JP7431063B2 - Data collection program and evaluation program - Google Patents

Description

The present invention relates to a data collection program and an evaluation program.

Concrete structures generally deteriorate over time (material deterioration of concrete) due to the invasion of various deterioration factors from the outside. For this reason, how to improve the quality near the surface of a constructed concrete structure, such as the density and resistance to the intrusion of deteriorating factors (hereinafter referred to as surface quality), is important in ensuring and improving the durability of the entire structure. becomes extremely important.

As a method for evaluating the surface quality of concrete structures, which play such an important role, we visually observe the concrete surface after the formwork has been removed from the concrete poured into the formwork (hereinafter referred to as the demolded surface). There is a known technique for classifying and evaluating workmanship using four numerical values (for example, see Non-Patent Document 1).

Furthermore, a technique for evaluating the demolding surface of a concrete structure using machine learning is also known (see, for example, Patent Document 1).

Draft guide for ensuring the quality of concrete structures (bridge piers, abutments, box culverts, retaining walls) December 2015 Tohoku Regional Development Bureau, Ministry of Land, Infrastructure, Transport and Tourism

Japanese Patent Application Publication No. 2016-142601

The technology of Patent Document 1 uses reference information that has been machine learned in advance by associating an image of a concrete structure after demolding with an evaluation of the image by a skilled engineer, and captured image data. , which evaluates the construction quality of concrete structures.

However, in the technology described in Patent Document 1, a skilled engineer visually evaluates the captured image, so if the image is unclear or there is a shadow, the effects of Therefore, the evaluation may be inappropriate. As a result, the accuracy of evaluating the construction quality of concrete structures may be reduced.

In one aspect, the present invention provides a data collection program capable of collecting training data for accurately evaluating concrete structures, and an evaluation program capable of accurately evaluating concrete structures. The purpose is to

In one embodiment, the data collection program is a data collection program that collects training data for machine learning, and includes external appearance symptoms in a predetermined range of the surface of the concrete structure input into a terminal used by a certified expert. and the location information of the terminal at the time of the input, and obtain an image taken using the terminal after accepting the evaluation and the location information of the terminal at the time of the imaging. and when the position information of the terminal at the time of the input and the position information of the terminal at the time of the imaging are within a predetermined range, the evaluation and the image are associated with each other and used as training data. It is a program that is executed by a computer.

In another aspect, the data collection program is a data collection program that collects training data for machine learning, and includes evaluations of multiple items regarding appearance symptoms in a predetermined range of the surface of a concrete structure, and an image of the predetermined range. and, causing a computer to execute a process of acquiring teacher data with a different number of teacher data for each of the plurality of evaluation items, and using the acquired teacher data as teacher data to be used for the machine learning , The teacher data includes images of the same location captured under different imaging conditions .

In addition, the evaluation program performs machine learning on teacher data collected using the data collection program of the present application to generate an evaluation model for the surface of a concrete structure, and uses the generated evaluation model and the concrete structure to be evaluated. This is a program that causes a computer to execute a process of outputting an evaluation result regarding the surface of the concrete structure to be evaluated based on an image taken of the surface of the concrete structure.

According to the data collection program described in this specification, it is possible to collect training data for accurately evaluating concrete structures. Moreover, according to the evaluation program described in this specification, a concrete structure can be evaluated with high accuracy.

FIG. 1 is a schematic diagram showing an evaluation system according to an embodiment. FIG. 2(a) is a diagram showing the hardware configuration of the server, and FIG. 2(b) is a diagram showing the hardware configuration of the user terminal. FIG. 2 is a functional block diagram of a user terminal and a server. 3 is a flowchart illustrating a process of collecting teacher data by a user terminal. It is a diagram showing a state in which a concrete structure is divided into ranges. FIGS. 6(a) to 6(e) are diagrams for explaining types of evaluation items. FIG. 7(a) is a diagram showing a user authentication screen, and FIG. 7(b) is a diagram showing an error screen. FIG. 8(a) is a diagram showing an evaluation input screen, and FIG. 8(b) is a diagram showing a modified example of the evaluation input screen. It is a table showing the relationship between the evaluation score and the state of the demolded surface for five evaluation items. FIG. 3 is a diagram showing an image input screen. 3 is a flowchart showing evaluation model generation processing by the server. FIG. 2 is a diagram showing an example of a data structure of a teacher data DB. 7 is a flowchart showing evaluation processing using an evaluation model. FIG. 14(a) is a diagram showing an evaluation image capturing screen, and FIG. 14(b) is a diagram showing an evaluation result display screen. 15(a) and 15(b) are diagrams for explaining modification example 1. FIG. FIGS. 16(a) and 16(b) are diagrams for explaining modification example 2. FIG.

