Property determination method and learning image capture device

Abstract

To provide a property determination method and a learning image capture device that can accurately determine the properties of excavated soil (e.g., mud soil) at a start of shield tunneling.SOLUTION: A property determination method for determining properties of excavated soil discharged from a shield tunneling machine includes a simulated soil preparation process S11 for producing simulated soil that simulates the properties of the excavated soil, a training data preparation process S12 for preparing a dataset of training data consisting of image data of the simulated soil and property information related to the properties of the simulated soil, a learning process S13 for training the learning model by using the dataset to output property information by inputting image data of the excavated soil into the learning model, and a determination process S30 for determining the property information of the excavated soil from the image data of the excavated soil by using the trained learning model.SELECTED DRAWING: Figure 5

Description

[Technical field]

[0001] The present invention relates to a property determination method and a learning image capturing device. [Background technology]

[0002] One of the shield methods for constructing tunnels using a shield machine is the earth pressure shield method. The shield machine used in these methods has a cutter head, a chamber into which the soil (excavated soil) excavated by the cutter head is taken, and a screw conveyor that discharges the mud in the chamber (soil with added mud additives to give it impermeability and plastic flowability) to the rear. In this method, the soil excavated by the cutter head is taken into the chamber, filling it, and the soil is discharged onto the conveyor via the screw conveyor while stabilizing the face using the earth pressure in the chamber. Therefore, in earth pressure shield tunneling methods, the mud in the chamber must maintain an appropriate plastic fluidity in order to stabilize the tunnel face, and it is important to understand the properties of the mud in question when managing tunnel construction. In relation to this, there is a technique for determining the fluidity of mud discharged from a shield tunneling machine using machine learning technology (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2021-116529 A Summary of the Invention [Problem to be solved by the invention]

[0004] In the technology described in Patent Document 1, image data for learning is obtained by imaging mud discharged from a shield tunneling machine. In other words, in the technology described in Patent Document 1, machine learning is performed by imaging mud discharged in actual construction, so there is a problem that it is difficult to complete learning at the start of shield tunneling, and the learning model cannot be started in a state suitable for the ground. From this perspective, the present invention provides a property determination method and learning image capture device that can accurately determine the properties of excavated soil (one example is mud) at the start of shield tunneling. [Means for solving the problem]

[0005] The property determination method according to the present invention is a method for determining the properties of excavated soil discharged from a shield tunneling machine. This property determination method includes a simulated soil preparation step, a learning data preparation step, a learning step, and a determination step. In the simulated soil preparation step, simulated soil simulating the properties of the excavated soil is prepared. In the learning data preparation step, a data set of learning data consisting of image data of the simulated soil and property information related to the properties of the simulated soil is prepared. In the learning step, the data set is used to train the learning model so that the image data of the excavated soil is input to the learning model to output property information. In the determination process, a trained learning model is used to determine property information of the excavated soil from image data of the excavated soil. In the property determination method according to the present invention, learning can be completed using imitation soil before the start of shield tunneling. Therefore, the learning model can be made suitable for the ground at the start of shield tunneling, so that the properties of the excavated soil (for example, mud soil) can be determined with high accuracy from the start of tunneling.

[0006] In the artificial soil preparation step, the properties of the artificial soil may be determined based on the results of a soil investigation included in design documents. In addition, in the simulated soil preparation process, the simulated soil may be prepared using soil and sand collected during the construction of the shaft, or purchased soil and sand may be sifted and mixed to reproduce the grain size test results from a soil survey. In this way, it is possible to easily create imitation soil suitable for learning. In the learning data preparation step, an image of the imitation soil transported by a transport means having a plurality of endless conveyors may be captured. In this way, the simulated soil can be photographed repeatedly, which reduces the burden of photographing the simulated soil. In addition, the simulated soil is stirred at the connection part of the conveyor, so the simulated soil in different states (variations) can be photographed.

