JP7471178B2 - Image region classification model creation device, concrete evaluation device, and concrete evaluation program - Google Patents

Description

The present invention relates to an image region classification model creation device, a concrete evaluation device, and a concrete evaluation program.

In 2019, the revision of "JIS A5308 Ready Mixed Concrete" added high-fluidity concrete, which is controlled by slump flow, to ordinary concrete. The test method for the material separation resistance of high-fluidity concrete requires visual confirmation of the presence or absence of uneven distribution of coarse aggregate after the slump flow test, and the reality is that no objective method of judgment has been established.

Conventionally, the following techniques have been applicable to evaluate the material separation resistance of highly fluid concrete:

Patent Document 1 discloses a test method for evaluating the material separation resistance of high-flow concrete when it is fresh. In this test method, a base plate and a slump cone for a slump flow test are prepared, and a plurality of annular sample dividing members with different diameters are prepared for dividing the slump flow concrete sample formed on the base plate into a plurality of regions by inserting them into the slump flow concrete sample. In this test method, the slump cone is placed on the base plate, a concrete sample is filled in the slump cone, and the slump cone is pulled up and removed to form a slump flow of the concrete sample on the base plate. In this test method, the plurality of sample dividing members are inserted into the slump flow concrete sample so as to be approximately concentric with the center of the slump flow on the base plate, thereby dividing the slump flow concrete sample into a plurality of approximately concentric regions. In this test method, concrete samples are taken from each of the plurality of regions, and the material separation resistance of the concrete is evaluated for each of the plurality of taken samples.

Patent Document 2 also discloses a method for evaluating material separation in high-fluidity concrete. In this method, when preparing high-fluidity concrete containing a high-performance water-reducing agent, the concrete is loaded into a specified mold, and a specified water-absorbent sheet is placed in contact with the entire top surface of the concrete in the mold. In this method, the concrete is left in this contact state for a certain period of time, after which the water-absorbent sheet is peeled off, and the amount of material adsorbed onto the sheet is calculated from the difference in weight of the sheet before and after the test, and this amount of adsorption is used as an index of material separation in the concrete.

Furthermore, Patent Document 3 discloses a concrete evaluation method used when filling a structure with high-fluidity concrete. In this concrete evaluation method, a flow rate measuring device is used to measure the flow time and maximum shear strain rate of high-fluidity concrete when the mix ratio is changed. In addition, this concrete evaluation method has a first step of determining the concrete separation limit characteristic that indicates the relationship between the flow time and the maximum shear strain rate, assuming that the maximum shear strain rate is the separation limit point that indicates the material separation of concrete. In addition, this concrete evaluation method has a second step of performing numerical calculations and flow analysis based on the properties of the actual high-fluidity concrete that is actually filled in the structure and the shape of the structure, and calculating the actual maximum shear strain rate. In addition, this concrete evaluation method has a third step of determining the actual flow time of the actual high-fluidity concrete that is actually filled in the structure using the flow rate measuring device. This concrete evaluation method also includes a fourth step of comparing the distribution of the practical maximum shear strain rate determined in the second step during the practical flow time determined in the third step with the concrete separation limit characteristics obtained in the first step, and determining the risk of material separation in the practical high-flow concrete and the risk of blockage to the structure from the distribution of the practical maximum shear strain rate based on a preset judgment criterion.

JP 2003-322602 A Japanese Patent Application Laid-Open No. 09-021797 JP 2013-170833 A

However, the techniques described in Patent Documents 1 and 2 all require special testing equipment, which is problematic in that it is extremely time-consuming. The technique described in Patent Document 3 also requires flow analysis, which is problematic in that it is extremely time-consuming. Note that this problem does not only apply to high-flow concrete, but to concrete in general.

The present disclosure has been made in consideration of the above circumstances, and aims to provide an image region classification model creation device, a concrete evaluation device, and a concrete evaluation program that can evaluate the material separation resistance of concrete more easily than when special tests are performed.

The image region classification model creation device according to the present invention described in claim 1 includes an acquisition unit that acquires a slump flow image, which is an image of a slump flow of concrete, and identification information that enables identification of coarse aggregate regions and mortar regions in the slump flow image; an image processing unit that performs predetermined image processing on the slump flow image acquired by the acquisition unit to emphasize the characteristics of the coarse aggregate regions and the mortar regions; and a creation unit that creates an image region classification model by training a segmentation model in which the slump flow image image-processed by the image processing unit and the identification information corresponding to the slump flow image are used as learning data, the slump flow image is used as input information, and information classifying the coarse aggregate regions and the mortar regions in the slump flow image is used as output information.

According to the image region classification model creation device of the present invention described in claim 1, a slump flow image and identification information that enables identification of coarse aggregate regions and mortar regions in the slump flow image are acquired, and the acquired slump flow image is subjected to predetermined image processing that emphasizes the characteristics of the coarse aggregate regions and mortar regions. The image region classification model is created by training a segmentation model in which the image-processed slump flow image and identification information corresponding to the slump flow image are used as learning data, the slump flow image is used as input information, and information classifying the coarse aggregate regions and mortar regions in the slump flow image is used as output information. By using the image region classification model, it is possible to evaluate the material separation resistance of concrete more easily compared to the case of performing special tests, etc.

The image region classification model creation device according to the present invention described in claim 2 is the image region classification model creation device described in claim 1, in which the image processing is at least one of a process for reducing non-uniform illuminance, a noise removal process, and a contrast adjustment process.