Hereinafter, one embodiment will be described in detail based on FIGS. 1 to 14.

FIG. 1 schematically shows the configuration of an evaluation system 100 according to an embodiment. Evaluation system 100 in FIG. 1 is a system for evaluating surface layer quality of concrete structures. The evaluation system 100 includes a server 10 and a user terminal 70, as shown in FIG. The server 10 and user terminal 70 are connected to a network 80 such as the Internet.

The server 10 acquires teacher data regarding concrete structures collected by the user terminal 70 and generates an evaluation model using machine learning. Further, the server 10 uses the generated evaluation model to analyze the image of the surface of the concrete structure (evaluation image) captured by the user terminal 70, and evaluates the surface quality.

FIG. 2(a) shows the hardware configuration of the server 10. As shown in FIG. 2A, the server 10 includes a CPU (Central Processing Unit) 90, a ROM (Read Only Memory) 92, a RAM (Random Access Memory) 94, and a storage unit (HDD (Hard Disk Drive) in this case). 96, a network interface 97, a portable storage medium drive 99, and the like. Each component of the server 10 is connected to a bus 98. In the server 10, the CPU 90 executes a program (including an evaluation program) stored in the ROM 92 or HDD 96, or a program (including an evaluation program) read from the portable storage medium 91 by the portable storage medium drive 99. As a result, the functions of each part shown in FIG. 3 are realized. Note that the functions of each part in FIG. 3 may be realized by, for example, an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

The user terminal 70 is a terminal such as a smartphone or a tablet terminal used by an engineer (hereinafter referred to as an expert) who is familiar with concrete technology such as a concrete inspector. The user terminal 70 receives an input of an evaluation of a concrete structure by an expert, and captures an image of a location corresponding to the evaluation in accordance with the expert's operation. Further, the user terminal 70 associates the evaluation with the image, generates teacher data, and transmits it to the server 10. Further, the user terminal 70 captures images of the concrete structure (evaluation images) in response to operations by experts and non-expert users, and transmits the captured evaluation images to the server 10 . Then, the evaluation result of the surface quality of the concrete structure transmitted from the server 10 is received and displayed.

FIG. 2(b) schematically shows the hardware configuration of the user terminal 70. As shown in FIG. 2(b), the user terminal 70 includes a CPU 190, a ROM 192, a RAM 194, a storage unit (HDD) 196, a network interface 197, a display unit 193, an input unit 195, a camera 187, and a drive for a portable storage medium. It is equipped with 199 mag. The display section 193 includes a liquid crystal display and the like, and the input section 195 includes a touch panel and the like. Each component of the user terminal 70 is connected to a bus 198. In the user terminal 70, the CPU 190 executes a program (including a data collection program) stored in the ROM 192 or HDD 196, or a program (including the data collection program) read from the portable storage medium 191 by the portable storage medium drive 199. By executing this, the functions of each part shown in FIG. 3 are realized.

FIG. 3 shows a functional block diagram of the user terminal 70 and the server 10. Hereinafter, the functions possessed by the user terminal 70 and the server 10 will be explained based on FIG. 3.

(About the functions of the user terminal 70)
When the CPU 190 executes the program, the user terminal 70 includes an authentication section 50, an evaluation acquisition section 52, an image acquisition section 54, a teacher data generation section 56, an evaluation image acquisition section 58, an evaluation image transmission section 60, and an evaluation acquisition section. 62.

The authentication unit 50 authenticates whether the user operating the user terminal 70 is an expert on concrete structures. At the time of this authentication, the authentication unit 50 refers to the expert DB 40. The expert DB 40 stores combinations of expert user names and passwords. The authentication unit 50 checks whether the user name and password input by the user match one of the user name and password combinations stored in the expert DB 40, and if they match, determines that the authentication is OK. When the authentication unit 50 determines that the authentication is OK, it notifies the evaluation acquisition unit 52 of the determination. Note that the authentication unit 50 may perform the above authentication by capturing an image of the user's face using the camera 187 and comparing the image with face image data stored in the expert DB 40. Further, the authentication unit 50 may perform the above authentication using biometric authentication using a fingerprint, an iris, or the like.

If the user is an expert, the evaluation acquisition unit 52 displays an evaluation input screen (see FIG. 8(a)) on the display unit 193 as a first screen for receiving input of evaluation results of concrete structures by the user. to be displayed. Then, when the user inputs an evaluation regarding a predetermined range of the concrete structure on the evaluation input screen, the input evaluation is acquired and transmitted to the teacher data generation section 56.