[0007] The learning image capturing device according to the present invention is a device for creating image data used for learning the properties of excavated soil discharged from a shield tunneling machine. This educational image capturing device comprises a conveying means for transporting simulated soil that simulates the properties of the excavated soil by arranging multiple conveyors endlessly, a covering means for covering part or all of the multiple conveyors, and an imaging means arranged within the covering means for capturing images of the simulated soil transported by the conveying means. By using the learning image capturing device according to the present invention, learning can be completed using imitation soil before the start of shield tunneling. Therefore, the learning model can be made suitable for the ground at the start of shield tunneling, so that the properties of the excavated soil (for example, mud) can be accurately determined from the start of tunneling. Effect of the Invention

[0008] According to the present invention, the properties of excavated soil can be determined with high accuracy at the start of shield tunneling. [Brief description of the drawings]

[0009] [Figure 1] 1 is a configuration diagram of a property determination system according to an embodiment of the present invention. [Figure 2A]FIG. 1 is an external view of a learning image capturing device, with a cover means drawn in phantom lines so that the internal configuration can be seen. [Figure 2B] FIG. [Diagram 3] FIG. 2 is a cross-sectional view of a shield tunneling machine. [Figure 4] FIG. 2 is a functional configuration diagram of a mud property learning device and a mud property determining device. [Diagram 5] 1 is an example of a flowchart illustrating steps of a property determination method according to an embodiment of the present invention. [Figure 6] FIG. 13 is a diagram for explaining the photographing conditions of the simulated soil in the verification test. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, the embodiment of the present invention will be described in detail with reference to the drawings. Each drawing is merely a schematic illustration to allow a sufficient understanding of the present invention. Therefore, the present invention is not limited to the illustrated examples. In each drawing, common or similar components are denoted by the same reference numerals, and duplicated explanations thereof will be omitted.

[0011] <Configuration of the property determination system according to the embodiment> A property determination system 1 according to an embodiment will be described with reference to Fig. 1. Fig. 1 is a configuration diagram of the property determination system 1 according to an embodiment. The property determination system 1 is a system that determines the properties (such as material composition and hardness, for example) of mud (excavated soil and sand that has been given impermeability and plastic flowability by adding a mud-adding material) discharged from a shield machine 8 used in an earth pressure shield tunneling method. The property determination system 1 uses machine learning technology to determine the properties from image data of the mud.

[0012] The property determination system 1 has a pre-process, which is a process before the start of shield tunneling, and a determination process, which determines the properties of the mud discharged from the shield tunneling machine 8 at the actual construction site. In the pre-process, learning about the mud properties is carried out using simulated soil that simulates the properties of mud. The pre-process mainly comprises a simulated soil production process for producing simulated soil that simulates the properties of mud, a learning data preparation process for photographing the simulated soil and preparing learning data, and a learning process for training a learning model using the learning data.

[0013] As shown in FIG. 1, the property determination system 1 mainly includes a learning image capture device 2, a management terminal 3, a mud property learning device 4, a judgment image capture device 5, and a mud property determination device 6. These devices can communicate via a network such as the Internet or a LAN (Local Area Network). The communication method may be either wired or wireless. The property determination system 1 can also be configured as a cloud system. Note that the configuration of the property determination system 1 shown in FIG. 1 is merely an example. In the preliminary process, the learning image capture device 2, the management terminal 3, and the mud property learning device 4 are used, and in the determination process, the judgment image capture device 5 and the mud property determination device 6 are used.

[0014] The learning image capturing device 2 will be described with reference to Figures 2A and 2B. Figure 2A is an external view of the learning image capturing device 2, and Figure 2B is a plan view of the conveying means 21 constituting the learning image capturing device 2. In Figure 2A, the covering means 22 is drawn with imaginary lines so that the internal configuration can be seen. 2A is a device that creates image data for learning by using imitation soil that mimics the properties of mud (for example, material composition and hardness). The learning image capturing device 2 mainly includes a conveying means 21, a covering means 22, and an imaging means 23.

[0015] The conveying means 21 shown in Fig. 2A conveys the imitation soil. The conveying means 21 is composed of a plurality of conveyors 25 (one example is a belt conveyor), and the conveyors 25 are arranged endlessly. The conveying means 21 in this embodiment is composed of three conveyors 25 (a first conveyor 25A, a second conveyor 25B, and a third conveyor 25C), and are arranged in a triangle shape as shown in Fig. 2B. The arrow in Fig. 2B indicates the rotation direction of the conveyors 25. The first conveyor 25A is disposed horizontally, and the height of the front end 25Aa and the rear end 25Ab are the same (see FIG. 2A). The imitation soil transported by the first conveyor 25A falls from the front end 25Aa.