According to the image region classification model creation device of the present invention described in claim 2, by performing image processing by at least one of a process for reducing non-uniform illuminance, a noise removal process, and a contrast adjustment process, it is possible to suppress a decrease in the performance of the image region classification model caused by at least one of the effects of non-uniform illuminance, noise, and contrast.

The image region classification model creation device according to the present invention described in claim 3 is the image region classification model creation device described in claim 1 or claim 2, in which the slump flow image is a grayscale image.

According to the image region classification model creation device of the present invention described in claim 3, by making the slump flow image a grayscale image, an image region classification model can be created more easily compared to when the slump flow image is a color image, and an image region classification model can be created with higher accuracy compared to when the slump flow image is a binary monochrome image.

The image region classification model creation device according to the present invention described in claim 4 is the image region classification model creation device described in any one of claims 1 to 3, in which the identification information is information created using the slump flow image that has been image-processed by the image processing unit.

According to the image region classification model creation device of the present invention described in claim 4, the identification information can be easily created by using a slump flow image that has been image-processed by an image processing unit.

The concrete evaluation device according to the present invention described in claim 5 includes an image processing unit that performs predetermined image processing on a slump flow image, which is an image of a slump flow of concrete to be evaluated, to emphasize the characteristics of coarse aggregate regions and mortar regions; an input unit that inputs the slump flow image that has been image-processed by the image processing unit to an image region classification model created by an image region classification model creation device described in any one of claims 1 to 4; a derivation unit that derives distribution state information showing the distribution state of coarse aggregate in the slump flow image using output information output from the image region classification model in response to the input of the slump flow image by the input unit; and an evaluation unit that evaluates the material separation resistance of the concrete to be evaluated using the distribution state information derived by the derivation unit.

According to the concrete evaluation device of the present invention described in claim 5, a slump flow image is subjected to predetermined image processing to emphasize the characteristics of the coarse aggregate area and the mortar area, the image-processed slump flow image is input to the image domain classification model created by the image domain classification model creation device of the present invention, and distribution state information showing the distribution state of the coarse aggregate in the slump flow image is derived using output information output from the image domain classification model by inputting the slump flow image, and the derived distribution state information is used to evaluate the material separation resistance of the concrete to be evaluated, making it possible to evaluate the material separation resistance of the concrete more easily than when special tests, etc. are performed.

The concrete evaluation device according to the present invention described in claim 6 is the concrete evaluation device described in claim 5, in which the image processing is at least one of a process for reducing non-uniform illuminance, a noise removal process, and a contrast adjustment process.

According to the concrete evaluation device of the present invention described in claim 6, by performing image processing using at least one of a process for reducing non-uniform illuminance, a noise removal process, and a contrast adjustment process, it is possible to input an appropriate slump flow image according to the applied process into the image region classification model, and as a result, it is possible to evaluate the material separation resistance of concrete with higher accuracy.

The concrete evaluation device according to the present invention described in claim 7 is the concrete evaluation device described in claim 5 or claim 6, in which the slump flow image is a grayscale image.

According to the concrete evaluation device of the present invention described in claim 7, by making the slump flow image a grayscale image, it is possible to evaluate the material separation resistance of concrete more easily compared to when the slump flow image is a color image, and it is possible to evaluate the material separation resistance of concrete with higher accuracy compared to when the slump flow image is a binary monochrome image.

The concrete evaluation program according to the present invention described in claim 8 performs predetermined image processing on a slump flow image, which is an image of the slump flow of the concrete to be evaluated, to emphasize the characteristics of the coarse aggregate area and the mortar area, inputs the slump flow image that has been subjected to the image processing to an image domain classification model created by an image domain classification model creation device described in any one of claims 1 to 4, derives distribution state information showing the distribution state of the coarse aggregate in the slump flow image using output information output from the image domain classification model in response to the input of the slump flow image, and evaluates the material separation resistance of the concrete to be evaluated using the derived distribution state information.

According to the concrete evaluation program of the present invention described in claim 8, a slump flow image is subjected to predetermined image processing to emphasize the characteristics of the coarse aggregate area and the mortar area, the image-processed slump flow image is input to the image domain classification model created by the image domain classification model creation device of the present invention, and distribution state information showing the distribution state of the coarse aggregate in the slump flow image is derived using output information output from the image domain classification model by inputting the slump flow image, and the derived distribution state information is used to evaluate the material separation resistance of the concrete to be evaluated, making it possible to evaluate the material separation resistance of the concrete more easily than when special tests, etc. are performed.

As explained above, the present invention makes it possible to evaluate the material separation resistance of concrete more easily than by conducting special tests, etc.