After the evaluation acquisition section 52 acquires the evaluation, the image acquisition section 54 displays an image input screen (see FIG. 10) as a second screen for capturing an image of the concrete structure on the display section 193. When the user images a predetermined range of the concrete structure using the image input screen, the image acquisition unit 54 acquires the captured image and transmits it to the teacher data generation unit 56.

The teacher data generation unit 56 generates teacher data by associating the evaluation regarding the predetermined range input by the user with the image captured of the predetermined range. The teacher data generation unit 56 transmits the generated teacher data to the teacher data acquisition unit 20 of the server 10. Note that the teacher data generation unit 56 may transmit the generated teacher data to the teacher data acquisition unit 20 each time it generates the teacher data, or it may send the generated teacher data to the teacher data acquisition unit 20 at the stage where a predetermined number of teacher data have been generated. The teacher data may be sent together to the teacher data acquisition unit 20.

The evaluation image acquisition unit 58 displays an evaluation image capture screen (see FIG. 14(a)) on the display unit 193 in response to a request from a user (who may or may not be an expert). indicate. Furthermore, when the user uses the evaluation image capture screen to image a predetermined range (range of evaluation target) of the concrete structure, the evaluation image acquisition unit 58 captures the captured image (evaluation image). It is acquired and transmitted to the evaluation image transmission section 60.

The evaluation image transmission section 60 transmits the evaluation image received from the evaluation image acquisition section 58 to the evaluation section 24 of the server 10 .

When the evaluation results are sent from the evaluation section 24 of the server 10, the evaluation acquisition section 62 acquires the evaluation results and displays them on the display section 193 (see the evaluation result display screen in FIG. 14(b)). .

(About the functions of the server 10)
The server 10 functions as a teacher data acquisition section 20, a model generation section 22, and an evaluation section 24 when the CPU 90 executes a program.

The teacher data acquisition unit 20 acquires the teacher data transmitted from the teacher data generation unit 56 and stores it in the teacher data DB 30. Here, the teacher data DB 30 stores images and evaluations corresponding to the images in association with each other.

When a predetermined number of teacher data are stored in the teacher data DB 30, the model generation unit 22 performs machine learning using the teacher data to generate an evaluation model. When an evaluation image of a concrete structure is input to this evaluation model, the range of the concrete structure shown in the evaluation image can be evaluated. The model generation section 22 passes the generated model to the evaluation section 24.

The evaluation unit 24 uses the evaluation model generated by the model generation unit 22 to evaluate the range of the concrete structure shown in the evaluation image. The evaluation unit 24 transmits the evaluation results to the evaluation acquisition unit 62 of the user terminal 70.

(Teacher data collection process by user terminal 70)
Next, the teacher data collection process by the user terminal 70 will be described in detail along the flowchart of FIG. 4 and with appropriate reference to other drawings. Note that the process in FIG. 4 is a process that is executed when the expert performs an operation to input teacher data via the input unit 195 of the user terminal 70. It is assumed that the expert is located in front of the concrete structure, is in a state where he can evaluate the concrete structure, and is also in a state where he can take an image of the concrete structure. For example, when there is a concrete structure (for example, a concrete column) as shown in FIG. 5, the user divides the concrete structure into rectangular areas (A11, A12, A13, . . . ) of 1 to 2 m square. The expert then evaluates each range on five evaluation items. Here, the five evaluation items are surface bubbles, sand streaks, overlapping lines, slag leakage at formwork joints, and sinking cracks, as shown in FIGS. 6(a) to 6(e).

In the process of FIG. 4, first, in step S10, the authentication unit 50 performs user authentication. Specifically, the authentication unit 50 displays a user authentication screen as shown in FIG. 7(a) on the display unit 193. This user authentication screen is provided with fields for entering a user name and password, and a "send" button. The user enters the user name and password on the user authentication screen and presses the "Send" button. In this case, the authentication unit 50 obtains the input user name and password, refers to the expert DB 40, and determines whether the obtained combination of user name and password is stored in the expert DB 40. If the combination of the acquired user name and password is stored in the expert DB 40, the authentication is OK, and if the combination is not stored, the authentication is NG. Note that user authentication may be realized by a method other than the method using a user ID and password. For example, the user authentication may be performed using biometric authentication using fingerprints, iris, etc., as described above, or may be performed using facial recognition.

Next, in step S12, the authentication unit 50 determines whether or not the authentication was successful, that is, whether the authentication was OK. If the determination in step S12 is negative, the process moves to step S14, and the authentication section 50 displays an error screen as shown in FIG. 7(b) on the display section 193. After that, all the processing in FIG. 4 ends.

