Damage location prediction device, damage location prediction method, damage location prediction program, and trained model generation method

Abstract

To enable prediction of locations where pavement damage is likely to occur.SOLUTION: A prediction device 10 includes an acquisition unit 101 that acquires input data based on data of reflected waves of microwaves emitted toward a paved road while traveling on the road, a prediction unit 102 that mechanically learns, as teacher data, location information that indicates characteristics that may cause damage in the input data and the road and predicts locations where damage is likely to occur on the road from the input data by using a learned model that outputs locations where damage may occur on the road when the input data is input, and a presentation unit 103 that presents the prediction result of the prediction unit 102.SELECTED DRAWING: Figure 5

Description

Technical Field

[0001] The present invention relates to a damage location prediction device, a damage location prediction method, a damage location prediction program, and a learned model generation method.

Background Art

[0002] The roads on which vehicles travel daily are damaged by the action of the load received from the vehicles, the temperature change at the location where the road is installed, and the influence of rainwater on the road. Specific examples of pavement damage include cracks, potholes, etc. A pothole is a round hole or depression formed by the peeling of the road surface layer. In order to quickly grasp such road damage and maintain and manage the road, the work of grasping the road surface conditions is carried out daily. Patent Document 1 discloses a learned model generation method and a road surface condition determination device for the purpose of being able to determine the road surface conditions with high accuracy at low cost.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] The technology disclosed in Patent Document 1 can determine the presence or absence of road surface conditions such as potholes, but it does not predict locations where pavement damage is likely to occur in the future.

[0005] In view of the above points, the present disclosure has been made, and an object thereof is to provide a damage location prediction device, a damage location prediction method, a damage location prediction program, and a learned model generation method capable of predicting locations where pavement damage is likely to occur.

Means for Solving the Problems

[0007] A damage location prediction device according to a second aspect of the present invention is a damage location prediction device according to a first aspect, wherein the prediction unit predicts the location in the area where the wheel makes contact with the road.

[0008] A third aspect of the present invention is a damage location prediction device according to the first or second aspect, wherein the prediction unit predicts the location using at least one of a plurality of data obtained by slicing the input data in the direction of travel of the road, a plurality of data obtained by slicing the input data in the direction of transverse travel of the road, or a plurality of data obtained by slicing the road in the depth direction.

[0009] A fourth aspect of the present invention is a damage location prediction device according to the third aspect, wherein the prediction unit predicts the location by combining a first result, which is the result of a prediction using a plurality of data sliced ​​in the depth direction of the road from the input data, and a second result, which is the result of a prediction using a plurality of data sliced ​​in the transverse direction of the road from the input data.

[0010] A fifth aspect of the present invention is a damage location prediction device according to the fourth aspect, wherein the prediction unit predicts the location by logical AND of the first result and the second result.

[0011] A sixth aspect of the present invention is a damage location prediction device according to the fourth aspect, wherein the prediction unit predicts the location by narrowing down the possibilities based on the simultaneous probability of the first result and the second result.

[0012] A seventh aspect of the present invention relates to a method for predicting damage locations, in which a processor acquires input data based on data of reflected microwaves irradiated toward a paved road while driving on the road, performs machine learning using the input data and information on locations on the road that show characteristics where pavement damage is likely to occur as training data, and uses a trained model that outputs locations on the road where pavement damage is likely to occur when the input data is input to predict locations on the road where damage is likely to occur from the input data, and performs processing to present the results of the prediction.

[0013] The eighth aspect of the present invention provides a damage location prediction program that causes a computer to acquire input data based on data of reflected microwave waves irradiated toward a paved road while driving on the road, and to perform machine learning on the input data and information on locations on the road that show characteristics where pavement damage is likely to occur, using a trained model that outputs locations on the road where pavement damage is likely to occur when the input data is input, to predict locations on the road where pavement damage is likely to occur from the input data, and to present the results of the prediction.