1 is a block diagram showing an example of a configuration of an information processing system according to an embodiment. 1 is a block diagram showing an example of a hardware configuration of an information processing device according to an embodiment. FIG. 2 is a block diagram showing an example of a functional configuration during learning of an image region classification model of the information processing device according to the embodiment. 1A to 1C are diagrams illustrating a learning method for an image region classification model by an information processing device according to an embodiment. FIG. 2 is a block diagram showing an example of a functional configuration of an image region classification model of the information processing device according to the embodiment when the model is in operation. FIG. 2 is a schematic diagram showing an example of a configuration of a learning image information database according to the embodiment; FIG. 2 is a schematic diagram showing an example of a configuration of a labeling information database according to the embodiment. 11 is a flowchart illustrating an example of a labeling information creation process according to the embodiment. FIG. 2 is a front view showing an example of a configuration of a labeling screen according to the embodiment. 11 is a flowchart illustrating an example of an image region classification model creation process according to the embodiment. 4 is a flowchart illustrating an example of a concrete evaluation process according to the embodiment. 1A to 1C are diagrams for explaining the concrete evaluation process according to the embodiment, and are diagrams showing an example of specific images at each stage and a coarse aggregate distribution graph. FIG. 13 is a front view showing an example of a configuration of an evaluation result screen according to the embodiment. 1A to 1C are diagrams for explaining the effect of the concrete evaluation processing according to the embodiment, and are diagrams showing an example of slump flow images before and after image processing, output results from an image region classification model, and a coarse aggregate distribution graph. FIG. 13 shows an example of a result of extraction of a coarse aggregate region by conventional edge extraction processing, in which the left figure shows an example of a slump flow image before image processing, and the right figure shows an example of a result of extraction of the coarse aggregate. FIG. 1 shows an example of the extraction result of a coarse aggregate region using a conventional watershed segmentation method, where the left figure shows an example of a slump flow image before image processing, and the right figure shows an example of the extraction result of coarse aggregate. FIG. 13 is a diagram for explaining the effect of the concrete evaluation process according to the embodiment, in which the left diagram shows an example of a learning image used to train an image region classification model, and the right diagram shows an example of an evaluation target image used in an actual evaluation. FIG. 17 is a diagram for explaining the effect of the concrete evaluation processing according to the embodiment, and is a diagram showing an example of a specific image when various image processing operations are selectively performed when the evaluation target image shown in FIG. 16 is used as an input image for an image region classification model.

Hereinafter, an example embodiment for carrying out the present invention will be described in detail with reference to the drawings. Note that in this embodiment, a case where an information processing device is applied as the image region classification model creation device and concrete evaluation device of the present invention will be described.

First, the configuration of the information processing system 1 according to this embodiment will be described with reference to FIG. 1. FIG. 1 is a block diagram showing an example of the configuration of the information processing system 1 according to this embodiment.

As shown in FIG. 1, the information processing system 1 according to this embodiment includes an information processing device 10, which plays a central role in this system, and an information storage device 90. The information processing device 10 according to this embodiment learns an image region classification model, which will be described later, and evaluates concrete using the image region classification model. The information storage device 90 according to this embodiment stores various information handled by the information processing device 10.

The information storage device 90 according to this embodiment includes a non-volatile memory unit 92. The memory unit 92 is realized by a hard disk drive (HDD), a solid state drive (SSD), a flash memory, or the like. The memory unit 92, which serves as a storage medium, stores a learning image information database 92A and a labeling information database 92B. The learning image information database 92A and the labeling information database 92B will be described in detail later.

The information processing device 10 and the information storage device 90 are connected via a network N, and the information processing device 10 can communicate with the information storage device 90 via the network N. In this embodiment, the network N is an in-company communication line such as a LAN (Local Area Network) or a WAN (Wide Area Network), but is not limited to this form. The network N may be a public communication line such as the Internet or a telephone line, or a combination of these in-company and public communication lines may be applied. In this embodiment, the network N is a wired communication line, but is not limited to this form. A wireless communication line may be applied, or a combination of wired and wireless communication lines may be applied.

Next, the hardware configuration of the information processing device 10 according to this embodiment will be described with reference to FIG. 2. FIG. 2 is a block diagram showing an example of the hardware configuration of the information processing device 10 according to this embodiment. Examples of the information processing device 10 include various computers such as a personal computer and a server computer.

The information processing device 10 according to this embodiment includes a CPU (Central Processing Unit) 11, a memory 12 as a temporary storage area, a non-volatile storage unit 13, an input unit 14 such as a keyboard and mouse, a display unit 15 such as a liquid crystal display, a media read/write device (R/W) 16, and a communication interface (I/F) unit 18. The CPU 11, memory 12, storage unit 13, input unit 14, display unit 15, media read/write device 16, and communication interface (I/F) unit 18 are connected to each other via a bus B. The media read/write device 16 reads information written to a recording medium 17 and writes information to the recording medium 17.

The storage unit 13 is realized by a HDD, SSD, flash memory, etc. The storage unit 13 as a storage medium stores a labeling information creation program 13A, an image area classification model creation program 13B, and a concrete evaluation program 13C. The labeling information creation program 13A is stored in the storage unit 13 by setting the recording medium 17 in which the labeling information creation program 13A is written to the medium reading and writing device 16, and the medium reading and writing device 16 reading the labeling information creation program 13A from the recording medium 17. The image area classification model creation program 13B is stored in the storage unit 13 by setting the recording medium 17 in which the image area classification model creation program 13B is written to the medium reading and writing device 16, and the medium reading and writing device 16 reading the image area classification model creation program 13B from the recording medium 17. Furthermore, the concrete evaluation program 13C is stored in the storage unit 13 by setting the recording medium 17 on which the concrete evaluation program 13C is written in the medium reading and writing device 16, and the medium reading and writing device 16 reading the concrete evaluation program 13C from the recording medium 17. The CPU 11 reads the labeling information creation program 13A, the image area classification model creation program 13B, and the concrete evaluation program 13C from the storage unit 13, expands them in the memory 12, and sequentially executes the processes of each program.

The storage unit 13 also stores an image region classification model 13D. The image region classification model 13D according to this embodiment is configured as a segmentation model in which a slump flow image, which is an image obtained by photographing the slump flow of concrete, is used as input information, and information classifying the coarse aggregate region and the mortar region in the slump flow image is used as output information.

