Sediment movement area estimation method, trained model, and sediment movement area estimation device

Abstract

To estimate a sediment movement area with high accuracy.SOLUTION: The present invention is a sediment movement area estimation method which uses a computer to estimate a sediment movement area. The sediment movement area estimation method includes, at least: a difference generation step in which an arithmetic unit provided in the computer generates difference information from differences between a digital elevation model and a digital surface model measured before a landslide disaster and a digital surface model measured after the landslide disaster; and an estimation step in which the arithmetic unit estimates the movement area on the basis of the difference information.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a method for estimating the moving area of earth and sand, a learned model, and an apparatus for estimating the moving area of earth and sand.

Background Art

[0002] Conventionally, after a large-scale earth and sand disaster occurs, it is required to quickly calculate the amount of earth and sand that has moved in order to grasp the overall situation of the disaster and formulate a restoration plan. For example, Patent Document 1 discloses a technique for estimating the amount of earth and sand that has moved using ortho-image data obtained by converting aerial photo image data into an orthographic projection. In this patent document, a technique for creating three-dimensional mesh data (topographic data indicating the moving area of earth and sand) based on the ortho-image data and estimating the amount of earth and sand that has moved is described.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, ortho-image data is affected by weather, seasons, and the like. Therefore, there is a risk that the estimation accuracy of the moving area of earth and sand will decrease.

[0005] Therefore, a main object of the present invention is to provide a method for estimating the moving area of earth and sand, a learned model, and an apparatus for estimating the moving area of earth and sand with high accuracy.

Means for Solving the Problems

[0006] The present invention is a method for estimating the moving area of earth and sand using a computer, comprising A difference generation step in which an arithmetic unit included in the computer generates difference information based on the difference between each of a digital elevation model and a digital surface model measured before a landslide disaster and a digital surface model measured after the landslide disaster; An estimation step in which the arithmetic unit estimates the movement area based on the difference information, and provides a landslide movement area estimation method including at least the above. Each of the digital elevation model and the digital surface model may be acquired using airborne laser scanning or UAV laser scanning. In the estimation step, the arithmetic unit may estimate using a learned model generated by performing machine learning that takes the difference information as an input and outputs the movement area of the sediment. The learned model includes a deep neural network (DNN) and a convolutional neural network (CNN). The estimation step includes a first estimation step and a second estimation step. In the first estimation step, the arithmetic unit inputs the difference information into the DNN to output the movement area. In the second estimation step, the arithmetic unit may input the difference information and the output result of the DNN into the CNN to output the movement area. The DNN outputs the movement area in mesh units. The CNN may output the movement area in pixel units. The difference information includes first difference information and second difference information. The first difference information is generated by the difference between the digital elevation model measured before the landslide disaster and the digital surface model measured after the landslide disaster. The second difference information may be generated by the difference between the digital surface model measured before the landslide disaster and the digital surface model measured after the landslide disaster. In the second estimation step, the arithmetic unit may further input an MPI (Morphometric Protection Index) into the CNN to output the movement area. Also, the present invention is A neural network comprising at least one input layer, at least one intermediate layer, and at least one output layer, First difference information generated from the difference between a digital elevation model measured before a landslide and a digital surface model measured after the landslide, and second difference information generated from the difference between a digital surface model measured before the landslide and a digital surface model measured after the landslide are input into the input layer, and the presence or absence of sediment movement is output from the output layer, whereby the weighting coefficients between the nodes of each layer are learned. Provided is a learned model for causing a computer to function so as to perform an operation based on the learned weighting coefficients based on difference information generated from the difference between each of a digital elevation model and a digital surface model measured before a landslide and a digital surface model measured after the landslide, and output a sediment movement area. Further, the present invention Comprises at least a calculation unit for estimating a sediment movement area, The calculation unit A difference generation step of generating difference information from the difference between each of a digital elevation model and a digital surface model measured before a landslide and a digital surface model measured after the landslide, Provided is a sediment movement area estimation device that performs at least an estimation step of estimating the movement area based on the difference information.

Advantages of the Invention

[0007] According to the present invention, it is possible to provide a sediment movement area estimation method, a learned model, and a sediment movement area estimation device for accurately estimating a sediment movement area. Note that the effects described in this specification are merely examples and are not limiting, and there may be other effects.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Embodiments for Carrying Out the Invention

[0009] Hereinafter, preferred embodiments for implementing the present invention will be described. The embodiments described below show an example of typical embodiments of the present invention, and the scope of the present invention is not construed narrowly thereby. Unless otherwise specified, in the drawings, "up" means the upward or upper side in the drawing, "down" means the downward or lower side in the drawing, "left" means the leftward or left side in the drawing, and "right" means the rightward or right side in the drawing. Also, for the drawings, the same or equivalent elements or members are denoted by the same reference numerals, and redundant descriptions are omitted.

[0010] The description of the present invention will be made in the following order. 1. First Embodiment According to the Present Invention (Example 1 of Earth and Sand Movement Area Estimation Method) (1) Aerial Laser Scanning or UAV Laser Scanning (2) Outline of the Present Invention (3) Trained Model (4) Hardware Configuration (5) Evaluation Results 2. Second Embodiment According to the Present Invention (Example 2 of Earth and Sand Movement Area Estimation Method) (1) Technical Content (2) Evaluation Results 3. Third Embodiment According to the Present Invention (Earth and Sand Movement Area Estimation Device)

[0011] <1. First Embodiment According to the Present Invention (Example 1 of Earth and Sand Movement Area Estimation Method)> <(1) Aerial Laser Scanning or UAV Laser Scanning> Conventionally, after a large-scale landslide disaster occurs, it is required to quickly calculate the amount of sediment that has moved in order to grasp the overall situation of the disaster and formulate a restoration plan. For calculating this amount of sediment, three-dimensional topographic and ground feature data measured using airborne laser scanning or UAV laser scanning can be used. Airborne laser scanning is a technology that emits laser light from an aircraft towards the ground surface to measure the elevations of the terrain and ground features. UAV laser scanning is a technology that emits laser light from an unmanned aerial vehicle (UAV) towards the ground surface to measure the elevations of the terrain and ground features. Note that the terrain refers to the shape of the land. The ground features refer to objects existing on the land, such as buildings, trees, and rocks.

