the joint surface, are output as learning data (teacher data).

[0032] The relationship between the categories, labels, and classes of images of concrete joints is shown in Figure 6. Figure 6 shows the relationship between the categories, labels, and classes of images of concrete joints. Here, classes are the final classifications displayed as results during inference. In this embodiment, categories are defined as combinations of (a) the surface roughness of the concrete joint, (b) the color of the image, (c) the weather at the time of photography, and (d) the presence or absence of water on the surface of the concrete joint. However, (a) the surface roughness of the concrete joint is important for evaluating concrete joints. In other words, (b) the color of the image, (c) the weather at the time of photography, and (d) the presence or absence of water on the surface of the concrete joint may be output as inference results, but these are unnecessary information for determining the treatment of the concrete joint. Therefore, in this embodiment, information other than surface roughness is removed before output (information unrelated to the determination of the treatment of the concrete joint is removed before output).

[0033] In this embodiment, the classes are the surface roughness of the joint surface, and four classes are defined: "shallow, medium, deep, and untreated." As with the categories, the surface roughness "shallow, medium, and deep" are determined based on the results of quantitative evaluation of the centerline average roughness Ra by measuring the unevenness of the joint surface using the light-section method. Note that the classes may also directly express the judgment results of the surface treatment, such as "OK, NG" or "good, poor." The classes are defined in advance to correspond to the categories, for example.

[0034] (Learning model) The workflow for learning a learning model will be described with reference to Fig. 7. Fig. 7 is a diagram showing an example of the workflow for learning a learning model. In step S21, the learning data (teaching data) created in Fig. 5 is input. The learning data here is a labeled image with four pieces of information: (a) the surface roughness of the joint, (b) the color of the image, (c) the weather at the time of shooting, and (d) the presence or absence of water on the surface of the joint.

[0035] Next, in step S22, machine learning processing is performed on the learning model using the learning data (teacher data) to create a trained learning model. Next, in step S23, the trained learning model is output. The learning model output in step S23 is configured to classify the concrete joint surface shown in the image into categories by inputting an image. The categories here are defined by a combination of (a) the surface roughness of the joint surface, (b) the color of the image, (c) the weather at the time of shooting, and (d) the presence or absence of water on the surface of the joint surface.

[0036] (Determination of joint surface) The workflow for determining a construction joint surface will be described with reference to Fig. 8. Fig. 8 is a diagram showing an example of the workflow for determining a construction joint surface. Note that a trained learning model is stored in advance in a storage means of the tablet terminal 2 (see Fig. 1A).

[0037] In step S31, the image acquisition unit 10 is used to capture an image of the construction joint surface to be determined, and an image of the construction joint surface is acquired. The location information of the tablet terminal 2 is also acquired. The location information is acquired from GPS information built into the tablet terminal 2. The tablet terminal 2 stores the acquired image and location information in its own storage means. Note that by adding the location information of the tablet terminal 2 to the acquired image and saving it, the acquired image can be used as evidence. An example of an acquired image is shown in FIG. 9A.

[0038] Next, in step S32, the determination unit 20 performs a division process on the acquired image (see FIG. 9A) to create divided images. An example of the divided image is shown in FIG. 9B. In the division process, the number of divisions can be set for the acquired image, such as "4," "9," "16," "25," or "36." Here, the explanation will be given assuming the number of divisions is "36." Note that the division process of the acquired image is not an essential step and can be omitted.

[0039] Next, in step S33, the determination unit 20 performs a determination process using the segmented image and the learning model to determine the construction joint surface that appears in the segmented image. Specifically, the determination unit 20 inputs the segmented image into the learning model to classify it into categories, and then determines a class based on the categories. The categories are defined by a combination of (a) the surface roughness of the construction joint surface, (b) the color of the image, (c) the weather at the time of shooting, and (d) the presence or absence of water on the construction joint surface. The classes are defined only by the surface roughness of the construction joint surface. The determination process for the segmented image is repeated the number of times equal to the number of divisions.

[0040] Next, in step S34, the determination unit 20 performs a process of combining the divided images on which the determination results (classes) are reflected, and combines the divided images on which the determination results (classes) are reflected. An image of the combined image obtained by combining the divided images on which the determination results (classes) are reflected is shown in Fig. 9C.

[0041] Furthermore, in step S35, the determination unit 20 uses the combined image to create a display screen that displays the determination result, and causes the display screen to be displayed on the display of the tablet terminal 2.

[0042] Furthermore, in step S36, the determination unit 20 stores the display screen including the determination result in its own storage means. By adding the position information of the tablet terminal 2 to the display screen (or the combined image) and storing it, the display screen (or the combined image) can be used as evidence.

[0043] As described above, the construction joint surface evaluation system 1 according to this embodiment learns the relationship between the image of the construction joint surface and the surface roughness of the construction joint surface, including factors that affect the judgment of the construction joint surface (for example, the color of the image, the weather at the time of shooting, the presence or absence of water on the surface of the construction joint surface, etc.). Therefore, factors that affect the judgment of the construction joint surface are taken into consideration, and the construction joint surface can be evaluated more accurately than before.

[0044] A verification test was conducted to confirm the evaluation accuracy of the construction joint surface evaluation system 1, and some of the results are shown in Figure 9D. Figure 9D shows the results of the verification test to confirm the evaluation accuracy. The evaluation was judged for each divided image, and if the judgment result was correct, it was marked "correct," and if it was incorrect, it was marked "incorrect." When the evaluation judgment results were verified, it was found that, although there was some variation, it was possible to perform evaluations equivalent to those using the conventional light-section method. [Explanation of symbols]

[0045] 1. Construction joint surface evaluation system 2. Tablet devices 3. Personal Computers 10 Image acquisition unit 20 Judgment section

Claims

1. A joint surface evaluation system for evaluating concrete joint surfaces, An image acquisition unit that acquires an image of the joint surface; A determination unit that determines the surface roughness of the joint surface from the image using a learning model, The learning model is learned using learning data in which at least one of the color of the image, the weather at the time of shooting, and the presence or absence of water on the surface, and the surface roughness of the joint surface are linked to the image as labels. A construction joint surface evaluation system characterized by:

2. The learning model is learned using learning data in which the color of the image, the weather at the time of shooting, the presence or absence of water on the surface, and the surface roughness of the joint surface are linked to the image as labels. The construction joint surface evaluation system according to claim 1 .

3. A method for evaluating a concrete joint surface, An image acquisition step of causing an image acquisition unit to acquire an image of the joint surface; A determination step of determining the surface roughness of the joint surface from the image using a learning model, The learning model is learned using learning data in which at least one of the color of the image, the weather at the time of shooting, and the presence or absence of water on the surface, and the surface roughness of the joint surface are linked to the image as labels. A method for evaluating a construction joint surface.

4. A construction joint surface evaluation program for causing a computer to execute the construction joint surface evaluation method according to claim 3.

Images