In this embodiment, a model based on Fast Random Forest is applied as the segmentation model, but this is not limited to this. Other segmentation models such as a Semantic Segmentation model and an Instance-aware Segmentation model may also be applied as the image region classification model 13D.

The communication I/F unit 18 is also connected to a network N and to an imaging device 80 that captures slump flow images, and the information processing device 10 can input slump flow images from the imaging device 80. The imaging device 80 according to this embodiment captures color still images, but is not limited to this. An imaging device that captures grayscale still images may be used as the imaging device 80, or an imaging device that captures binary monochrome still images may be used as the imaging device 80. Also, an imaging device that captures moving images may be used as the imaging device 80.

Next, the functional configuration of the information processing device 10 according to this embodiment when learning the image region classification model 13D will be described with reference to Figures 3 and 4. Figure 3 is a block diagram showing an example of the functional configuration of the information processing device 10 according to this embodiment when learning the image region classification model 13D. Also, Figure 4 is a diagram provided for explaining the learning method of the image region classification model 13D by the information processing device 10 according to this embodiment.

As shown in FIG. 3, the information processing device 10 during learning of the image region classification model 13D includes an acquisition unit 11A, an image processing unit 11B, and a creation unit 11C. The CPU 11 of the information processing device 10 executes the image region classification model creation program 13B to function as the acquisition unit 11, the image processing unit 11B, and the creation unit 11C.

The acquisition unit 11A according to this embodiment acquires a slump flow image (hereinafter referred to as a "learning image") and identification information (in this embodiment, this is labeling information described later, hereinafter referred to as "labeling information") that enables identification of coarse aggregate areas and mortar areas in the learning image. In this embodiment, a grayscale image is applied as the learning image. Therefore, when using a slump flow image obtained by photographing with the photographing device 80 as the learning image, since the image is a color image, the information processing device 10 needs to convert the color image into a grayscale image. However, the learning image is not necessarily limited to a grayscale image, and for example, a color image or a binary monochrome image may be applied as the learning image.

The image processing unit 11B according to this embodiment performs predetermined image processing on the learning image acquired by the acquisition unit 11A to emphasize the characteristics of the coarse aggregate area and the mortar area. Note that, in this embodiment, all of the processes of reducing non-uniform illuminance, noise removal, and contrast adjustment are applied as the image processing, but this is not limited to this. A form in which one or a combination of two of these processes is applied as the image processing may also be used. Also, in this embodiment, a top hat filter process is applied as the process of reducing non-uniform illuminance, a Gaussian blur process is applied as the noise removal process, and a histogram equalization process and a histogram matching process are applied as the contrast adjustment process, but it goes without saying that this is not limited to this form.

The creation unit 11C according to this embodiment creates an image region classification model 13D by training a segmentation model in which the learning images that have been image-processed by the image processing unit 11B and the labeling information corresponding to the learning images are used as learning data, the slump flow image is used as input information, and information classifying the coarse aggregate regions and mortar regions in the slump flow image is used as output information.

In addition, in this embodiment, the labeling information is information created using a learning image that has been image-processed by the image processing unit 11B. In this embodiment, the learning image after image processing by the image processing unit 11B is displayed on the display unit 15. In this state, the user of the information processing device 10 uses the input unit 14 to label the coarse aggregate area and the mortar area with different values, thereby creating the labeling information. This allows the user of the information processing device 10 to label while referring to the learning image in which the characteristics of the coarse aggregate area and the mortar area are emphasized, and as a result, the labeling information can be created more accurately and easily.

That is, as an example, as shown in FIG. 4, the image processing unit 11B performs the above-mentioned image processing on the learning image 50A, thereby extracting a plurality of features 50B. Also, as described above, the learning image 50A that has been image-processed by the image processing unit 11B is used to obtain labeling information 50C corresponding to the learning image 50A. Then, the creation unit 11C uses the learning image 50A and the labeling information 50C after the image processing as learning data to train the above-mentioned segmentation model, thereby creating the image region classification model 13D.

Next, referring to FIG. 5, the functional configuration of the image region classification model 13D of the information processing device 10 according to this embodiment when in operation will be described. FIG. 5 is a block diagram showing an example of the functional configuration of the image region classification model 13D of the information processing device 10 according to this embodiment when in operation.

As shown in FIG. 5, when the image region classification model 13D is in operation, the information processing device 10 includes an image processing unit 11D, an input unit 11E, a derivation unit 11F, and an evaluation unit 11G. The CPU 11 of the information processing device 10 executes the concrete evaluation program 13C to function as the image processing unit 11D, the input unit 11E, the derivation unit 11F, and the evaluation unit 11G.

The image processing unit 11D according to this embodiment performs predetermined image processing for emphasizing the characteristics of the coarse aggregate area and the mortar area on a slump flow image (in this embodiment, a slump flow image obtained by photographing with the photographing device 80, hereinafter referred to as the "evaluation target image"), which is an image of the slump flow of the concrete to be evaluated. Note that in this embodiment, as the above-mentioned image processing, all of the processing for reducing non-uniform illuminance, the noise removal processing, and the contrast adjustment processing are applied as in the image processing by the image processing unit 11B, but this is not limited thereto. As in the image processing by the image processing unit 11B, one or a combination of two of these processing may be applied as the image processing. In this embodiment, as in the learning image, a grayscale image is applied as the evaluation target image, but this is not limited thereto. As in the learning image, for example, a color image or a binary monochrome image may be applied as the evaluation target image.