On the other hand, if the determination in step S12 is affirmative, that is, if the user is an expert, the process moves to step S16, and the evaluation acquisition unit 52 displays an evaluation input screen. The evaluation input screen is a screen as shown in FIG. 8(a). On the screen shown in Figure 8(a), for five evaluation items (surface bubbles, sand streaks, overlapping lines, slag leakage at formwork joints, sinking cracks), a score of "4.0" and a score of "3.0" were obtained. ” point, “2.0” point, or “1.0” point can be input. Note that the expert may evaluate multiple evaluation items or only one evaluation item on the evaluation input screen shown in FIG. 8(a).

FIG. 9 is a table showing the relationship between the evaluation points and the condition of the demolded surface for the five evaluation items described above. In FIG. 9, the condition of the demolding surface is classified into four levels from 4 points to 1 score, and the higher the score, the more beautifully finished the demolding surface is. An expert who observes and evaluates the demolding surface can evaluate the demolding surface based on the table of FIG. However, the relationship between the condition of the demolded surface and the evaluation score in Figure 9 is generally "good = 4 points", "fair = 3 points", "slightly bad = 2 points", and "poor = 1 point". Experts may give an intuitive evaluation, such as giving a score of 4 if they think the demolded surface is beautiful, or 3 if they think the demolding surface is of an ordinary level. Note that the relationship between the condition of the demolding surface and the evaluation score may be reversed (the better the score, the lower the score).

Note that by setting the evaluation input screen as shown in FIG. 8(b), the expert may be allowed to evaluate in 0.5 point increments within the range of 4 points to 1 point. This takes into account the fact that even when an engineer familiar with concrete technology evaluates, there are concrete surface conditions located at the boundaries of each evaluation point. Moreover, the table of FIG. 9 is an example, and the evaluation may be in other than four levels. Furthermore, the words and phrases associated with the evaluation scores may be other than "good, average, bad". For example, words that can distinguish the quality and beauty of the demolding surface of a concrete structure may be used, such as "less, average, more" or "clean, average, dirty."

Returning to FIG. 4, in step S18, the evaluation acquisition unit 52 waits until the "register" button is pressed by the expert. When the expert inputs the evaluation of a predetermined range of the concrete structure (for example, range A11 in FIG. 5) on the evaluation input screen of FIG. 8(a) and presses the "register" button (step S18: affirmative), the evaluation is obtained. The unit 52 moves to step S20 and acquires the evaluation input by the expert. Note that the evaluation acquisition unit 52 transmits the acquired evaluation information to the teacher data generation unit 56. Here, when the expert is operating the user terminal 70, position information can be acquired and recorded using the GNSS (Global Navigation Satellite System) built into the user terminal 70. preferable. Furthermore, if position information is acquired and recorded using GNSS when acquiring an image in step S26, which will be described later, it can be determined whether the location evaluated by the expert matches the location where the image was taken. . In other words, it can be proven that an expert evaluated the actual concrete structure by looking at it at the construction site. At this time, the evaluation time, imaging time, etc. may be recorded using a clock function built into the user terminal 70.

Next, in step S22, the image acquisition unit 54 displays an image input screen. The image input screen is a screen as shown in FIG. 10, and is provided with a frame for displaying a live view image captured by the camera 187 and an "imaging" button. The expert presses the "imaging" button while displaying the range of the concrete structure evaluated immediately before (for example, range A11 in FIG. 5) within the frame.

Next, in step S24, the image acquisition unit 54 waits until the "imaging" button is pressed by the expert. When the expert presses the "imaging" button (step S24: affirmative), the image acquisition unit 54 executes imaging with the camera 187 in step S26, and captures the evaluated range (for example, range A11 in FIG. 5). Get the image. Note that the image acquisition unit 54 transmits the acquired image to the teacher data generation unit 56.

Next, in step S28, the teacher data generation unit 56 generates teacher data by associating the evaluation with the image. Here, it is assumed that the expert inputs a score of "2.0" for the evaluation item "sand streaks" on the evaluation input screen shown in FIG. 8(a), but does not input evaluation points for other evaluation items. . In this case, the teacher data generation unit 56 generates teacher data by associating the image with the evaluation item "sand streaks" = evaluation score "2.0". On the other hand, the teacher data generation unit 56 does not generate teacher data for other evaluation items. In addition, on the evaluation input screen, there are cases where a score of "2.0" is inputted into the evaluation item "sand streaks" and a point "1.0" is inputted into the evaluation item "stacked lines." In such a case, the teacher data generation unit 56 generates teacher data by associating the image with the evaluation item "sand streaks" = evaluation score "2.0", and also generates the teacher data by associating the image with the evaluation item "sand streaks" = evaluation score The teacher data may be generated by associating an image with the point "1.0".