[0016] The second conveyor 25B is installed downstream of the first conveyor 25A. The second conveyor 25B is installed at an angle to the horizontal plane, and the front end 25Ba is higher than the rear end 25Bb (see FIG. 2A). The rear end 25Bb of the second conveyor 25B is located below the front end 25Aa of the first conveyor 25A, and the simulated soil transported by the first conveyor 25A falls onto the rear end 25Bb of the second conveyor 25B. The simulated soil that falls onto the second conveyor 25B is transported by the second conveyor 25B and falls from the front end 25Ba. The third conveyor 25C is installed downstream of the second conveyor 25B. The third conveyor 25C is similar to the second conveyor 25B and is installed tilted with respect to the horizontal plane, with the leading end 25Ca being higher than the trailing end 25Cb (see FIG. 2A). The trailing end 25Cb of the third conveyor 25C is located below the leading end 25Ba of the second conveyor 25B, and the simulated soil transported by the second conveyor 25B falls onto the trailing end 25Cb of the third conveyor 25C. The simulated soil that falls onto the third conveyor 25C is transported by the third conveyor 25C and falls from the leading end 25Ca onto the first conveyor 25A. With this configuration, the imitation soil circulates on the transport means 21 while repeatedly dropping at the connecting parts of the conveyors 25 (that is, the imitation soil repeatedly moves between the conveyors 25, and moves continuously on the conveyors 25).

[0017] The covering means 22 shown in FIG. 2A is for reproducing the environment in which mud is actually photographed during shield construction (for example, the environment inside the shield tunneling machine 8). The covering means 22 has a structure capable of blocking light (natural light) from the outside, for example, and is installed so as to cover a part or all of the multiple conveyors 25. The covering means 22 in this embodiment is installed so as to cover the first conveyor 25A. The covering means 22 has, for example, a bag shape, and is placed from above on a stand 24 installed on the first conveyor 25A. An imaging means 23 is installed inside the covering means 22. Incidentally, a lighting means (not shown) for illuminating the simulated soil to be photographed may be installed inside the covering means 22. The imaging means 23 captures an image of the dummy soil transported by the transport means 21. The imaging means 23 is, for example, a digital camera, and outputs image data to the management terminal 3 (see FIG. 1).

[0018] The management terminal 3 shown in FIG. 1 is operated by an administrator of the property determination system 1, etc. The management terminal 3 acquires image data of the simulated soil from the learning image capturing device 2, and stores the image data. The manager also registers information (property information) related to the properties of the simulated soil that has been created in the management terminal 3. For example, the manager registers information such as the mix of materials used in the created simulated soil in the management terminal 3. The manager also measures the hardness and softness of the simulated soil, and registers the measurement results in the management terminal 3. The management terminal 3 associates the acquired image data of the simulated soil with information related to the properties of the simulated soil, and transmits the image data to the mud property learning device 4. In addition, various information related to tunnel construction may be stored in the management terminal 3. For example, information on design documents for tunnel construction (including soil investigation results) is stored in the management terminal 3. The soil investigation results are used to determine the properties of the artificial soil to be created.

[0019] The mud property learning device 4 shown in FIG. 1 is a device that plays a central role in the processing of the learning process in which a learning model is trained using learning data. The mud property learning device 4 acquires image data of the simulated soil and information on the properties of the simulated soil (property information) from the management terminal 3 (see the white arrow in FIG. 1). The mud property learning device 4 then learns the relationship between the mud image data and the properties using the image data of the simulated soil and the property information of the simulated soil. The learning method will be described in detail later. The learned model learned by the mud property learning device 4 is transmitted to a mud property determination device 6 used in actual tunnel construction, and is used to determine the properties of the mud discharged from a shield machine 8 (see the dotted arrow in FIG. 1).