The input unit 11E according to this embodiment inputs the evaluation target image that has been subjected to image processing by the image processing unit 11D to the image region classification model 13D. By inputting the evaluation target image after this image processing, information for classifying the evaluation target image into coarse aggregate regions and mortar regions is output from the image region classification model 13D.

Therefore, the derivation unit 11F according to this embodiment uses the information output from the image region classification model 13D to derive distribution state information indicating the distribution state of the coarse aggregate in the image to be evaluated. Then, the evaluation unit 11G according to this embodiment uses the distribution state information derived by the derivation unit 11F to evaluate the material separation resistance of the concrete to be evaluated.

Next, the learning image information database 92A according to this embodiment will be described with reference to FIG. 6. As shown in FIG. 6, the learning image information database 92A according to this embodiment stores image IDs (identifications) and image information in association with each other.

The image ID is information that is assigned in advance as different information for each learning image 50A in order to individually identify the learning image 50A. The image information is image information that indicates the corresponding learning image 50A itself. Note that in this embodiment, an image obtained by photographing a learning slump flow by the photographing device 80 is used as the learning image 50A, but an image obtained by using a photographing device other than the photographing device 80 may also be used as the learning image 50A.

Next, the labeling information database 92B according to this embodiment will be described with reference to FIG. 7. As shown in FIG. 7, the labeling information database 92B according to this embodiment stores image IDs and labeling information in association with each other.

The image ID is the same as the image ID in the learning image information database 92A, and the labeling information is the same as the labeling information 50C described above. That is, the labeling information database 92B stores labeling information 50C corresponding to each of the learning images 50A registered in the learning image information database 92A in a one-to-one correspondence.

Next, the operation of the information processing device 10 according to this embodiment will be described with reference to Figures 8 to 13.

First, the operation of the information processing device 10 when executing the labeling information creation process will be described with reference to Figs. 8 and 9. Fig. 8 is a flowchart showing an example of the labeling information creation process according to this embodiment. Also, Fig. 9 is a front view showing an example of the configuration of the labeling screen according to this embodiment.

When a user inputs an instruction to start execution of the labeling information creation program 13A via the input unit 14, the CPU 11 of the information processing device 10 executes the labeling information creation program 13A, thereby executing the labeling information creation process shown in FIG. 8. Note that, in order to avoid confusion, a case will be described here where the learning image information database 92A has already been constructed.

In step 200 of FIG. 8, the CPU 11 reads out one set of image ID and image information (hereinafter referred to as the "learning image for labeling") from the learning image information database 92A. In step 202, the CPU 11 executes image processing by the image processing unit 11B described above on the read learning image for labeling.

In step 204, the CPU 11 controls the display unit 15 to display a labeling screen having a predetermined configuration using the labeling learning image that has undergone the above processing, and in step 206, the CPU 11 waits until the specified information is input.

Figure 9 shows an example of a labeling screen according to this embodiment. As shown in Figure 9, the labeling screen according to this embodiment displays a message encouraging the user to label the coarse aggregate and mortar, and also displays a learning image for labeling that has undergone the above-mentioned image processing.

When the labeling screen shown in FIG. 9 is displayed on the display unit 15 as an example, the user of the information processing device 10 inputs different values for the coarse aggregate area and the mortar area for the displayed learning image for labeling.

In this embodiment, the different values are input as "1" for the coarse aggregate area and "0" for the mortar area, but this is not limited to this. For example, the opposite may be true, where "0" is input for the coarse aggregate area and "1" is input for the mortar area. Also, for example, "A" may be input for the coarse aggregate area and "B" for the mortar area. In short, any form may be applied as long as different information is input for the coarse aggregate area and the mortar area.

In addition, in this embodiment, labeling of coarse aggregate regions and mortar regions is performed on a pixel-by-pixel basis in the learning image during labeling, but this is not limited to this. For example, labeling may be performed for each predetermined pixel group in the learning image during labeling.

When the user of the information processing device 10 finishes labeling the coarse aggregate region and the mortar region in the learning image during labeling, the user designates the end button 15A via the input unit 14. In response to this, step 206 becomes a positive judgment, and the process proceeds to step 208.

In step 208, the CPU 11 stores information indicating the labeling results for the labeling-time learning image obtained by the above processing in the labeling information database 92B together with the corresponding image ID as labeling information for the labeling-time learning image.

In step 210, the CPU 11 determines whether the above processing has been completed for all learning images 50A stored in the learning image information database 92A. If the determination is negative, the process returns to step 200, whereas if the determination is positive, the labeling information creation process ends. When repeatedly executing the processing of steps 200 to 210, the CPU 11 applies the learning images 50A that have not been processed up to that point as learning images during labeling.

The above labeling information creation process results in the construction of a labeling information database 92B, as shown in FIG. 7 as an example.

Next, the operation of the information processing device 10 when executing the image region classification model creation process will be described with reference to FIG. 10. FIG. 10 is a flowchart showing an example of the image region classification model creation process according to this embodiment.

When a user inputs an instruction to start execution of the image region classification model creation program 13B via the input unit 14, the CPU 11 of the information processing device 10 executes the image region classification model creation program 13B, thereby executing the image region classification model creation process shown in FIG. 10. Note that, in order to avoid confusion, a case will be described here where the learning image information database 92A and the labeling information database 92B have already been constructed.