[0012] An explanation of airborne laser scanning will be given. An aircraft for performing airborne laser scanning is equipped with a laser ranging device, a global navigation satellite system (GNSS) receiver, and an inertial measurement unit (IMU).

[0013] The laser ranging device emits laser light towards the ground surface and receives the reflected light from the ground surface. Based on the difference between the time when the laser light is emitted and the time when the reflected light is received, the distance from the aircraft to the ground surface can be measured. Thereby, the elevations of the terrain and ground features are measured. The laser ranging device can emit laser light, for example, 50,000 to 100,000 times per second and can measure the ground surface at intervals of 60 cm or less. Therefore, high-density three-dimensional topographic and ground feature data can be obtained.

[0014] The GNSS receiver receives radio waves from GNSS (Global Navigation Satellite System) satellites. Thereby, the GNSS receiver can measure the position of the aircraft.

[0015] The IMU measures the acceleration and angular velocity of the aircraft. Thereby, the attitude of the aircraft is obtained. As a result, the emission direction of the laser light can be corrected.

[0016] A method for estimating the amount of sediment moved using three-dimensional topographic and ground feature data measured by aerial laser surveying will be described. A computer generates difference information between the three-dimensional topographic and ground feature data measured before the disaster and the three-dimensional topographic and ground feature data measured after the disaster. By generating this difference information, the increase or decrease in elevation for each coordinate can be calculated. By multiplying the increase or decrease in elevation by the area, the amount of sediment moved can be estimated.

[0017] However, three-dimensional topographic and ground feature data often contains error information. For example, error information occurs when the ground itself moves due to the occurrence of a disaster, when the laser light that should reach the ground surface is blocked by tree branches and leaves and does not reach the ground surface, or when the reflected light intensity is weak and cannot be received. If the amount of sediment is estimated based on three-dimensional topographic and ground feature data containing this error information, there is a risk that areas where sediment has not moved will be misestimated as areas where sediment has moved. As a result, there is a risk that an error will occur in the estimated amount of sediment.

[0018] Therefore, visual inspection of orthoimage data is used to identify the areas where sediment has moved (sediment movement areas). In the difference information before and after the disaster, the sediment movement areas are visually identified. However, this identification requires experience and know-how. After a disaster occurs, it is necessary to quickly grasp the overall situation of the disaster and formulate a recovery plan, so it is necessary to quickly estimate the amount of sediment that has moved. Therefore, a large number of personnel without experience and know-how may be mobilized. As a result, there is a risk of variations in discrimination accuracy and delays in calculating the amount of sediment. In addition, since it requires labor costs and time, there is also a problem of increased costs. Furthermore, orthoimage data is affected by weather, seasons, etc. Therefore, there is a risk that the estimation accuracy of sediment movement areas will decrease.

[0019] Note that although UAV laser surveying is measured unmanned and aerial laser surveying is measured manned, UAV laser surveying also measures the elevation of terrain and ground features in the same way as aerial laser surveying. Therefore, the description of UAV laser surveying will be omitted.

[0020] <(2) Summary of the Invention> The three-dimensional topographic and ground feature data obtained using airborne laser surveying or UAV laser surveying includes a digital surface model (DSM), a digital elevation model (DEM), and orthoimage data. The digital surface model (DSM) refers to a model of the height of the earth's surface including ground features. The digital elevation model (DEM) refers to a model of the height of the ground excluding the height of ground features. The orthoimage data refers to image data obtained by converting aerial photographic image data into an orthographic projection.

[0021] The inventors have conducted extensive research on techniques for accurately estimating the sediment movement area and completed the present invention. The present invention is a method for estimating the sediment movement area using a computer, wherein an arithmetic unit included in the computer generates difference information by calculating the difference between each of the digital elevation model and the digital surface model measured before a sediment disaster and the digital surface model measured after the sediment disaster, and an estimation step in which the arithmetic unit estimates the movement area based on the difference information. The present invention provides a method for estimating a sediment movement area, which includes at least the above-described steps.

[0022] A method for estimating a sediment movement area according to an embodiment of the present invention will be described with reference to FIG. 1. FIG. 1 is a flowchart showing an example of a method for estimating a sediment movement area according to an embodiment of the present invention. As shown in FIG. 1, the method for estimating a sediment movement area according to an embodiment of the present invention includes at least a difference generation step (step S1) and an estimation step (step S2).

[0023] In the differential generation stage (step S1), an arithmetic unit provided in the computer generates differential information based on the difference between each of the digital elevation model (DEM) and the digital surface model (DSM) measured before the landslide disaster and the digital surface model (DSM) measured after the landslide disaster. That is, the arithmetic unit generates first differential information based on the difference between the digital elevation model (DEM) measured before the landslide disaster and the digital surface model (DSM) measured after the landslide disaster. Also, the arithmetic unit generates second differential information based on the difference between the digital surface model (DSM) measured before the landslide disaster and the digital surface model (DSM) measured after the landslide disaster.

[0024] The arithmetic unit can be, for example, a CPU (Central Processing Unit), a GPU (Graphic Processing Unit), etc. Due to the landslide disaster, the collapse and deposition of sediment occur. Therefore, the change in elevation is useful for estimating the moving area of sediment. As described above, the digital elevation model (DEM) is obtained using airborne laser scanning or UAV laser scanning.

[0025] The processing in this differential generation stage will be described with reference to FIG. 2. FIG. 2 is a schematic diagram for explaining the sediment movement area estimation method according to an embodiment of the present invention.

[0026] FIG. 2A shows the situation before the landslide disaster. An example of ground features shows forests and grasslands. In this example, the height of the ground in the digital elevation model (DEM) is constant. The height of the earth's surface in the digital surface model (DSM) varies according to the height of the ground features.

[0027] Before the landslide disaster, both the digital elevation model (DEM) and the digital surface model (DSM) can be measured using airborne laser scanning or UAV laser scanning.

[0028] Figure 2B shows the situation after the landslide disaster. Due to the landslide disaster, part of the forest and part of the grassland have disappeared. The landslide disaster causes changes in the position and height of the ground features. Therefore, the changes in the position and height of the ground features are useful for estimating the movement area of the sediment.