[0020] The assessment-use image capturing device 5 shown in Fig. 1 is a device that captures images of mud discharged from a shield tunneling machine 8. The assessment-use image capturing device 5 is, for example, a digital camera, and outputs image data to a mud property determining device 6. The assessment-use image capturing device 5 is installed, for example, behind the shield tunneling machine 8, in a location where it is possible to capture images of mud. With reference to FIG. 3, the photographing state of the mud by the judgment image photographing device 5 will be described. FIG. 3 is a cross-sectional view of a shield tunneling machine 8. The shield tunneling machine 8 shown in FIG. 3 is equipped with a cutter head (cutter spoke) 81 provided at the front end (face K side) and a partition wall 82 provided at the rear of the cutter head 81, and a chamber 83 is formed between the cutter head 81 and the partition wall 82. A plurality of cutter bits 81a, 81a, . . . are fixed to the cutter head 81. In addition, a screw conveyor 84 for discharging the mud 7 in the chamber 83 is provided inside the shield tunneling machine 8. The screw conveyor 84 penetrates the partition wall 82 and is connected to the chamber 83. In addition, a shield jack 85 for applying thrust to the shield tunneling machine 8 is provided, and a motor 86 for applying a rotational force to the cutter head 81 is provided.

[0021] In the earth pressure shield tunneling method, mud additives are added to the soil excavated by the cutter bits 81a, 81a, ... and mixed in the chamber 83 to generate impermeable and plastically fluid mud 7, which fills the chamber 83. While maintaining this state, mud pressure is generated in the mud 7 in the chamber 83 by the thrust of the shield jack 85, which counters the earth pressure and groundwater pressure of the face K and excavates while balancing the excavation amount and discharge amount of the shield machine 8. A belt conveyor 87 is provided behind the shield tunneling machine 8 (toward the tunnel entrance E) to transport the mud 7 (the mixed soil and sand described above in this embodiment) discharged from the discharge outlet of the screw conveyor 84 further rearward. The judgment image capturing device 5 is installed so as to capture an image of the mud 7 transported by the belt conveyor 87. As a result, the ground is excavated by the shield tunneling machine 8, and the soil and sand taken into the chamber 83 is discharged as mud 7 through the screw conveyor 84 and transported rearward by the belt conveyor 87. The properties of the transported mud 7 are then judged by the mud property determining device 6.

[0022] The mud property determination device 6 shown in FIG. 1 is a device that plays a central role in the process of determining the properties of the mud 7 discharged from a shield machine 8 at an actual construction site. The mud property determination device 6 is operated by a tunnel construction worker or the like. The mud property determination device 6 acquires image data of the mud 7 from the determination image capture device 5 and stores the image data. This mud property determination device 6 uses a trained learning model to determine the properties of the mud 7 shown in the image data captured by the determination image capture device 5. The trained learning model used for the determination is trained using image data of simulated soil. The mud property determination device 6 inputs the image data of the mud 7 into the trained learning model to obtain the properties of the mud 7. The determination method will be described in detail later. The mud property determination device 6 may notify a worker or the like of the determined properties of the mud 7. For example, the mud property determination device 6 may display information (property information) related to the determined property on its own display unit, or may transmit the property information to a mobile terminal (not shown) carried by a worker, etc. The worker performs construction management of the tunnel based on the determination result (property information of the mud 7).

[0023] The functional configuration of the mud property learning device 4 and the mud property determining device 6 will be described with reference to Fig. 4. Fig. 4 is a functional configuration diagram of the mud property learning device 4 and the mud property determining device 6 according to the embodiment. The mud property learning device 4 comprises a learning data acquisition unit 41 and a learning processing unit 42, and the learning processing unit 42 has a learning model. The learning model is a learning system in machine learning, which can derive good results by comparing the results of classification, prediction, and judgment based on given data with the actual results that are the correct answers, and adjusting various parameters. For example, convolutional neural networks, random forests, gradient boosting, etc. can be used as the learning model algorithm.

[0024] The learning data acquisition unit 41 shown in FIG. 4 acquires a data set of learning data for training a learning model. The learning data is configured as a set of image data of the simulated soil and property information related to the properties of the simulated soil. The property information is annotation data of the image data, and for example, a plurality of indices indicating the hardness or softness of the simulated soil are linked to the image data. The indices indicating the hardness or softness are, for example, five indices "hard, slightly hard, medium, slightly soft, soft", and are classified based on the measured slump value, slump flow value, table flow value, moisture content (moisture content), adhesion strength by vane shear test, fine grain content, etc. of the simulated soil.