10, the CPU 11 reads out a set of image information (hereinafter referred to as "learning image") and labeling information having a common image ID from the learning image information database 92A and the labeling information database 92B. In step 302, the CPU 11 performs a process of extracting only the slump flow image in a circular shape from the read learning image (hereinafter referred to as "region of interest extraction process"), and then performs image processing by the image processing unit 11B described above on the extracted circular learning image. Note that in this embodiment, as the region of interest extraction process, the learning image is divided into a circle inscribed in the outer periphery of the approximately circular slump flow region in the learning image, and only the inside of the circle is extracted, but this is not limited to this. For example, a circle that includes only the slump flow region in the learning image and has the largest diameter may be prepared in advance, and the circle may be used instead of the inscribed circle.

In step 304, the CPU 11 uses the learning image that has undergone the above processing as input information and the read labeling information as output information to train the image region classification model 13D.

In step 306, the CPU 11 determines whether the above processing has been completed for all learning images 50A stored in the learning image information database 92A. If the determination is negative, the process returns to step 300, whereas if the determination is positive, the image region classification model creation process ends. When repeatedly executing the processing from step 300 to step 306, the CPU 11 applies the learning images 50A that have not been processed up to that point as learning images during learning.

Next, the operation of the information processing device 10 when performing the concrete evaluation process will be described with reference to Figs. 11 to 13. Fig. 11 is a flowchart showing an example of the concrete evaluation process according to this embodiment. Fig. 12 is a diagram for explaining the concrete evaluation process according to this embodiment, showing specific images at each stage and an example of a coarse aggregate distribution graph. Furthermore, Fig. 13 is a front view showing an example of the configuration of the evaluation result screen according to this embodiment.

When a user inputs an instruction to start execution of the concrete evaluation program 13C via the input unit 14, the CPU 11 of the information processing device 10 executes the concrete evaluation program 13C, thereby executing the concrete evaluation process shown in FIG. 11. Note that, in order to avoid confusion, a case will be described here in which an evaluation target image showing a slump flow image that is the evaluation target and that has been converted to grayscale and obtained by photographing with the photographing device 80 is stored in the storage unit 13. Also, in order to avoid confusion, a case will be described here in which a learned image region classification model 13D has already been created.

In step 400 in FIG. 10, the CPU 11 reads out the evaluation target image from the storage unit 13. In step 402, the CPU 11 performs the above-described region of interest extraction process on the read out evaluation target image, and then executes image processing by the image processing unit 11D described above. By this region of interest extraction process and image processing, if the read out evaluation target image is evaluation target image 60A shown in FIG. 12 as an example, then evaluation target image 60B after image processing as shown in FIG. 12 as an example is obtained.

In step 404, the CPU 11 inputs the evaluation target image that has undergone the above processing to the image region classification model 13D. By inputting this evaluation target image, information in which the coarse aggregate region and the mortar region are labeled (hereinafter referred to as "image region classification information") 60C, as shown in FIG. 12 as an example, is output from the image region classification model 13D.

Therefore, in step 406, the CPU 11 acquires the image region classification information output from the image region classification model 13D. In step 408, the CPU 11 uses the acquired image region classification information to derive distribution state information indicating the distribution state of the coarse aggregate in the image to be evaluated. In this embodiment, the distribution state information is derived as follows.

First, as shown in FIG. 12 as an example, the CPU 11 assumes multiple concentric circles 60D (three in the example shown in FIG. 12) centered on a central point for the evaluation target image that has been made circular by the region of interest extraction process. Next, the CPU 11 calculates the ratio of the area of the coarse aggregate area to the total area of the divided area (hereinafter referred to as the "relative area ratio") for each of the multiple divided areas divided by each of the assumed concentric circles 60D. Then, the CPU 11 uses the calculated relative area ratio for each divided area to create a graph 60E shown in FIG. 12 as an example (hereinafter referred to as the "coarse aggregate distribution graph") as the distribution state information.

In step 410, the CPU 11 uses the distribution state information obtained by the process of step 408 to evaluate the material separation resistance of the concrete corresponding to the evaluation target image. In this embodiment, the evaluation of the material separation resistance of the concrete is performed by a process that derives information indicating which level of the material separation resistance of the concrete is located in a predetermined number of levels (10 levels in this embodiment) (hereinafter referred to as a "material separation resistance level value"). However, this is not limited to this form, and for example, instead of the material separation resistance level value, a process that derives an image that graphically shows the level of material separation resistance may be applied.

In step 412, the CPU 11 controls the display unit 15 to display a predetermined evaluation result screen using the coarse aggregate distribution graph 60E and the material separation resistance level value obtained by the above processing. In step 414, the CPU 11 waits until the specified information is input.

Figure 13 shows an example of an evaluation result screen according to this embodiment. As shown in Figure 13, the evaluation result screen according to this embodiment displays a coarse aggregate distribution graph 60E and a material separation resistance level value. Therefore, by referring to the evaluation result screen, a user of the information processing device 10 can easily understand the material separation resistance performance of the concrete being evaluated.

When the evaluation result screen shown in FIG. 13 is displayed on the display unit 15 as an example, the user of the information processing device 10 refers to the displayed coarse aggregate distribution graph 60E and material separation resistance level value, and then selects the end button 15A via the input unit 14. In response to this, step 414 becomes a positive judgment, and this concrete evaluation process is terminated.

Actual fresh tests were conducted on high-flow concrete, and tests on material separation resistance were conducted using the concrete evaluation process according to this embodiment for normal slump flow and slump flow with material separation. Note that here, the test for normal slump flow is referred to as test T1, and the test for slump flow with material separation is referred to as test T2.