[0029] The area where there is no change in the ground height and the forest has disappeared is the area where the forest has been logged independently of the landslide disaster. The area where the ground has risen is the area where sediment has accumulated due to the landslide disaster. The area where the ground is depressed is the area where the ground has collapsed due to the landslide disaster.

[0030] Region a is the area where the forest has disappeared and the sediment has collapsed due to the landslide disaster. Since the forest has disappeared, the height in the Digital Surface Model (DSM) can be regarded as the height in the Digital Elevation Model (DEM). Also, since the Digital Elevation Model (DEM) before the landslide disaster has been obtained, the depth of the collapsed ground can be measured by the difference between the Digital Elevation Model (DEM) before the landslide disaster and the Digital Surface Model (DSM) after the landslide disaster.

[0031] Region b is the area where the grassland has disappeared and the sediment has collapsed due to the landslide disaster. Since the grassland has disappeared, the height in the Digital Surface Model (DSM) can be regarded as the height in the Digital Elevation Model (DEM). Also, since the Digital Elevation Model (DEM) before the landslide disaster has been obtained, the depth of the collapsed ground can be measured by the difference between the Digital Elevation Model (DEM) before the landslide disaster and the Digital Surface Model (DSM) after the landslide disaster.

[0032] In addition, the area where the ground height in the Digital Elevation Model (DEM) before the landslide disaster is lower than the ground feature height in the Digital Surface Model (DSM) after the landslide disaster and the difference between their respective heights is relatively large can be estimated as the area where the ground features have not moved.

[0033] In this way, the area where the sediment has moved can be estimated by the difference between the Digital Elevation Model (DEM) measured before the landslide disaster and the Digital Surface Model (DSM) measured after the landslide disaster.

[0034] Return to the description of FIG. 1. In the estimation step (step S2), the arithmetic unit estimates the moving area of the earth and sand based on the difference information. For this estimation, a trained model obtained by machine learning can be used. The description of this trained model will be given later.

[0035] In the present invention, a digital elevation model (DEM) measured after an earth and sand disaster is not used. The area where the earth and sand has moved is estimated from the difference between each of the digital elevation model (DEM) and the digital surface model (DSM) measured before the earth and sand disaster and the digital surface model (DSM) measured after the earth and sand disaster.

[0036] In order to generate a digital elevation model (DEM) after an earth and sand disaster, it usually takes more than one week in many cases. According to the present invention, since a digital elevation model (DEM) is not required as input information after an earth and sand disaster, the period for estimating the earth and sand moving area can be significantly shortened.

[0037] In addition, since the digital elevation model (DEM) includes the height of the ground, when a person skilled in the art calculates the earth and sand moving area, it is conceivable to compare the digital elevation model (DEM) before the earth and sand disaster with the digital elevation model (DEM) after the earth and sand disaster. On the other hand, it is extremely difficult to arrive at the idea of estimating the earth and sand moving area without using the digital elevation model (DEM) after the earth and sand disaster.

[0038] In addition, the digital surface model (DSM) includes not only the height of the ground but also the height of ground features such as forests. Therefore, simply comparing the digital surface model (DSM) with the digital elevation model (DEM) has an inhibiting factor.

[0039] From the above, it is extremely difficult for a person skilled in the art to arrive at the present invention. Despite such inventiveness difficulties, the inventors of the present invention have completed the present invention. As a result, useful effects described later are produced. From this, it is clear that the present invention has novelty and progressiveness.

[0040] The method for estimating the earth and sand movement area according to an embodiment of the present invention does not directly use the three-dimensional terrain and feature data obtained by airborne laser surveying or UAV laser surveying, but uses the data generated by preprocessing including a difference generation step (step S1). Thereby, the estimation accuracy of the earth and sand movement area is improved.

[0041] Although not shown in the figure, after the estimation step (step S2), the arithmetic unit may calculate the amount of earth and sand that has moved. For example, the arithmetic unit can calculate the amount of earth and sand based on the movement area of the earth and sand and the elevation difference information.

[0042] <(3) Trained model> In the estimation step (step S2), a trained model generated by performing machine learning in which the arithmetic unit provided in the computer uses the difference information as an input and outputs the movement area of the earth and sand can be used.

[0043] Note that machine learning is one of the learning methods for discovering certain rules from a certain data and realizing inferences and predictions for unknown data based on those rules. Known techniques can be used for machine learning. The present invention can be realized by using a computer having a machine learning function, a trained model, and the like.

[0044] In the generation of this trained model, it is preferable to select, as teacher data, the data generated by preprocessing including the difference generation step (step S1), rather than directly using the three-dimensional terrain and feature data obtained by airborne laser surveying or UAV laser surveying as teacher data. Thereby, the estimation accuracy of the trained model can be improved.

[0045] The machine learning method is not particularly limited. For example, decision tree learning such as ID3 and random forest, correlation rule learning, supervised learning, unsupervised learning, etc. may be used. Alternatively, various neural networks such as artificial neural network (ANN), deep neural network (DNN), convolutional neural network (CNN), recurrent neural network (RNN), etc. may be used. Alternatively, genetic programming (GP), inductive logic programming (ILP), fuzzy algorithm, evolutionary algorithm (EA), reinforcement learning, support vector machine (SVM), clustering, Bayesian network, etc. may be used. Furthermore, a combination of these methods or a development using deep learning technology may also be used.

[0046] For example, the case of machine learning using a neural network will be described with reference to FIG. 3. FIG. 3 is a schematic diagram showing a configuration example of a learned model according to an embodiment of the present invention.

[0047] This learned model performs an operation based on the learned weight coefficients based on the difference information generated by the difference between each of the digital elevation model and the digital surface model measured before the landslide disaster and the digital surface model measured after the landslide disaster, and is a learned model for causing a computer to function so as to output the moving area of the soil and sand.

[0048] As shown in FIG. 3, the learned model according to an embodiment of the present invention is composed of a neural network including an input layer L1, at least one intermediate layer L2, and an output layer L3.

[0049] Each layer has nodes. In this configuration example, the input layer L1 has two nodes, the intermediate layer L2 has five nodes, and the output layer L3 has two nodes. Note that the number of intermediate layers L2 and the number of nodes in each layer can be adjusted so as to obtain a preferable output result.