[0014] A trained model generation method according to a ninth aspect of the present invention involves a processor performing machine learning using input data based on data of reflected microwave waves irradiated toward a paved road while driving on the road, and information on locations on the road that exhibit characteristics where pavement damage is likely to occur, as training data, and generating a trained model through machine learning that outputs locations on the road where pavement damage is likely to occur when the input data is input. [Effects of the Invention]

[0015] According to the present disclosure, it is possible to provide a damage location prediction device, a damage location prediction method, a damage location prediction program, and a learned model generation method that can predict locations where pavement damage is likely to occur.

Brief Description of the Drawings

[0016] [Figure 1] It is a diagram showing a schematic configuration of a damage location prediction system including the prediction device according to the present embodiment. [Figure 2] It is a diagram for explaining the detection of the reflection response waveform. [Figure 3] It is a diagram showing the state of reflection of electromagnetic waves irradiated from an electromagnetic wave device. [Figure 4] It is a block diagram showing the hardware configuration of the prediction device. [Figure 5] It is a block diagram showing an example of the functional configuration of the prediction device. [Figure 6] It is a block diagram showing the hardware configuration of the learning device. [Figure 7] It is a block diagram showing an example of the functional configuration of the learning device. [Figure 8] It is a flowchart showing the flow of prediction processing by the prediction device. [Figure 9] It is a diagram showing an example of an image serving as a basis for Y volume data and Z volume data. [Figure 10] It is a diagram for explaining Y volume data and Z volume data. [Figure 11] It is a diagram showing an example of presentation of a prediction result by the prediction device.

Modes for Carrying Out the Invention

[0017] Hereinafter, an example of an embodiment of the present disclosure will be described while referring to the drawings. In each of the drawings, the same or equivalent components and parts are given the same reference numerals. Also, the dimensional ratios in the drawings are exaggerated for convenience of explanation and may be different from the actual ratios.

[0018] Figure 1 is a diagram showing the schematic configuration of a damage location prediction system including a prediction device according to this embodiment. As shown in Figure 1, the damage location prediction system 1 according to this embodiment is mounted on a vehicle 90.

[0019] Damage location prediction system 1 is a system that predicts locations where damage is likely to occur in the pavement of a road 80 paved with asphalt or the like, such as locations where hot poles are likely to occur. Damage location prediction system 1 consists of a prediction device 10 and an electromagnetic wave device 30. The prediction device 10 is a device for predicting locations where damage is likely to occur in the pavement of the road 80.

[0020] The electromagnetic wave device 30 comprises multiple electromagnetic wave irradiating units and receiving units arranged on a line. The electromagnetic wave device 30 is installed, for example, at the rear lower part of the vehicle 90, such that the direction of travel of the vehicle 90 is the bridge axis direction and the line direction of the electromagnetic wave device 30 is perpendicular to the bridge axis. The electromagnetic wave irradiating units irradiate electromagnetic waves such as microwaves toward the road 80. The receiving units receive reflected waves reflected from various parts inside the pavement of the road 80.

[0021] As the electromagnetic wave irradiation unit, any known electromagnetic wave radar system can be used without particular limitation, but it is preferable in terms of work efficiency and accuracy to use a radar system with multiple transmitting and receiving sensors arranged in parallel. Furthermore, it is preferable in terms of work efficiency to use an array antenna, which is an array-shaped antenna, for the transmitting and receiving sensors.

[0022] As shown in Figure 2, the electromagnetic wave device 30 scans the evaluation target area 95 of the road 80 in the direction of vehicle travel, irradiating electromagnetic waves from the surface towards the interior (depth) of the road 80, and receiving the reflected waves. This allows the reflected wave intensity corresponding to the depth to be detected for each grid in the evaluation target area 95. One grid can be, for example, 1 cm × 1 cm, and one line width can be 2.0 m. In this case, reflected response waveforms for 200 grids are detected per line.