The mixing conditions for the slump flow used in Test T1 and Test T2 are shown in Table 1. In Table 1, W represents the amount of water, C represents the amount of cement, and s/a represents the fine aggregate ratio.

Figure 14 is a diagram used to explain the effect of the concrete evaluation process according to this embodiment, showing an example of slump flow images before and after image processing, output results from an image region classification model, and a coarse aggregate distribution graph.

As shown in Figure 14, even though the image region classification model applied in this test was only trained twice, it was able to effectively distinguish between the coarse aggregate region and the mortar region for both Test T1 and Test T2. In addition, it can be easily seen from the coarse aggregate distribution graph that the distribution of coarse aggregate is approximately uniform (approximately flat) for Test T1, whereas the distribution of coarse aggregate is significantly biased from the center to the edge of the slump flow for Test T2. It is believed that by increasing the number of times the image region classification model is trained, it will be possible to more accurately determine material separation resistance.

On the other hand, FIG. 15A is a diagram showing an example of the extraction result of the coarse aggregate area using conventional edge extraction processing, where the left diagram shows an example of a slump flow image before image processing, and the right diagram shows an example of the coarse aggregate extraction result. As shown in FIG. 15A, in this case, water marks were also extracted in addition to the coarse aggregate area, and good results could not be obtained.

Figure 15B shows an example of the results of extracting coarse aggregate regions using the conventional watershed segmentation method, with the left figure showing an example of a slump flow image before image processing, and the right figure showing an example of the results of extracting coarse aggregate. As shown in Figure 15B, in this case, only a portion of the coarse aggregate region was extracted, and good results could not be obtained in this case either.

In this embodiment, therefore, all of the processes for reducing non-uniform illuminance, noise removal, and contrast adjustment are applied as image processing to the slump flow image to be input to the image region classification model 13D. The effects of these processes will be described below with reference to Figs. 16 and 17. Fig. 16 is a diagram for explaining the effects of the concrete evaluation process according to this embodiment, in which the left diagram shows an example of a learning image used to learn the image region classification model, and the right diagram shows an example of an evaluation target image used in the actual evaluation. Fig. 17 is a diagram for explaining the effects of the concrete evaluation process according to this embodiment, showing an example of a specific image when various image processes are selectively performed when the evaluation target image shown in Fig. 16 is used as an input image for the image region classification model.

As shown in FIG. 17(A), when no image processing was performed on the grayscale image of the slump flow image, it was difficult to extract the coarse aggregate area, as shown at the right end of the figure. Also, as shown in FIG. 17(B), even when a process for reducing non-uniform illuminance (here, processing using a top hat filter) and a process for removing noise (here, processing using a Gaussian blur filter) were performed on the grayscale image of the slump flow image, it was difficult to extract the coarse aggregate area, as shown at the right end of the figure.

In contrast, as shown in FIG. 17(C), when a contrast adjustment process (histogram equalization process in this case) was performed on the grayscale image of the slump flow image in addition to a process for reducing non-uniform illumination and a noise removal process, the coarse aggregate area could be extracted more clearly than when the contrast adjustment process was not performed.

Furthermore, as shown in FIG. 17(D), when a histogram matching process, which is a type of contrast adjustment process, was performed on the grayscale image of the slump flow image in addition to a process for reducing non-uniform illumination, a noise removal process, and a histogram equalization process, the area of coarse aggregate could be extracted more clearly compared to the case where the histogram matching process was not performed.

As described above, according to this embodiment, the apparatus includes an acquisition unit 11A that acquires a slump flow image, which is an image of a slump flow of concrete, and identification information (in this embodiment, labeling information) that enables identification of coarse aggregate regions and mortar regions in the slump flow image, an image processing unit 11B that performs predetermined image processing on the slump flow image acquired by the acquisition unit 11A to emphasize the characteristics of the coarse aggregate regions and the mortar regions, and a creation unit 11C that creates an image region classification model by training a segmentation model in which the slump flow image processed by the image processing unit 11B and the identification information corresponding to the slump flow image are used as learning data, the slump flow image is used as input information, and information that classifies the coarse aggregate regions and mortar regions in the slump flow image is used as output information. Therefore, by using the image region classification model, it is possible to evaluate the material separation resistance of concrete more easily than when special tests are performed.

In addition, according to this embodiment, the image processing performed by the image processing unit 11B includes all of the processes of reducing non-uniform illuminance, noise removal, and contrast adjustment. Therefore, it is possible to suppress the degradation of the performance of the image region classification model due to the influence of each of non-uniform illuminance, noise, and contrast.

In addition, according to this embodiment, a grayscale image is used as the slump flow image used to train the image region classification model. Therefore, compared to when the slump flow image is a color image, an image region classification model can be created more easily, and compared to when the slump flow image is a binary monochrome image, an image region classification model can be created with higher accuracy.

In addition, according to this embodiment, the identification information is information created using a slump flow image that has been image-processed by the image processing unit 11B. This makes it easier to create the identification information.

According to this embodiment, the apparatus includes an image processing unit 11D that performs predetermined image processing for emphasizing the characteristics of the coarse aggregate area and the mortar area on a slump flow image, which is an image of the slump flow of the concrete to be evaluated, an input unit 11E that inputs the slump flow image that has been image-processed by the image processing unit 11D to the created image area classification model, a derivation unit 11F that uses output information output from the image area classification model by inputting the slump flow image by the input unit 11E to derive distribution state information showing the distribution state of the coarse aggregate in the slump flow image, and an evaluation unit 11G that evaluates the material separation resistance of the concrete to be evaluated using the distribution state information derived by the derivation unit 11F. Therefore, the material separation resistance of the concrete can be evaluated more easily compared to the case of performing special tests, etc.