[0025] The learning processing unit 42 uses a data set of learning data to train the learning model by machine learning. The method of training the learning model is not particularly limited, and it is preferable to train the learning model by a method suitable for the algorithm of the learning model. For example, image data of the learning data is input to the input layer, and the intermediate layer is adjusted based on the error between the result output from the output layer and the property information of the learning data. In other words, the learning processing unit 42 trains the learning model by machine learning so that the property information of the mud 7 is output by inputting image data of the mud 7. The number of learning data included in the data set is not particularly limited, and the learning is terminated when, for example, an expected accuracy (correct answer rate) is reached. The trained learning model is transmitted to the mud property determination device 6.

[0026] The mud property determination device 6 includes a determination data acquisition unit 61 and an estimation processing unit 62, and the estimation processing unit 62 has a trained learning model. The trained learning model is a learning model that has been machine-trained by the mud property learning device 4. In other words, the trained learning model has been trained so that appropriate property information is output by inputting image data of the mud 7. The determination data acquisition unit 61 acquires determination data that is the basis for judging property information of the mud 7. The determination data is image data obtained by photographing the mud 7 discharged from the shield tunneling machine 8 by the determination image photographing device 5. The estimation processing unit 62 uses the determination data (image data of the mud 7) to estimate the properties of the mud 7. A trained learning model is used to estimate the properties of the mud 7, and image data of the mud 7 is input to an input layer of the trained model, and property information of the mud 7 (one example is an index indicating hardness "hard, slightly hard, medium, slightly soft, soft") is output from an output layer.

[0027] <About the property determining method using the property determining system according to the embodiment> With reference to Fig. 5 (and also with reference to Figs. 1 to 4 as appropriate), a property determination method using the property determination system 1 according to the embodiment will be described. Fig. 5 is an example of a flowchart showing the steps of the property determination method using the property determination system 1. As shown in Fig. 5, the property determination method has, for example, a preliminary step "S10", a start of shield excavation "S20", and a determination step "S30". The preliminary step "S10" also has a simulated soil production step "S11", a learning data preparation step "S12", and a learning step "S13". The preliminary step "S10" is work carried out before the start of shield excavation.

[0028] (Simultaneous soil preparation process "S11") A simulated soil is created that mimics the properties of mud discharged in actual tunnel construction (for example, material composition and hardness). When creating the simulated soil, it is advisable to refer to the results of the soil survey included in the design documents to determine the properties of the simulated soil. In addition, the simulated soil may be created by sifting soil collected when constructing the departure shaft or arrival shaft of the shield tunneling machine or purchased soil, and using the soil mixed to reproduce the grain size test results in the soil survey. The manager (engineer) also registers information (property information) regarding the properties of the prepared simulated soil in the management terminal 3. For example, information such as the mix of materials for the prepared simulated soil is registered in the management terminal 3. The manager (engineer) also measures the hardness of the simulated soil and registers the measurement results in the management terminal 3.

[0029] (Learning data preparation process "S12") The prepared imitation soil is photographed using the learning image capturing device 2 shown in Fig. 2A, and image data of the imitation soil is created. Specifically, the transport means 21 is operated to capture an image of the imitation soil being transported on the conveyor 25. The image data of the imitation soil is sent to the management terminal 3 and stored. If the properties of the imitation soil change during the long period of imaging, the imitation soil may be sampled during the imaging process, and the properties may be measured, and the results may be registered in the management terminal 3 as property information. The administrator (engineer) operates the management terminal 3 to associate the acquired image data of the simulated soil with the property information of the simulated soil, and transmits them as a set to the mud property learning device 4. This completes the preparation of a data set of learning data consisting of image data of the simulated soil and property information related to the properties of the simulated soil in the mud property learning device 4. Note that the image data of the simulated soil and the property information of the simulated soil may be transmitted separately to the mud property learning device 4, and annotation (tagging) may be performed in the mud property learning device 4.

[0030] (Learning process "S13", start of shield excavation "S20", judgment process "S30") The mud property learning device 4 learns the learning model using a data set of learning data consisting of image data of the simulated soil and property information of the simulated soil. It is preferable to complete the learning of the learning model before shield tunneling starts. After that, shield tunneling by the shield machine 8 starts. While the shield tunneling is continuing, the image capturing device for judgment 5 captures images of the mud 7 discharged from the shield tunneling machine 8, and the mud property judgment device 6 judges the properties of the mud 7 based on the image data of the captured mud 7. A trained learning model trained using imitation soil is used to judge the properties.