Next, in step S30, the teacher data generation unit 56 transmits the teacher data to the teacher data acquisition unit 20 of the server 10. Note that the teacher data may be transmitted when a predetermined number of teacher data have been collected, or when all the processes in FIG. 4 have been completed.

Next, in step S32, the teacher data generation unit 56 determines whether or not the process is finished. For example, if the expert inputs an instruction to end, the determination in step S32 is affirmative, and the entire process in FIG. 4 ends. On the other hand, if the expert inputs a command to continue, the determination in step S32 is negative, the process returns to step S16, and the processes and determinations in steps S16 to S32 described above are repeatedly executed.

(Evaluation model generation process by server 10)
Next, the evaluation model generation process by the server 10 will be described along the flowchart of FIG. 11.

In the process of FIG. 11, in step S50, the teacher data acquisition unit 20 waits until teacher data is transmitted from the user terminal 70. When the teacher data is transmitted from the user terminal 70 (step S50: affirmative), the process moves to step S52, and the teacher data acquisition unit 20 stores the received teacher data in the teacher data DB 30. Here, as an example, as shown in FIG. 12, images are stored in the teacher data DB 30 in association with evaluation points for each evaluation item. Although FIG. 12 shows that one image is stored for each evaluation item and evaluation point, in reality, a large number of images are stored.

Next, in step S54 of FIG. 11, the teacher data acquisition unit 20 determines whether the number of evaluations (that is, the number of images) for each evaluation item has reached the required number. If the determination at step S54 is negative, the process returns to step S50. On the other hand, if the determination in step S54 is affirmative, the process moves to step S56.

Here, in this embodiment, the required number of teacher data for each evaluation score of "4 points", "3 points", "2 points", and "1 point" of each evaluation item is at least 100 data, preferably 500 data. More than 1,000 pieces of data, preferably 1,000 pieces of data or more, should be collected. By collecting such a large amount of training data, it is possible to generate an evaluation model that can accurately evaluate the condition of a concrete surface that lacks feature quantities.

Furthermore, in this embodiment, the required number of teacher data may be different for each evaluation item. For example, in the case of the evaluation item ``surface bubbles'', areas with bubbles in the image tend to appear blackish, while areas without bubbles tend to appear whitish, and the presence or absence of bubbles can be determined by this difference in black and white. Therefore, it is a visual symptom that is relatively easy to evaluate among the five evaluation items. Similarly, in the case of the evaluation item ``Leakage from formwork joints,'' a seam without any leakage is a thin straight line, and a seam with a leakage will have symptoms such as a difference in color around the straight line, so it is not a thin straight line. It is a visual symptom that is relatively easy to evaluate if you pay attention to whether or not it is present. On the other hand, the evaluation items ``sand streaks'' and ``overlapping lines'' are difficult to distinguish from the ``color unevenness'' frequently observed on the demolded surface, and are the appearance symptoms that are difficult to evaluate among the five evaluation items. Regarding the evaluation item "sinking and cracking," although minor symptoms are observed relatively frequently, symptoms equivalent to a score of "1" or "2" rarely occur, so a large amount of data is needed. The appearance symptoms are difficult to collect.

Therefore, in this embodiment, the required number D1 of training data for the evaluation item "Surface bubbles" and the evaluation item "Slag leakage at formwork joints" is the same number, and the evaluation item "Sand streaks" and the evaluation item "Overlapping line '' can be made larger than D1 in consideration of the difficulty of evaluation. Further, the required number D3 of training data for the evaluation item "sinking cracks" can be made smaller than D1 in consideration of the difficulty of data collection. Note that the required number D1 of the teacher data for "surface bubbles" and "leakage at formwork joints" does not necessarily have to be the same number, and it is sufficient that they fall within a certain range. Similarly, data number ranges may be set for each of the required number D2 and the required number D3.

In step S56, the model generation unit 22 performs machine learning using the teacher data stored in the teacher data DB 30 to generate an evaluation model. Machine learning can be, for example, supervised learning using a neural network using a large amount of training data. Furthermore, existing techniques such as so-called deep learning can be used for machine learning. After generating the evaluation model, the model generation section 22 transmits the generated evaluation model to the evaluation section 24 . With the above, the entire process in FIG. 11 is completed.

(Evaluation processing using evaluation model)
Next, evaluation processing using the evaluation model will be described in detail along the flowchart of FIG. 13. FIG. 13 shows parallel processing by the user terminal 70 and the server 10. The process in FIG. 13 is a process for evaluating the image captured by the user terminal 70 using the evaluation model generated in the process in FIG.