[0023] Electromagnetic waves emitted from the electromagnetic wave device 30 are reflected at the boundaries of the layers of the road 80. However, if there is an abnormality at the boundary of the layers of the road 80 or within the layers, the electromagnetic waves emitted from the electromagnetic wave device 30 will be reflected in a pattern different from that of a normal state. Figure 3 shows the reflection of electromagnetic waves emitted from the electromagnetic wave device 30. If the road 80 is paved with asphalt and consists of a surface layer, base layer, upper subbase and lower subbase, then if each layer is in a normal state, the electromagnetic waves will be reflected as shown in Figure 3(a). On the other hand, if cracks occur at the boundaries of the layers or voids occur within the layers, the electromagnetic waves will be partially diffusely reflected as shown in Figure 3(b). Areas where electromagnetic waves are diffusely reflected as shown in Figure 3(b) appear normal at first glance in the surface layer, but abnormalities have occurred in the layers below, so the pavement is already damaged or is expected to be damaged in the future.

[0024] The prediction device 10 uses data of reflected waves from the road 80 to which the electromagnetic wave device 30 irradiates the road 80 when predicting areas where road pavement damage is likely to occur. The reflected wave data can be processed in 3D and used as image data. The prediction device 10 also uses a trained model generated by training the reflected wave data of the electromagnetic waves and information on damaged areas of the road pavement using machine learning techniques as training data when predicting areas where road pavement damage is likely to occur. By inputting the reflected wave information from the road 80 into the trained model, the prediction device 10 can predict areas where road pavement damage is likely to occur by outputting the trained model.

[0025] The electromagnetic wave device 30 outputs information on the reflected response waveform (reflected wave intensity according to depth) for each acquired grid to the prediction device 10. The electromagnetic wave device 30 is not limited to being mounted on the vehicle 90, but may also be in other forms, such as being held by a worker or in the form of a handcart. Similarly, the prediction device 10 is not limited to being mounted inside the vehicle 90, but may also be in other forms, such as being held by a worker.

[0026] Figure 4 is a block diagram showing the hardware configuration of the prediction device.

[0027] As shown in Figure 4, the prediction device 10 has the following components: CPU (Central Processing Unit) 11, ROM (Read Only Memory) 12, RAM (Random Access Memory) 13, storage 14, input unit 15, display unit 16, optical disc drive unit 17, and communication interface (communication I / F) 18. Each component is connected to the others via a bus 19 so that they can communicate with each other.

[0028] The CPU 11 is a central processing unit that executes various programs and controls various parts. Specifically, the CPU 11 reads a program from the ROM 12 or storage 14 and executes the program using the RAM 13 as a working area. The CPU 11 controls each of the above components and performs various calculations according to the program recorded in the ROM 12 or storage 14. In this embodiment, the ROM 12 or storage 14 stores a prediction program (damage location prediction program) for predicting where damage to the road pavement 80 will occur.

[0029] ROM12 stores various programs and data. RAM13 temporarily stores programs or data as a working area. Storage14 consists of an HDD (Hard Disk Drive) or SSD (Solid State Drive) and stores various programs, including the operating system, and various data.

[0030] The input unit 15 includes a pointing device such as a mouse and a keyboard, and is used for various types of input. The display unit 16 is, for example, a liquid crystal display and displays various types of information. The display unit 16 may also function as the input unit 15 by employing a touch panel system.

[0031] The optical disc drive unit 17 reads data stored on various recording media such as CD-ROMs (Compact Disc Read Only Memory) or Blu-ray discs, and writes data to these recording media.

[0032] The communication interface 18 is an interface for communicating with other devices, and standards such as Ethernet®, FDDI, and Wi-Fi® can be used.

[0033] When executing the above prediction program, the prediction device 10 uses the above hardware resources to implement various functions. The functional configuration implemented by the prediction device 10 will now be described.

[0034] Figure 5 is a block diagram showing an example of the functional configuration of the prediction device 10.

[0035] As shown in Figure 5, the prediction device 10 has an acquisition unit 101, a prediction unit 102, and a presentation unit 103 as its functional configuration. Each functional configuration is realized by the CPU 11 reading and executing a prediction program stored in the ROM 12 or storage 14.

[0036] The acquisition unit 101 acquires data of the reflected electromagnetic waves emitted by the electromagnetic wave device 30 toward the road 80. The acquisition unit 101 may also acquire image data generated by three-dimensional processing of the reflected electromagnetic wave data.