[0050] For machine learning, supervised learning using a DNN (Deep Neural Network) can be adopted. The explanatory variables input to the input layer L1 are the first difference information x1 and the second difference information x2. The first difference information x1 can be generated by the difference between the digital elevation model measured before the landslide disaster and the digital surface model measured after the landslide disaster. The second difference information x2 can be generated by the difference between the digital surface model measured before the landslide disaster and the digital surface model measured after the landslide disaster.

[0051] In supervised learning, the estimation target area of the landslide movement area can be divided into meshes in units of 5 m, and the correct and incorrect answers can be learned by the neural network for each mesh. The target variable output from the output layer L3 is the output information y. The output information y is a continuous value in the range of 0 to 1. When this value is 0.5 or more, it is a mesh where the earth and sand have moved, and when it is less than 0.5, it is a mesh where the earth and sand have not moved. In an actually moved mesh of earth and sand, 1 is assigned as the teacher label. In a mesh where the earth and sand have not actually moved, 0 is assigned as the teacher label.

[0052] In this way, by inputting the first difference information x1 and the second difference information x2 into the input layer L1 and outputting the presence or absence of earth and sand movement (output information y) from the output layer L3, the weighting coefficients between the nodes of each layer are learned.

[0053] <(4) Hardware Configuration> The hardware configuration of the computer used in the method for estimating the earth and sand movement area according to an embodiment of the present invention will be described with reference to FIG. 4. FIG. 4 is a block diagram showing a configuration example of a computer 50 used in the method for estimating the earth and sand movement area according to an embodiment of the present invention. As shown in FIG. 4, the computer 50 may include, as components, an arithmetic unit 101, a storage 102, a memory 103, and a display unit 104. Each component is connected, for example, by a bus as a data transmission path.

[0054] The arithmetic unit 101 is composed of, for example, a CPU, a GPU, etc. The arithmetic unit 101 controls each component included in the computer 50. The computer 50 functions by the arithmetic unit 101 reading a program that realizes each of the elevation difference generation stage (step S1), the ground object height generation stage (step S2), the color value conversion stage (step S3), and the estimation stage (step S4).

[0055] The storage 102 stores programs used by the arithmetic unit 101, control data such as arithmetic parameters, and image data, etc. The storage 102 is realized, for example, by using an HDD (Hard Disk Drive) or an SSD (Solid State Drive).

[0056] The memory 103 temporarily stores, for example, programs executed by the arithmetic unit 101. The memory 103 is realized, for example, by using a RAM (Random Access Memory).

[0057] The display unit 104 displays information. The display unit 104 can display, for example, the earth and sand movement area. The display unit 104 is realized, for example, by an LCD (Liquid Crystal Display) or an OLED (Organic Light-Emitting Diode), etc.

[0058] Although illustration is omitted, computer 50 may be provided with a communication interface. This communication interface has a function of communicating via an information communication network by using a communication technology such as Wi-Fi, Bluetooth (registered trademark), LTE (Long Term Evolution), or the like. For example, calculation unit 101 can estimate an earth and sand movement area based on image data obtained via this communication interface.

[0059] Computer 50 may be, for example, a server, or may be a smartphone terminal, a tablet terminal, a mobile phone terminal, a PDA (Personal Digital Assistant), a PC (Personal Computer), a portable music player, a portable game machine, or a wearable terminal (HMD: Head Mounted Display, glasses-type HMD, watch-type terminal, band-type terminal, etc.).

[0060] A program for realizing an estimation step (step S4) or the like may be stored in another computer device or computer system of computer 50. In this case, computer 50 can use a cloud service that provides the functions of this program. Examples of this cloud service include SaaS (Software as a Service), IaaS (Infrastructure as a Service), PaaS (Platform as a Service), and the like.

[0061] Furthermore, this program can be stored using various types of non-transitory computer readable media and supplied to a computer. Non-transitory computer readable media include various types of tangible storage media. Examples of non-transitory computer readable media include magnetic recording media (such as flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (such as magneto-optical disks), Compact Disc Read Only Memory (CD-ROM), CD-R, CD-R / W, semiconductor memories (such as mask ROM, Programmable ROM (PROM), Erasable PROM (EPROM), flash ROM, Random Access Memory (RAM)). Also, the above program may be supplied to a computer by various types of transitory computer readable media. Examples of transitory computer readable media include electrical signals, optical signals, and electromagnetic waves. Transitory computer readable media can supply the above program to a computer via wired communication channels such as electric wires and optical fibers, or wireless communication channels.

[0062] <(5) Evaluation Results> The performance of the learned model according to the present invention was evaluated. The area to be estimated was divided into 5m unit meshes, and a method of evaluating correct and incorrect answers for each mesh was used. It was defined as correct when the presence or absence of sediment movement visually determined by a human and the presence or absence of sediment movement estimated by the arithmetic unit matched.

[0063] The evaluation results of the learned model will be described with reference to FIG. 5. FIG. 5 is a table showing the evaluation results of the learned model according to an embodiment of the present invention.

[0064] "Region" indicates the region where the sediment disaster occurred. "Area" indicates a specific area in this region.

[0065] For each of these areas, the sediment movement area and the amount of produced sediment were evaluated. The sediment movement area is the area where sediment has moved. The amount of produced sediment is the amount of sediment produced.

[0066] The accuracy rate of the sediment movement area is the value obtained by dividing the number of correctly identified meshes by the total number of meshes. This accuracy rate was 82% for area A1, 83% for area A2, 91% for area H1, and 90% for area H2, resulting in a generally high accuracy rate.

[0067] Regarding the amount of produced sediment estimated by the calculation unit, it was 646,000 m 3 for area A1, 2,709,000 m 3 for area A2, 241,000 m 3 for area H1, and 711,000 m 3 for area H2.

[0068] Regarding the amount of produced sediment visually discriminated by humans, it was 719,000 m 3 for area A1, 2,955,000 m 3 for area A2, 335,000 m 3 for area H1, and 797,000 m 3 for area H2.

[0069] The error rate is the absolute value obtained by dividing the difference between the amount of sediment estimated by the calculation unit and the amount of sediment visually discriminated by humans by the amount of sediment visually discriminated by humans. This error rate was 10% for area A1, 8% for area A2, 28% for area H1, and 11% for area H2. The error rate for area H1 was slightly higher, but overall the error rate was low.