In the process of FIG. 13, first, in step S70, the evaluation image acquisition unit 58 of the user terminal 70 waits until there is a request from the user to capture an evaluation image. Note that the user in this case may be an expert or a non-expert. That is, it is not necessary to perform user authentication before the process shown in FIG. 13 is performed.

When a user inputs a request to capture an evaluation image, the process moves to step S72, and the evaluation image acquisition unit 58 displays an evaluation image capture screen on the display unit 193. It is assumed that the evaluation image capturing screen is a screen as shown in FIG. 14(a). The evaluation image capturing screen in FIG. 14A is provided with a frame for displaying a live view image captured by the camera 187 and an "imaging" button. The user presses the "imaging" button with the range of the concrete structure to be evaluated displayed within the frame.

Next, in step S74, the evaluation image acquisition unit 58 waits until the user presses the "imaging" button. When the user presses the "imaging" button (step S74: affirmative), the evaluation image acquisition unit 58 executes imaging with the camera 187 in step S76, and acquires an image of the range that the user wants to evaluate. Note that the evaluation image acquisition section 58 delivers the acquired image to the evaluation image transmission section 60.

Next, in step S78, the evaluation image transmitting unit 60 transmits the captured image (referred to as an evaluation image) to the evaluation unit 24 of the server 10.

On the other hand, on the server 10 side, in step S170, the evaluation unit 24 waits until it receives the evaluation image. Therefore, when the process of step S78 is executed on the user terminal 70 side, the process moves to step S172, and the evaluation unit 24 evaluates the evaluation image using the evaluation model. The evaluation unit 24 inputs the evaluation image to an evaluation model, and obtains evaluation points output from the evaluation model.

Next, in step S174, the evaluation unit 24 transmits the evaluation result (evaluation score) to the user terminal 70.

On the other hand, on the user terminal 70 side, the evaluation acquisition unit 62 waits until the evaluation results are transmitted from the server 10 in step S80. Therefore, when the process of step S174 is executed on the server 10 side, the process moves to step S82, and the evaluation acquisition section 62 acquires the evaluation result and displays it on the display section 193. In this case, the evaluation acquisition unit 62 displays an evaluation result display screen as shown in FIG. 14(b), for example. The evaluation result display screen in FIG. 14(b) shows the evaluation results (2 points) and an evaluation image.

With the above processing, all the processing in FIG. 13 is completed.

As described above in detail, according to the present embodiment, the evaluation acquisition unit 52 receives input of the evaluation of appearance symptoms in a predetermined range of the surface of the concrete structure, and the image acquisition unit 54 After obtaining the evaluation, an image of a predetermined range is obtained. Then, the teacher data generation unit 56 associates the evaluation result with the image and sets it as teacher data. That is, in this embodiment, at a concrete structure construction site, an expert evaluates a predetermined range in front of the actual concrete structure, and then images the predetermined range. Here, since the concrete structure has a gray color that reflects the color of cement and a smooth surface, it is difficult to accurately evaluate it simply by observing the image because there are few features. In particular, the concrete poured into the formwork is darkened immediately after the formwork is removed, and it is difficult to accurately evaluate the surface condition by observing images. In contrast, in this embodiment, accurate evaluation can be made by visually evaluating the surface condition of the concrete structure, so by using the evaluation, accurate teacher data can be collected. . Therefore, in this embodiment, a highly accurate evaluation model can be generated using accurate teaching data. Thereby, by inputting the evaluation image into the evaluation model, it is possible to accurately evaluate the surface of the concrete structure shown in the evaluation image. In this case, since the evaluation model is a model that reflects the knowledge of an expert, it is possible to obtain an evaluation similar to that of an expert from an evaluation image taken by a user other than an expert.

Further, in this embodiment, the user terminal 70 displays the evaluation input screen (FIG. 8(a)) to accept the evaluation input, and then displays the image input screen (FIG. 10) to input the image. get. By making screen transitions in this way, experts are unable to evaluate the captured image, so they will inevitably evaluate the actual object.

Furthermore, in this embodiment, since it is determined whether the user is an expert through user authentication processing, it is possible to prevent evaluation results by non-experts from being included in the teacher data.

Further, in the present embodiment, the teacher data acquisition unit 20 performs an evaluation on predetermined evaluation items regarding appearance symptoms observed on the surface of the concrete structure in a predetermined range of the concrete structure, and an image of the evaluated range. (S52 in FIG. 11). Then, when the necessary number of teacher data set for each evaluation item can be acquired, the acquired teacher data is used as the teacher data to be used for the machine learning (S54 in FIG. 11: affirmative, S56). In other words, the required number of teacher data can be changed for each evaluation item. Thereby, the teacher data required for machine learning can be appropriately collected, taking into account the accuracy of machine learning, the ease of collecting teacher data, etc.