[0037] The prediction unit 102 inputs the reflected wave data acquired by the acquisition unit 101 into a pre-trained model 21 and outputs prediction results from the pre-trained model 21 to predict areas on the road 80 where damage to the pavement is likely to occur. The prediction unit 102 may also perform 3D processing on the reflected electromagnetic wave data to convert it into image data and input the image data into the pre-trained model 21. Alternatively, if the acquisition unit 101 acquires image data generated by 3D processing of the reflected electromagnetic wave data, the prediction unit 102 may input the image data acquired by the acquisition unit 101 into the pre-trained model 21.

[0038] The pre-trained model 21 used by the prediction unit 102 for prediction is a model that, upon input of image data generated from reflected electromagnetic waves directed toward the road, extracts and outputs locations on the road where damage is likely to occur. For example, the pre-trained model 21 outputs locations where damage is likely to occur if the probability of damage exceeding a predetermined threshold.

[0039] The prediction unit 102 may predict areas where damage to the road 80 pavement is likely to occur in the entire area where electromagnetic waves are irradiated onto the road 80, or it may predict areas where damage to the road 80 is likely to occur in the area where the wheels of a vehicle make contact with the road 80. The area where the wheels of a vehicle make contact with the road 80 refers to the area where the wheels mainly make contact with the road 80 due to the normal driving of a vehicle, and refers to the area where ruts may form after many vehicles have driven over it.

[0040] When making predictions by inputting image data into a trained model 21, the prediction unit 102 may input image data obtained by slicing a 3D image in a predetermined direction into the trained model 21 and have the trained model 21 make predictions. The direction of travel of the vehicle 90 on the road 80 is the X direction, the direction perpendicular to the X direction and horizontal, i.e., the transverse direction of the road 80 is the Y direction, and the depth direction of the road 80 is the Z direction. Images sliced ​​in the Y direction are called Y slice images, and a collection of multiple Y slice images is called Y volume data. Similarly, images sliced ​​in the Z direction are called Z slice images, and a collection of multiple Z slice images is called Z volume data.

[0041] In this case, the prediction unit 102 may input the Y volume data and Z volume data into the trained model 21 and narrow down the results extracted by the trained model 21. The prediction unit 102 may perform narrowing by logical AND of the extracted locations of both as a narrowing process. Alternatively, the prediction unit 102 may perform narrowing by joint probability as a narrowing process. The joint probability is (accuracy of Z volume data × accuracy of Y volume data)1 / 2 This is how it is determined. By performing a filtering process, the prediction unit 102 can reduce unnecessary extraction locations such as false detections.

[0042] Furthermore, the trained model 21 may be a model that has been trained using not only the Y volume data and Z volume data, but also X volume data, which is a collection of X slice images obtained by slicing images in the X direction.

[0043] In the example shown in Figure 5, the trained model 21 is located outside the predictor 10, but the trained model 21 may also be located inside the predictor 10.

[0044] The display unit 103 displays information on the locations where road 80 is likely to be damaged, as predicted by the prediction unit 102, to the display unit 16.

[0045] With this configuration, the prediction device 10 can predict information about areas where damage to the road pavement is likely to occur based on data of reflected electromagnetic waves emitted by the electromagnetic wave device 30 toward the road 80, and present the predicted results.

[0046] Next, we will describe the learning device that trains the pre-trained model used for prediction by the prediction device 10 and generates the pre-trained model.

[0047] Figure 6 is a block diagram showing the hardware configuration of the learning device.

[0048] As shown in Figure 6, the learning device 20 has the following components: CPU 21, ROM 22, RAM 23, storage 24, input unit 25, display unit 26, optical disc drive unit 27, and communication interface (communication I / F) 28. Each component is connected to the others via a bus 29 so that they can communicate with each other.

[0049] The CPU 21 is a central processing unit that executes various programs and controls various parts. Specifically, the CPU 21 reads a program from the ROM 22 or storage 24 and executes the program using the RAM 23 as a working area. The CPU 21 controls each of the above components and performs various calculations according to the program recorded in the ROM 22 or storage 24. In this embodiment, the ROM 22 or storage 24 stores a learning program that trains a model to predict where damage to the road pavement 80 will occur and generates a trained model.