[0070] Furthermore, by using this trained model, the period required to estimate the sediment movement area was significantly shortened. For example, if the area of the area to be estimated is about 30 km 2 and there are 5 people creating the digital elevation model (DEM), it takes about 7 days to create the digital elevation model (DEM) after the sediment disaster. Furthermore, when humans visually estimate the sediment movement area based on the created digital elevation model (DEM), it takes an additional about 2 days.

[0071] On the one hand, by using this learned model, even if the area of the area to be estimated is about 30 km 2 , the period until the landslide movement area is estimated is about one day. The larger the area of the area to be estimated and the higher the collapse density, the higher the effectiveness of the present invention.

[0072] The above content described for the landslide movement area estimation method according to the first embodiment of the present invention can be applied to other embodiments of the present invention as long as there is no particular technical contradiction.

[0073] <2. Second Embodiment According to the Present Invention (Example 2 of Landslide Movement Area Estimation Method)> <(1) Technical Content> The present invention can estimate the landslide movement area with higher accuracy by using a two-stage estimation process using a deep neural network (DNN) and a convolutional neural network (CNN).

[0074] That is, the learned model includes a deep neural network (DNN) and a convolutional neural network (CNN), and the estimation stage includes a first estimation stage and a second estimation stage. In the first estimation stage, the arithmetic unit inputs the difference information into the DNN to output the movement area. Then, in the second estimation stage, the arithmetic unit inputs the difference information and the output result of the DNN into the CNN to output the movement area.

[0075] This will be described with reference to FIG. 6. FIG. 6 is a flowchart showing an example of the landslide movement area estimation method according to an embodiment of the present invention. As shown in FIG. 6, the estimation stage (see FIG. 1) includes a first estimation stage (step S21) and a second estimation stage (step S22).

[0076] Note that the number of estimation stages is not limited to two, and may be three or more. Also, neural networks other than DNN and CNN may be used.

[0077] First, in the difference generation stage (step S1), the arithmetic unit of the computer generates difference information based on the difference between each of the digital elevation model (DEM) and the digital surface model (DSM) measured before the landslide disaster, and the digital surface model (DSM) measured after the landslide disaster.

[0078] This difference information includes first difference information and second difference information. The first difference information is generated based on the difference between the digital elevation model (DEM) measured before the landslide disaster and the digital surface model (DSM) measured after the landslide disaster. The second difference information is generated based on the difference between the digital surface model (DSM) measured before the landslide disaster and the digital surface model (DSM) measured after the landslide disaster.

[0079] Let the elevation of the digital elevation model (DEM) measured before the landslide disaster be z DEM1 , let the elevation of the digital surface model (DSM) measured before the landslide disaster be z DSM1 , and let the elevation of the digital surface model (DSM) measured after the landslide disaster be z DSM2 At this time, the first difference information a1 and the second difference information a2 can be calculated using the following mathematical formulas (1) and (2) respectively.

[0080] a1 = z DSM2 - z DEM1 ···(1) a2 = z DSM2 - z DSM1 ···(2)

[0081] The first difference information a1 and the second difference information a2 will be further described with reference to FIG. 7. FIG. 7 is a schematic diagram for explaining a method for estimating a landslide movement area according to an embodiment of the present invention.

[0082] FIG. 7A shows the situation after the landslide disaster. Due to the landslide disaster, a part of the forest, which is a ground feature, has disappeared. Since the position and height of the ground feature change due to the landslide disaster, the change in the position and height of the ground feature is useful for estimating the landslide movement area.

[0083] Region R1 is an area where forests have disappeared due to landslides, and the ground is collapsing and eroding. Region R2 is an area where forests have disappeared and sediment has accumulated.

[0084] Figure 7B is an image diagram showing the first difference information a1. In this figure, the elevation z of the digital elevation model (DEM) measured before the landslide DEM1 and the elevation z of the digital surface model (DSM) measured after the landslide DSM2 are shown.

[0085] In the region where the first difference information a1 is large in the positive direction (the upward direction in the figure), since the forest has not disappeared, it can be estimated that the possibility of sediment movement is low. In region R1 where the first difference information a1 is small in the negative direction, it can be estimated that the ground is collapsing and eroding. In region R2 where the first difference information a1 is small in the positive direction, it can be estimated that sediment is accumulating.

[0086] Figure 7C is an image diagram showing the second difference information a2. In this figure, the elevation z of the digital surface model (DSM) measured before the landslide DSM1 and the elevation z of the digital surface model (DSM) measured after the landslide DSM2 are shown.

[0087] In the region where the second difference information a2 is small, since the forest has not disappeared, it can be estimated that the possibility of sediment movement is low. In regions R1 and R2 where the second difference information a2 is large in the negative direction, it can be estimated that the forest has disappeared, that is, the ground is collapsing and eroding or sediment is accumulating.

[0088] From this, in region R1, since the forest has disappeared and the ground is collapsing and eroding, the first difference information a1 becomes a slightly negative value, and the second difference information a2 becomes a large negative value. The sediment movement estimation method according to this embodiment can accurately estimate this region R1.

[0089] Return to the description of FIG. 6. Next, in the first estimation step (step S21), the calculation unit inputs the difference information to the DNN and causes it to output the moving area. Specifically, the calculation unit estimates the presence or absence of sediment movement using the DNN with the first difference information and the second difference information as input data (explanatory variables).

[0090] The DNN outputs the moving area of sediment in mesh units. The DNN performs binary classification without considering the relationship with adjacent meshes, for example, in mesh units of 5 m, and outputs whether sediment movement has occurred. This output result can be, for example, a black-and-white image indicating the presence or absence of sediment movement in mesh units.

[0091] Next, in the second estimation step (step S22), the calculation unit inputs the difference information and the output result of the DNN to the CNN and causes it to output the moving area. Specifically, the CNN performs detailed semantic segmentation with the first difference information, the second difference information, and the output result of the DNN as inputs (explanatory variables), and outputs the moving area of sediment in pixel units. This output result can be, for example, a black-and-white image indicating the presence or absence of sediment movement in pixel units.

[0092] The CNN automatically extracts features from image data and realizes high-precision classification and recognition. This eliminates the need for manual feature extraction required by conventional methods and exhibits better performance.