In the above embodiment, the evaluation items were surface bubbles, sand streaks, overlapping lines, slag leakage at formwork joints, and sinking cracks among the appearance symptoms observed on the surface of concrete structures. However, other appearance symptoms such as temperature cracks, pockmarks, strips, flow stripes, construction stripes, cold joints, uneven color, and surface peeling may also be included in the evaluation items. That is, in the above embodiment, the evaluation items were external symptoms that were not due to construction defects (surface bubbles, sand streaks, overlapping lines, slag leakage at formwork joints, sinking cracks), but external appearance symptoms that were due to construction defects were evaluated as evaluation items. may be included in When the appearance symptoms of poor construction are used as evaluation items, the evaluation points may be defined as minor, moderate, severe, etc.

(Modification 1)
Concrete structures generally undergo material deterioration over a period of several years to several decades due to the invasion of various deterioration factors from the outside. When such deterioration occurs over time, it also appears as a change in the appearance of the surface of the concrete structure.

Therefore, for example, as shown in FIG. 15(a), for the same range of the surface of a concrete structure (for example, range ID=A11 in FIG. 5), immediately after construction, one year later, two years later, etc. Evaluation and imaging may be performed at different times, and the evaluation and images at each time may be associated with the evaluation of aging deterioration (degree of aging deterioration), and this may be used as training data.

In this case, an evaluation model is generated by machine learning changes in images, changes in evaluation, and degree of deterioration over time. For example, as shown in FIG. 15(b), by inputting the evaluation image and the time when the evaluation image was taken into the evaluation model, it is possible to determine the degree of future deterioration over time and the It is possible to output information such as what the state will be (predicted image) and what the evaluation will be after a predetermined period of time (evaluation prediction).

(Modification 2)
In addition, as in Modification 1 above, when evaluating and imaging the same area of the surface of a concrete structure at different times, such as immediately after construction, one year later, two years later, etc., the expert must Furthermore, technical opinions regarding maintenance other than evaluation may be input. For example, when the evaluation score for a certain evaluation item decreases by one point (for example, when the evaluation score changes from 4 to 3), the expert inputs a technical opinion such as "observation enhancement". In addition, experts may decide to "conduct a detailed investigation" if the score for a certain evaluation item drops by 2 points (for example, if the score changes from 4 to 2), or if it drops by 3 points (for example, if the score changes from 4 to 2). If the number changes from 4 to 1), enter a technical opinion such as "implement emergency response".

In this case, as shown in FIG. 16(a), data in which changes in images, changes in evaluation, and technical opinions are associated is used as training data. Then, an evaluation model is generated by performing machine learning using training data as shown in FIG. 16(a). For example, as shown in FIG. 16(b), two evaluation images taken of the same range at different times can be input to this evaluation model. By doing so, it becomes possible to obtain technical opinions regarding the relevant scope (intensified observation, implementation of detailed investigation, implementation of emergency response, etc.).

In the above embodiment, the case where the image input screen (FIG. 10) is displayed after the evaluation input screen (FIG. 8(a)) is displayed is not limited to this. Each screen may be displayed as appropriate depending on the request. In this case, the teacher data generation unit 56 inputs the date and time and location information when the evaluation was input on the evaluation input screen (FIG. 8(a)) and the date and time and location information when the image was captured using the image input screen (FIG. 10). If the position where the evaluation was input and the position where the image was captured are similar, and the date and time when the evaluation was input is immediately before the date and time when the image was captured, the teacher Data may also be generated. Even in this case, as in the above embodiment, it is possible to generate an evaluation model from teacher data that is a combination of an image and an accurate evaluation result evaluated by visually observing the actual object.

Note that in the above embodiment, a case has been described in which the evaluation system 100 includes the server 10 and the user terminal 70; however, the present invention is not limited to this, and the user terminal 70 may perform all the processing described in the above embodiment. good.

Note that the above processing functions can be realized by a computer. In that case, a program is provided that describes the processing contents of the functions that the processing device should have. By executing the program on the computer, the above processing functions are realized on the computer. A program describing the processing contents can be recorded on a computer-readable storage medium (excluding carrier waves).

When a program is distributed, it is sold in the form of a portable storage medium such as a DVD (Digital Versatile Disc) or a CD-ROM (Compact Disc Read Only Memory) on which the program is recorded. It is also possible to store the program in the storage device of the server computer and transfer the program from the server computer to another computer via a network.