[0050] ROM22 stores various programs and data. RAM23 temporarily stores programs or data as a working area. Storage24 consists of an HDD or SSD and stores various programs, including the operating system, and various data.

[0051] The input unit 25 includes a pointing device such as a mouse and a keyboard, and is used for various types of input. The display unit 26 is, for example, a liquid crystal display and displays various types of information. The display unit 26 may also function as the input unit 25 by employing a touch panel system.

[0052] The optical disc drive unit 27 reads data stored on various recording media such as CD-ROMs or Blu-ray discs, and writes data to the recording media.

[0053] The communication interface 28 is an interface for communicating with other devices, and standards such as Ethernet®, FDDI, and Wi-Fi® are used.

[0054] When executing the above learning program, the learning device 20 uses the above hardware resources to implement various functions. The functional configuration implemented by the learning device 20 will be described below.

[0055] Figure 7 is a block diagram showing an example of the functional configuration of the learning device 20.

[0056] As shown in Figure 7, the learning device 20 has an acquisition unit 201 and a learning unit 202 as its functional configuration. Each functional configuration is realized by the CPU 21 reading and executing a learning program stored in the ROM 22 or storage 24.

[0057] The acquisition unit 201 acquires training data for training a model and generating a trained model 21. The training data for generating the trained model 21 includes a pair of three-dimensional image data generated based on data of reflected electromagnetic waves irradiated by the electromagnetic wave device 30 toward the road 80, and information on the locations where damage has occurred on the road 80.

[0058] The learning unit 202 trains a model using the training data acquired by the acquisition unit 201 to generate a trained model 21. The learning unit 202 generates the trained model 21 by training in a direction that reduces the error between the locations on the road 80 that the trained model 21 predicts as having features that may cause damage, and the locations on the road 80 where damage actually occurs. The number of training data for training the model is not limited to a predetermined number, but it is desirable to use a number that does not cause overfitting.

[0059] The algorithm used by the learning unit 202 to train the model is not limited to a specific one. Examples of models or algorithms used in supervised learning as in this embodiment include regression analysis, decision trees, support vector machines (SVMs), neural networks, ensemble learning, and random forests. Alternatively, classification may be performed beforehand, and then supervised learning may be performed for each class. In this case, the classification may be supervised or unsupervised. In the example in Figure 7, the trained model 21 is located outside the learning device 20, but the trained model 21 may also be located inside the learning device 20.

[0060] When the learning unit 202 generates the trained model 21, it uses at least one of the X volume data, Y volume data, and Z volume data as training data. The learning unit 202 may train the model using only one volume data, but by training the model using multiple volume data, the accuracy of the prediction can be improved compared to training with only one volume data.

[0061] By using the trained model 21 generated by the learning device 20, the prediction device 10 can predict information about areas on the road 80 where damage to the road pavement is likely to occur, based on the data of reflected electromagnetic waves emitted by the electromagnetic wave device 30 toward the road 80, and present the predicted results.

[0062] Next, the operation of the prediction device 10 will be explained.

[0063] Figure 8 is a flowchart showing the flow of the prediction process performed by the prediction device 10. The CPU 11 reads the prediction program from the ROM 12 or storage 14, loads it into the RAM 13, and executes it to perform the prediction process.

[0064] In step S101, the CPU 11 acquires data on the reflected electromagnetic waves emitted by the electromagnetic wave device 30 toward the road 80.

[0065] After acquiring the reflected wave data, the CPU 11 then inputs the reflected wave data into the trained model in step S102 to predict areas on the road 80 where damage to the pavement is likely to occur.

[0066] The CPU 11 may input Y-volume data and Z-volume data into the trained model 21 when predicting areas on the road 80 where pavement damage is likely to occur. Figure 9 shows examples of images that form the basis of Y-volume data and Z-volume data. Figure 9(a) is an example of a Z-slice image that forms the basis of Z-volume data, and Figure 9(b) is an example of a Y-slice image that forms the basis of Y-volume data. Figure 10 also shows examples of Y-volume data and Z-volume data obtained by slicing a 3D image. Figure 10(a) is an example of Z-volume data consisting of multiple Z-slice images, and Figure 10(b) is an example of Y-volume data consisting of multiple Y-slice images.