[0031] As described above, in the property determination system 1 according to this embodiment, the learning model is trained using imitation soil before construction begins, and the learned learning model is used to determine the properties of the mud 7 from the start of construction. In other words, learning can be completed using imitation soil before the start of shield excavation. Therefore, the learning model can be made suitable for the ground at the start of shield excavation, so the properties of the excavated soil can be accurately determined from the start of excavation.

[0032] A test was carried out to verify the effectiveness of the judgment-use image capturing device 5 according to this embodiment, which will now be described. (1. Creation of the shooting location) In the verification test, as shown in Figure 2A, three belt conveyors were arranged in a triangle, and the horizontally arranged belt conveyors and camera were covered with a blackout curtain to be used as the judgment image capture device 5. By covering the camera with a blackout curtain, a situation was reproduced assuming a shielded site. The camera used was a commercially available network camera (hereinafter referred to as an AI camera) that allows manual image capture settings.

[0033] (2. Photographing the imitation soil) As shown in Figure 6, the AI ​​camera was set to take pictures at angles of 35°, 45°, and 55° relative to the conveyor belt. Markings were placed on the conveyor belt beforehand and pictures were taken. Figure 6 is a diagram to explain the shooting conditions of the simulated soil in the verification test.

[0034] (3. Summary) As a result of the verification test, the AI ​​camera was able to create a sufficient amount of image data for AI learning. In other words, it was found that the judgment image capture device 5 is effective for creating image data for AI learning. In addition, a simulation of learning and judgment was performed using the images captured in the verification test described above. As a result, the accuracy (correct answer rate), matching rate, and recall rate all achieved a value of 0.92 or higher. Here, accuracy is the ratio of the number of correct answers to the total number of samples. Matching rate is the degree to which there are no false positives, and is the ratio of the number of correct answers to the number of inferences for each property. Recall rate is the degree to which there are no missed detections, and is the ratio of the number of inferences to the number of correct answers for each property. These results confirmed that accurate inferences can be made using image data captured of dummy soil by the judgment image capture device 5.

[0035] Although the embodiment of the present invention has been described above, the present invention is not limited to this, and can be practiced within the scope of the claims. In the embodiment, a case has been described in which muddy soil is generated with impermeability and plastic flowability by adding muddy soil to excavated soil and mixing the soil. However, it is also possible to determine the properties of soil (excavated soil) to which no muddy soil is added by the property determination system 1. The muddy soil in the embodiment is an example of excavated soil. [Explanation of symbols]

[0036] 1. Characterization system 2. Learning Image Capture Device 3. Management terminal 4 Mud soil property learning device 5. Image capture device for judgment 6 Mud soil property determination device 8. Shield tunneling machine 21 Means of transportation 22 Covering means 23 Imaging means 25 Conveyor 41 Learning data acquisition unit 42 Learning processing unit 61 Judgment data acquisition unit 62 Estimation processing unit 81 Cutter Head 81a Cutter Bit 82 Bulkhead 83 Chamber 84 Screw Conveyor 85 Shield Jack 86 Motor 87 Conveyor Belt

Claims

1. A method for determining the properties of excavated soil discharged from a shield tunneling machine, comprising the steps of: A simulated soil preparation step of preparing simulated soil that simulates the properties of the excavated soil; A learning data preparation process for preparing a learning data set including image data of the simulated soil and property information related to the properties of the simulated soil; A learning process of training the learning model to output property information by inputting image data of the excavated soil to the learning model using the data set; A determination step of determining property information of the excavated soil from image data of the excavated soil using a trained learning model. A property determination method comprising:

2. In the simulated soil preparation process, the properties of the simulated soil are determined based on the soil investigation results included in the design documents.

2. The method for determining properties according to claim 1.

3. 2. The property determination method according to claim 1, wherein in the simulated soil preparation step, the simulated soil is prepared using soil and sand collected during construction of a shaft.

4. In the learning data preparation step, an image of the imitation soil transported by a transport means having a plurality of endless conveyors is taken.

2. The method for determining properties according to claim 1.

5. A learning image capturing device that creates image data used to learn the properties of excavated soil discharged from a shield tunneling machine, A conveying means for conveying simulated soil simulating the properties of the excavated soil by arranging a plurality of conveyors endlessly; A cover means for covering a part or all of the plurality of conveyors; An imaging means is disposed in the cover means and images the simulant soil transported by the transport means. A learning image capturing device characterized by:

Images