A computer that executes a program stores, for example, a program recorded on a portable storage medium or a program transferred from a server computer in its own storage device. Then, the computer reads the program from its own storage device and executes processing according to the program. Note that a computer can also directly read a program from a portable storage medium and execute processing according to the program. Furthermore, each time a program is transferred from the server computer, the computer can sequentially execute processing according to the received program.

The embodiments described above are examples of preferred implementations of the present invention. However, the present invention is not limited to this, and various modifications can be made without departing from the gist of the present invention. For example, in order to satisfy the required number D2 of training data for the evaluation item "sand streaks" and the evaluation item "overlap line", and/or to satisfy the required number D3 of training data for the evaluation item "sinking cracks", The same location may be imaged multiple times. In this case, it is preferable to take an image while changing the imaging conditions (angle, zoom magnification, etc.). In addition, in order to satisfy the required number D2 of training data for the evaluation item "sand streaks" and the evaluation item "overlap line", and/or to satisfy the required number D3 of training data for the evaluation item "sinking cracks" For these images, compared to the training data images for the evaluation item "Surface bubbles" and the evaluation item "Slag leakage at formwork joints", the compression ratio of the JPEG image was lowered and the image was taken in so-called fine mode. You may also do so. Thereby, teacher data can be collected efficiently.

10 Server 20 Teacher data acquisition unit 22 Model generation unit 24 Evaluation unit 50 Authentication unit 52 Evaluation acquisition unit 54 Image acquisition unit 56 Teacher data generation unit 70 User terminal

Claims (10)

A data collection program that collects training data for machine learning,
Accepting the evaluation of external appearance symptoms in a predetermined range of the surface of a concrete structure input into a terminal used by a certified expert and the location information of the terminal at the time of the input ,
acquiring an image captured using the terminal after accepting the evaluation and position information of the terminal at the time of the imaging ;
If the position information of the terminal at the time of the input and the position information of the terminal at the time of the imaging are within a predetermined range, the evaluation and the image are associated with each other and used as training data;
A data collection program that causes a computer to perform processing. displaying on the terminal a first screen that accepts the input of the evaluation, and executing a process of accepting the input;
After the evaluation is input on the first screen, a second screen for capturing the image is displayed on the terminal , and a process for acquiring the image is executed.
The data collection program according to claim 1. The evaluation includes evaluation of at least one of the appearance symptoms on the surface of the concrete structure, such as surface bubbles, sand streaks, overlapping lines, slag leakage at formwork joints, and sinking cracks. The data collection program described in 2. 4. The data collection program according to claim 1, wherein the expert is a concrete diagnostician. A data collection program that collects training data for machine learning,
Obtaining training data in which evaluations of a plurality of items regarding appearance symptoms in a predetermined range of the surface of a concrete structure are associated with images of the predetermined range, with different numbers of training data for each of the plurality of evaluation items, using the acquired teacher data as teacher data used for the machine learning;
Let the computer carry out the process ,
The teacher data includes images of the same location captured under different imaging conditions.
Data collection program. The evaluation of the plurality of items includes surface bubbles, sand streaks, overlapping lines, and leakage at formwork joints among the appearance symptoms on the surface of the concrete structure,
5. The number of training data regarding at least one of the sand streaks and the overlapping line is larger than the number of training data regarding at least one of the surface bubbles and slag leakage at the formwork joints. The data collection program described in . The evaluation of the plurality of items includes appearance symptoms on the surface of the concrete structure, such as surface bubbles, slag leakage from formwork joints, and sinking cracks,
7. The data collection program according to claim 5, wherein the number of training data regarding the sinking cracks is smaller than the number of training data regarding at least one of the surface bubbles and the leakage at the formwork joints. Machine learning is performed on training data collected using the data collection program according to any one of claims 1 to 7 to generate an evaluation model for the surface of a concrete structure,
outputting an evaluation result regarding the surface of the concrete structure to be evaluated based on the generated evaluation model and an image taken of the surface of the concrete structure to be evaluated;
An evaluation program that allows a computer to perform processing. In the process of generating the evaluation model, the evaluation model is generated using training data about the same range collected at different timings,
In the outputting process, a future change of the concrete structure to be evaluated is estimated and output based on the generated evaluation model and an image captured of the surface of the concrete structure to be evaluated. The evaluation program according to claim 8, characterized by: In the process of generating the evaluation model, the evaluation model is generated using teacher data about the same range collected at different timings and input opinions regarding maintenance management,
In the outputting process, an opinion regarding the maintenance and management of the concrete structure to be evaluated is estimated based on the generated evaluation model and images of the surface of the concrete structure to be evaluated taken at different timings. 10. The evaluation program according to claim 9, wherein the evaluation program outputs the evaluation program.

Images