In addition, according to this embodiment, the image processing performed by the image processing unit 11D includes all of the processes of reducing non-uniform illuminance, noise removal, and contrast adjustment. Therefore, suitable slump flow images corresponding to these processes can be input to the image region classification model, and the material separation resistance of concrete can be evaluated with higher accuracy.

Furthermore, according to this embodiment, a grayscale image is used as the slump flow image used to evaluate the concrete. Therefore, compared to when the slump flow image is a color image, the material separation resistance of the concrete can be evaluated more easily, and compared to when the slump flow image is a binary monochrome image, the material separation resistance of the concrete can be evaluated with higher accuracy.

In the above embodiment, for example, the hardware structure of the processing unit that executes each process of the acquisition unit 11A, image processing unit 11B, creation unit 11C, image processing unit 11D, input unit 11E, derivation unit 11F, and evaluation unit 11G can be the various processors shown below. As described above, the various processors include a CPU, which is a general-purpose processor that executes software (programs) and functions as a processing unit, as well as a programmable logic device (PLD), which is a processor whose circuit configuration can be changed after manufacture, such as an FPGA (Field-Programmable Gate Array), and a dedicated electrical circuit, such as an ASIC (Application Specific Integrated Circuit), which is a processor with a circuit configuration designed specifically to execute a specific process.

The processing unit may be configured with one of these various processors, or may be configured with a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs, or a combination of a CPU and an FPGA). The processing unit may also be configured with a single processor.

As an example of configuring the processing unit as a single processor, first, there is a form in which one processor is configured as a combination of one or more CPUs and software, as typified by computers such as client and server, and this processor functions as the processing unit. Secondly, there is a form in which a processor is used that realizes the functions of the entire system, including the processing unit, on a single IC (Integrated Circuit) chip, as typified by systems on chips (SoCs). In this way, the processing unit is configured as a hardware structure using one or more of the various processors mentioned above.

More specifically, the hardware structure of these various processors can be an electrical circuit that combines circuit elements such as semiconductor elements.

10 Information processing device 11 CPU
Description of the Reference Signs 11A Acquisition unit 11B Image processing unit 11C Creation unit 11D Image processing unit 11E Input unit 11F Derivation unit 11G Evaluation unit 12 Memory 13 Storage unit 13A Labeling information creation program 13B Image area classification model creation program 13C Concrete evaluation program 13D Image area classification model 14 Input unit 15 Display unit 15A End button 16 Medium reading and writing device 17 Recording medium 18 Communication I/F unit 50A Learning image 50B Features 50C Labeling information 60A Evaluation target image 60B Evaluation target image 60C Image area classification information 60D Concentric circle 60E Coarse aggregate distribution graph 80 Photography device 90 Information storage device 92 Storage unit 92A Learning image information database 92B Labeling information database

Claims (8)

an acquisition unit that acquires a slump flow image, which is an image obtained by capturing a slump flow of concrete, and identification information that enables identification of a coarse aggregate region and a mortar region in the slump flow image;
an image processing unit that performs predetermined image processing on the slump flow image acquired by the acquisition unit to emphasize characteristics of the coarse aggregate region and the mortar region;
a creating unit that creates an image region classification model by training a segmentation model using the slump flow image that has been image-processed by the image processing unit and the identification information corresponding to the slump flow image as learning data, the slump flow image as input information, and information classifying the coarse aggregate region and the mortar region in the slump flow image as output information;
An image region classification model creation device comprising: The image processing is at least one of a process for reducing non-uniform illuminance, a noise removal process, and a contrast adjustment process.
The image region classification model generating device according to claim 1 . The slump flow image is a grayscale image.
3. The image region classification model creating device according to claim 1. The identification information is information created using the slump flow image that has been image-processed by the image processing unit.
The image region classification model creating device according to any one of claims 1 to 3. an image processing unit that performs predetermined image processing on a slump flow image, which is an image obtained by photographing the slump flow of concrete to be evaluated, to emphasize characteristics of a coarse aggregate region and a mortar region;
an input unit for inputting the slump flow image that has been subjected to image processing by the image processing unit to an image region classification model created by the image region classification model creation device according to any one of claims 1 to 4;
a derivation unit that derives distribution state information indicating a distribution state of coarse aggregate in the slump flow image by using output information output from the image region classification model in response to an input of the slump flow image by the input unit;
An evaluation unit that evaluates the material separation resistance of the concrete to be evaluated using the distribution state information derived by the derivation unit;
A concrete evaluation device equipped with: The image processing is at least one of a process for reducing non-uniform illuminance, a noise removal process, and a contrast adjustment process.
The concrete evaluation device according to claim 5. The slump flow image is a grayscale image.
The concrete evaluation device according to claim 5 or 6. A slump flow image is an image of the slump flow of the concrete to be evaluated, and predetermined image processing is performed to emphasize the characteristics of the coarse aggregate area and the mortar area.
The slump flow image that has been subjected to the image processing is input to an image region classification model created by the image region classification model creation device according to any one of claims 1 to 4,
deriving distribution state information indicating a distribution state of coarse aggregate in the slump flow image using output information output from the image region classification model in response to the input of the slump flow image;
Using the derived distribution state information, the material separation resistance of the concrete to be evaluated is evaluated.
A concrete evaluation program for running processing on a computer.