[0093] Also, the CNN hierarchically extracts features from low-level features (edges, colors, textures, etc.) to high-level features (objects, shapes, patterns, etc.) of the input image. Thereby, complex patterns of the image can be captured.

[0094] Furthermore, in the convolutional layer of the CNN, since the filter (kernel) is applied to the entire image, the number of parameters is significantly reduced. Thereby, the calculation cost of the model is suppressed and learning is made more efficient.

[0095] The first difference information input to the CNN is, in a specific example, the information obtained by grayscale imaging. This image can be, for example, a grayscale image in which the first difference information a1 in the range of 0 to 20 m is converted to the range of 255 to 0. For example, the minimum value (0 m) of the first difference information a1 is converted so as to correspond to the maximum value (255, white) of the grayscale. The maximum value (20 m) of the first difference information a1 is converted so as to correspond to the minimum value (0, black) of the grayscale. The value of 20 m is suitable because, in many cases, the tree height is within that range.

[0096] The intention of this conversion is based on the idea that where the first difference information a1 indicates a large value, it represents the height of the tree because no earth and sand movement has occurred, and conversely, where it indicates a small value, there is a high possibility that earth and sand movement has occurred. In the present embodiment, the possibility of earth and sand movement is increased when the value after conversion is large (white as an image), but it may also be arranged such that the possibility of earth and sand movement is increased when the value after conversion is small (black as an image).

[0097] This conversion will be described with reference to FIG. 8. FIG. 8 is a graph showing an example of a method for converting the first difference information a1 according to the present invention into a grayscale image. The horizontal axis in FIG. 8 indicates the first difference information a1 before conversion, and the vertical axis indicates the pixel value of the grayscale image after conversion. When the value of the first difference information a1 is 0 m or less, the pixel value is 255. When the value of the first difference information a1 changes in the range of 0 to 20 m, the pixel value changes in the range of 0 to 255. When the value of the first difference information a1 is 20 m or more, the pixel value is 0.

[0098] When the value of the first difference information a1 changes in the range of 0 to 20 m, the pixel value can be calculated by the following mathematical formula (3). Here, x indicates the first difference information a1, and y indicates the pixel value of the grayscale image.

[0099] y = -12.75x + 255 ···(3)

[0100] This conversion is designed such that the converted value becomes larger in places that are more likely to be areas of sediment movement. Large values in the positive direction of the first difference information a1 indicate locations where forests remain, suggesting a low likelihood of sediment movement. Small values in the positive direction of the first difference information a1 indicate that the forest has disappeared and there may be sediment deposition. Negative values of the first difference information a1 indicate that the forest has disappeared and there may be ground collapse or erosion occurring.

[0101] The second difference information input to the CNN is also information that has been grayscale imaged. This image can be, for example, a grayscale image obtained by converting the second difference information a2 in the range of -20 to 0 m to the range of 255 to 0. The intention of this conversion is based on the idea that sediment movement is likely to have occurred where the forest has disappeared, and other ideas are the same as those for the first difference information. Through this conversion, areas where sediment movement is likely to have occurred are visualized as high values (white), and areas where sediment movement is less likely to have occurred are visualized as low values (black).

[0102] This conversion will be described with reference to FIG. 9. FIG. 9 is a graph showing an example of a method for converting the second difference information a2 according to the present invention into a grayscale image. The horizontal axis of FIG. 9 represents the second difference information a2 before conversion, and the vertical axis represents the pixel value of the grayscale image after conversion. When the value of the second difference information a2 is -20 m or less, the pixel value is 255. When the value of the second difference information a2 changes in the range of -20 to 0 m, the pixel value changes in the range of 255 to 0. When the value of the second difference information a2 is 0 m or more, the pixel value is 0.

[0103] When the value of the second difference information a2 changes in the range of -20 to 0 m, the pixel value is represented by the following mathematical formula (4). Here, x represents the second difference information a2, and y represents the pixel value of the grayscale image.

[0104] y = -12.75x ···(4)

[0105] This conversion is designed such that the converted value becomes larger in places that are more likely to be areas of sediment movement. A large negative value in the second difference information a2 indicates that sediment may be deposited or that ground collapse or erosion may be occurring.

[0106] In the second estimation stage (step S22), it is preferable that the arithmetic unit further inputs the MPI (Morphometric Protection Index) into the CNN to output the sediment movement area. That is, it is preferable that the arithmetic unit inputs the first difference information a1, the second difference information a2, the output result of the DNN, and the MPI into the CNN to output the sediment movement area.

[0107] MPI is an index for quantitatively evaluating how well a specific point is protected by the surrounding terrain in terrain analysis. In other words, MPI analyzes the terrain around each cell up to a certain distance and quantifies how well that point is protected by the surrounding terrain.

[0108] Specifically, using a digital elevation model (DEM), the height of the terrain around each cell is evaluated to calculate the degree of protection. MPI is a concept equivalent to positive openness and is an index for evaluating the spread of the field of view. Positive openness indicates how open the field of view is, and MPI evaluates how well a specific point is protected by the surrounding terrain based on this.

[0109] For example, the sky view factor (SVF) is measured in the range of 0° to 180°, and a large value (i.e., a wide open space) is obtained at high places such as ridges, while a small value (i.e., a narrow open space) is obtained at low places such as valleys. On the other hand, MPI is represented in the range of 0 to 1, taking a smaller value at ridges and a larger value at valleys. This indicates that it is an index showing how well a point is "protected" by the surrounding terrain, that is, how well the surrounding terrain protects against external factors such as wind and flow.

[0110] By inputting this MPI into a CNN, it is possible to learn and infer that "there is a high possibility that sediment is deposited in the valley near the sediment production area (the area where the ground is collapsing and eroding)".

[0111] MPI can be calculated, for example, by SAGA GIS (System for Automated Geoscientific Analyses). SAGA GIS is an open-source geographic information system (GIS) used for the editing and analysis of spatial data.

[0112] The calculation procedure of MPI will be described with reference to FIG. 10. FIG. 10 is a flowchart showing an example of the calculation procedure of MPI according to the present invention.

[0113] As shown in FIG. 10, first, in step S31, a digital elevation model (DEM) of the target area is prepared. This data can be acquired by aerial laser surveying, UAV laser surveying, or the like.