[0067] The CPU 11 may then input the Y volume data and Z volume data into the trained model 21 and refine the results extracted by the trained model 21. The CPU 11 may refine the results by performing a logical AND operation of the extracted locations from both volumes. Alternatively, the CPU 11 may refine the results by performing a joint probability operation. The joint probability is calculated as (Probability of Z volume data × Probability of Y volume data). 1 / 2 This is how it is determined. By performing a filtering process, the CPU 11 can reduce unnecessary extraction points such as false detections.

[0068] After predicting information about areas where road 80's pavement is likely to be damaged, the CPU 11 then presents the prediction results in step S103.

[0069] Figure 11 shows an example of the prediction results presented by the prediction device 10. Figure 11(a) is an example of the result of inputting Y volume data into the trained model 21, and Figure 11(b) is an example of the result of inputting Z volume data into the trained model 21. As shown in Figure 11, by inputting image data generated based on electromagnetic wave reflection data into the trained model 21, the prediction device 10 can extract locations similar to those actually detected by the analyst as having damage to the road pavement 80.

[0070] By performing the aforementioned process, the prediction device 10 can predict information about areas on the road 80 where damage to the road pavement is likely to occur, based on data of reflected electromagnetic waves emitted by the electromagnetic wave device 30 toward the road 80, and present the predicted results.

[0071] The prediction device 10 and the learning device 20 are equipped with CPUs 11 and 21 as part of their hardware configuration, but may also be equipped with a GPU (Graphics Processing Unit). By equipping the prediction device 10 and the learning device 20 with a GPU in addition to the CPUs 11 and 21, the prediction device 10 and the learning device 20 can perform parallel processing of prediction and learning processes using the CPUs 11 and 21 and the GPU. By performing parallel processing of prediction and learning processes using the CPUs 11 and 21 and the GPU, the prediction device 10 and the learning device 20 can be expected to improve the speed of detecting damaged areas in the pavement and improve the speed of learning the learning model.

[0072] The learning device 20 may further train the trained model 21 using the prediction results output by the prediction device 10 using the trained model 21. For example, if there are locations that the prediction unit 102 cannot detect in areas that show features that may cause damage to the pavement, the learning device 20 may improve the detection accuracy by performing additional training using the data of the undetected locations as input. The learning device 20 can improve the accuracy of the trained model 21 by further training the trained model 21.

[0073] Furthermore, it is conceivable that the characteristics that may cause damage to the pavement may vary depending on the road structure, weather conditions, traffic volume, and other environmental conditions in each region. Therefore, the learning device 20 may create a trained model 21 for each region or route to optimize the prediction of pavement damage. By creating a trained model 21 for each region or route, the prediction device 10 can detect locations where pavement damage is likely to occur with greater accuracy compared to when using a trained model 21 that does not take into account the differences between regions or routes.

[0074] In addition, the learning and prediction processes that the CPU reads and executes in each of the above embodiments may be executed by various processors other than the CPU. Examples of such processors include PLDs (Programmable Logic Devices) such as FPGAs (Field-Programmable Gate Arrays) whose circuit configuration can be changed after manufacturing, and dedicated electrical circuits that are processors with circuit configurations specifically designed to execute specific processes, such as ASICs (Application Specific Integrated Circuits). Furthermore, the learning and prediction processes may be executed by one of these various processors, or by a combination of two or more processors of the same or different types (for example, multiple FPGAs, and a combination of a CPU and an FPGA). More specifically, the hardware structure of these various processors is an electrical circuit that combines circuit elements such as semiconductor elements.