[0114] Next, in step S32, a range for analyzing the terrain around each cell (for example, a radius of 2000 meters) is set. The terrain characteristics within this range are evaluated.

[0115] Next, in step S33, the height of the terrain around each cell is evaluated, and its MPI is quantified. Specifically, the higher the height of the surrounding terrain, the more that point is considered to be protected.

[0116] Finally, in step S34, the calculated MPI values are converted into a grayscale image. The conversion from MPI to a grayscale image will be described with reference to FIG. 11. FIG. 11 is a graph showing an example of a method for converting MPI to a grayscale image according to the present invention. The horizontal axis of FIG. 11 represents MPI, and the vertical axis represents the values of the grayscale image.

[0117] Using the graph shown in FIG. 11, the MPI in the range of 0 to 1 is converted into a grayscale image in the range of 0 to 255. The equation of the straight line passing through the two points (0, ) and (1, 255) is shown by the following mathematical formula (5). Here, x represents the MPI, and y represents the value of the grayscale image.

[0118] y = 255x ···(5)

[0119] By this conversion, the area showing a high MPI is visualized as a high value (white), and the area showing a low MPI is visualized as a low value (black). For example, the valley shows a high value, and the ridge shows a low value. The higher the value of the grayscale image, the higher the possibility of being an earth and sand movement area.

[0120] <(2) Evaluation Results> To evaluate the performance of the learned model according to the present invention, the following method was used. The area to be estimated was divided into meshes, and the correct and incorrect answers were evaluated for each mesh. A case where the presence or absence of earth and sand movement visually determined by a human and the presence or absence of earth and sand movement estimated by the calculation unit match was defined as a correct answer.

[0121] The confusion matrix in this evaluation will be described with reference to FIG. 12. FIG. 12 is a table showing the confusion matrix in the evaluation of the learned model according to an embodiment of the present invention. The confusion matrix is a table used to evaluate the performance of the learned model. By comparing the inference result and the actual result, the number of correctly classified and the number of misclassified are classified into four categories and organized. Thereby, the accuracy and error tendency of the learned model can be clearly grasped. The following is an explanation of the components of the confusion matrix and their meanings.

[0122] TP (True Positive, hit) is a case where both the inference result and the correct value are "earth and sand movement present". TP is a case where the inference is correct and the mesh where there is actually earth and sand movement is correctly detected.

[0123] FN (False Negative, detection failure) occurs when the inference result is "no landslide movement" while the correct value is "landslide movement". FN means that the inference is incorrect, and although there was actually a landslide movement, it could not be detected.

[0124] FP (False Positive, false detection) occurs when the inference result is "landslide movement" while the correct value is "no landslide movement". FP means that the inference is incorrect, and although there was actually no landslide movement, it was erroneously detected as having a landslide movement.

[0125] TN (True Negative, correct detection) occurs when both the inference result and the correct value are "no landslide movement". TN means that the inference is correct and a mesh with no actual landslide movement was correctly judged.

[0126] Using this confusion matrix, the metrics "Accuracy (correct rate)", "Precision (precision rate)", "Recall (recall rate)", and "IoU" for evaluating the performance of the trained model were calculated.

[0127] Accuracy (correct rate) is the ratio of the meshes correctly inferred among all meshes. Accuracy can be calculated using the following formula (6).

[0128] Accuracy = (TP + TN) / (TP + FP + FN + TN) ···(6)

[0129] Precision (precision rate) is the ratio of the meshes with actual landslide movement among those inferred to have landslide movement. Precision can be calculated using the following formula (7).

[0130] Precision = TP / (TP + FP) ···(7)

[0131] Recall (recall rate) is the ratio of the meshes correctly inferred among those with actual landslide movement. Recall can be calculated using the following formula (8).

[0132] Recall = TP / (TP + FN) ···(8)

[0133] IoU (Intersection over Union) is the ratio of the meshes where the landslide is correctly inferred. IoU can be calculated using the following formula (9).

[0134] IoU = TP / (TP + FP + FN) ···(9)

[0135] By using these indicators, the accuracy of the landslide inference results was comprehensively evaluated. In the areas where landslide disasters actually occurred, the above method was applied to estimate the landslide area. The evaluation results of this estimation will be described with reference to FIG. 13. FIG. 13 is a table showing the evaluation results of the learned model according to an embodiment of the present invention.

[0136] This table shows the accuracy evaluation indicators for the landslide areas in the first embodiment and the second embodiment. The degree of performance improvement of the learned model according to the first embodiment and the learned model according to the second embodiment was confirmed.

[0137] Regarding the accuracy rate, the first embodiment is 0.889, while the second embodiment is 0.957, showing an improvement. Regarding the precision rate, the first embodiment is 0.440, while the second embodiment is 0.793, showing a significant improvement. Regarding the recall rate, the first embodiment is 0.727, while the second embodiment is 0.727, showing a slight improvement. Regarding IoU, the first embodiment is 0.376, while the second embodiment is 0.611, showing a significant improvement.

[0138] Thus, it was confirmed that in the second embodiment, all of these indicators are improved. This shows that the learned model according to the second embodiment has higher accuracy than the learned model according to the first embodiment.

[0139] Furthermore, while referring to FIGS. 14 and 15, the evaluation results are visually explained. FIGS. 14 and 15 are images showing examples of estimation results of the earth and sand movement area estimation method according to an embodiment of the present invention.

[0140] FIG. 14A shows the correct answer, that is, the actual earth and sand movement area. The white part indicates the area where earth and sand movement has occurred, and the black part indicates the area where earth and sand movement has not occurred.

[0141] FIG. 14B shows the estimation result using the learned model according to the first embodiment. The white part indicates the area where earth and sand movement is estimated by the learned model. When compared with the correct answer image in FIG. 14A, misdetection pixels (noise) can be confirmed.

[0142] FIG. 14C shows the estimation result using the learned model according to the second embodiment. The noise has significantly decreased, and the estimation result is approaching the correct answer image. By comparing with the image in FIG. 14B, it can be confirmed that the misdetection has decreased and the estimation accuracy of the earth and sand movement area has improved.

[0143] FIG. 15 is an enlarged view of a specific part of the image shown in FIG. 14. FIG. 15A is the correct answer image, accurately representing the range where earth and sand movement actually occurred. The white pixels indicate the actual earth and sand movement area. The gray contour line indicates the boundary of the earth and sand movement area.