[0075] Furthermore, while the above embodiments describe a configuration in which the learning process and prediction process programs are pre-stored (installed) in ROM or storage, the system is not limited to this. The programs may be provided in a form recorded on a non-transitory recording medium such as a CD-ROM (Compact Disk Read Only Memory), DVD-ROM (Digital Versatile Disk Read Only Memory), or USB (Universal Serial Bus) memory. Alternatively, the programs may be provided in a form that can be downloaded from an external device via a network. [Explanation of Symbols]

[0076] 10 Prediction device 20 Learning device 30 Electromagnetic wave device 80 road 90 vehicles

Claims

1. An acquisition unit that acquires input data based on data of reflected microwave waves irradiated toward the road while driving on a paved road, A prediction unit uses a trained model, which performs machine learning on the input data and information on locations on the road that show characteristics where pavement damage is likely to occur, as training data, and outputs locations on the road where damage is likely to occur in the future from the time the input data was acquired, to predict locations on the road where pavement damage is likely to occur in the future from the time the input data was acquired, based on the input data. A display unit that displays the prediction results of the prediction unit, Equipped with, The training data includes a pair of three-dimensional image data generated from the input data and information on locations where damage has occurred on the road, and the prediction unit inputs the three-dimensional image data generated from the input data into the trained model to predict locations on the road where pavement damage is likely to occur in the future from the time the input data was acquired, based on the input data, in a damage location prediction device.

2. The damage location prediction device according to claim 1, wherein the prediction unit predicts the location in the area where the wheel makes contact with the road.

3. The damage location prediction device according to claim 1 or 2, wherein the prediction unit predicts the location using at least one of a plurality of data obtained by slicing the input data in the direction of travel of the road, a plurality of data obtained by slicing the input data in the direction of transverse travel of the road, and a plurality of data obtained by slicing the road in the depth direction.

4. The damage location prediction device according to claim 3, wherein the prediction unit predicts the location by combining a first result, which is the result of a prediction using a plurality of data obtained by slicing the input data in the depth direction of the road, and a second result, which is the result of a prediction using a plurality of data obtained by slicing the input data in the transverse direction of the road.

5. The damage location prediction device according to claim 4, wherein the prediction unit predicts the location by logical AND of the first result and the second result.

6. The damage location prediction device according to claim 4, wherein the prediction unit predicts the location by narrowing down the results based on the simultaneous probability of the first result and the second result.

7. The processor, Input data is acquired based on the data of reflected microwaves irradiated toward the road while driving on the paved road. Using the aforementioned input data and information on locations on the road that exhibit characteristics suggesting that pavement damage may occur in the future from the time the input data was acquired, machine learning is performed on the trained model, which, when the aforementioned input data is input, outputs locations on the road where pavement damage may occur in the future from the time the input data was acquired, thereby predicting locations on the road where damage may occur in the future from the time the input data was acquired, based on the aforementioned input data. The results of the aforementioned prediction are presented below. Execute the process, The aforementioned training data includes a pair of three-dimensional image data generated from the input data and information on the locations where damage has occurred on the road. The processor inputs three-dimensional image data generated from the input data into the trained model, thereby predicting locations on the road where pavement damage is likely to occur in the future from the time the input data was acquired.

8. On the computer, Input data is acquired based on the data of reflected microwaves irradiated toward the road while driving on the paved road. Using the aforementioned input data and information on locations on the road that exhibit characteristics suggesting that pavement damage may occur in the future from the time the input data was acquired, machine learning is performed on the trained model, which, when the aforementioned input data is input, outputs locations on the road where pavement damage may occur in the future from the time the input data was acquired, thereby predicting locations on the road where pavement damage may occur in the future from the time the input data was acquired, based on the aforementioned input data. The results of the aforementioned prediction are presented below. Execute the process, The aforementioned training data includes a pair of three-dimensional image data generated from the input data and information on the locations where damage has occurred on the road. A damage location prediction program that, by inputting three-dimensional image data generated from the input data into the trained model, causes the computer to predict locations on the road where pavement damage is likely to occur in the future from the time the input data was acquired.

9. The processor, Machine learning is performed using input data based on the data of reflected microwave waves irradiated toward the road while driving on the paved road, and information on locations on the road that show characteristics where pavement damage is likely to occur in the future from the time the input data was acquired, as training data. The aforementioned training data includes a pair of three-dimensional image data generated from the input data and information on the locations where damage has occurred on the road. This machine learning method generates a trained model that, when inputting 3D image data generated from the input data, outputs locations on the road where pavement damage is likely to occur in the future from the time the input data was acquired. Method for generating pre-trained models.

Images