[0144] FIG. 15B shows the estimation result using the learned model according to the first embodiment. Misdetection pixels (noise) can be confirmed, and many white pixels are also displayed in places where there is no earth and sand movement. Also, many black pixels exist within the gray contour line. That is, FN (detection omission) is increasing.

[0145] Figure 15C shows the estimation result using the learned model according to the second embodiment. The number of misdetected pixels has significantly decreased, and the estimation result is approaching the correct data. The noise has decreased, enabling accurate estimation of the earth and sand movement area. Also, the number of black pixels within the gray contour line has decreased compared to Figure 15B. That is, FN (detection omission) has decreased.

[0146] By using the learned model according to the second embodiment, the period until estimating the earth and sand movement area has been significantly shortened. Calculated based on the Surveying Business Standard Steps of the Ministry of Land, Infrastructure, Transport and Tourism (2023 edition), it takes about 5 to 7 days for 5 workers to create ground data for about 20 - 30 km. 2 Furthermore, when visually interpreting the earth and sand movement area while measuring time again, it took about 2 days. Therefore, the conventional method requires a total of about 7 to 9 days from data creation to grasping the earth and sand movement area.

[0147] On the other hand, in the present invention, when the analysis was performed using a virtual desktop environment equipped with a GPU (NVIDIA Tesla M60), the inferences in the first estimation stage and the second estimation stage were each completed in a few minutes. Even including the creation of input data and the arrangement of inference results, the entire process could be completed in 1 day. This enables rapid response after a disaster occurs.

[0148] From the above points, it is shown that the present invention realizes a significant improvement in efficiency and speed compared to the conventional method in grasping the earth and sand movement area, and has high practical value.

[0149] The above - described content regarding the method for estimating the earth and sand movement area according to the second embodiment of the present invention can be applied to other embodiments of the present invention as long as there is no particular technical contradiction.

[0150] <3. The Third Embodiment of the Present Invention (Apparatus for Estimating Earth and Sand Movement Area)> The present invention at least includes an arithmetic unit for estimating the moving area of earth and sand, and the arithmetic unit performs at least a difference generation step of generating difference information based on the difference between each of the digital elevation model and the digital surface model measured before the earth and sand disaster and the digital surface model measured after the earth and sand disaster, and an estimation step of estimating the moving area based on the difference information, and provides an earth and sand moving area estimation device.

[0151] A configuration example of an earth and sand moving area estimation device according to an embodiment of the present invention may be the configuration example shown in FIG. 4. As shown in FIG. 4, the earth and sand moving area estimation device 50 at least includes an arithmetic unit 101 for estimating the moving area of earth and sand.

[0152] The arithmetic unit 101 performs a difference generation step of generating difference information based on the difference between each of the digital elevation model and the digital surface model measured before the earth and sand disaster and the digital surface model measured after the earth and sand disaster. Next, the arithmetic unit 101 performs an estimation step of estimating the moving area of earth and sand based on this difference information. In this estimation step, the arithmetic unit 101 can use the above-described learned model. This learned model can be recorded in, for example, the storage 102.

[0153] The above content described for the earth and sand moving area estimation device according to the third embodiment of the present invention can be applied to other embodiments of the present invention as long as there is no particular technical contradiction.

[0154] Note that the present invention can also have the following configuration. [1] An earth and sand moving area estimation method for estimating the moving area of earth and sand using a computer, a difference generation step in which an arithmetic unit included in the computer generates difference information based on the difference between each of the digital elevation model and the digital surface model measured before the earth and sand disaster and the digital surface model measured after the earth and sand disaster, and an estimation step in which the arithmetic unit estimates the moving area based on the difference information, and the earth and sand moving area estimation method includes at least these steps. [2] Each of the numerical elevation model and the numerical surface model is obtained using aerial laser surveying or UAV laser surveying. The method for estimating the earth and sand movement area according to [1]. [3] In the estimation step, the arithmetic unit estimates using a learned model generated by performing machine learning that takes the difference information as an input and outputs the movement area of the earth and sand. The method for estimating the earth and sand movement area according to [1] or [2]. [4] The learned model includes a deep neural network (DNN) and a convolutional neural network (CNN). The estimation step includes a first estimation step and a second estimation step. In the first estimation step, the arithmetic unit inputs the difference information into the DNN to output the movement area. In the second estimation step, the arithmetic unit inputs the difference information and the output result of the DNN into the CNN to output the movement area. The method for estimating the earth and sand movement area according to [3]. [5] The DNN outputs the movement area in mesh units. The CNN outputs the movement area in pixel units. The method for estimating the earth and sand movement area according to [4]. [6] The difference information includes first difference information and second difference information. The first difference information is generated by the difference between the numerical elevation model measured before the earth and sand disaster and the numerical surface model measured after the earth and sand disaster. The second difference information is generated by the difference between the numerical surface model measured before the earth and sand disaster and the numerical surface model measured after the earth and sand disaster. The method for estimating the earth and sand movement area according to [4] or [5]. [7] In the second estimation step, the arithmetic unit further inputs MPI (Morphometric Protection Index) into the CNN to output the movement area. The method for estimating the earth and sand movement area described in [6]. [8] It is composed of a neural network including an input layer, at least one intermediate layer, and an output layer. The first difference information generated by the difference between the digital elevation model measured before the earth and sand disaster and the digital surface model measured after the earth and sand disaster, the digital surface model measured before the earth and sand disaster, and the The second difference information generated by the difference between the digital surface model measured after the earth and sand disaster and the digital surface model measured after the earth and sand disaster are input into the input layer, and the presence or absence of earth and sand movement is output from the output layer, so that the The weighting coefficients between the nodes of each layer are learned. Based on the difference information generated by the difference between each of the digital elevation model and the digital surface model measured before the earth and sand disaster and the digital surface model measured after the earth and sand disaster, a A learned model for causing a computer to function so as to perform an operation based on the learned weighting coefficient and output the earth and sand movement area. [9] It includes at least an arithmetic unit for estimating the earth and sand movement area. The arithmetic unit A difference generation stage for generating difference information by the difference between each of the digital elevation model and the digital surface model measured before the earth and sand disaster and the digital surface model measured after the earth and sand disaster, An earth and sand movement area estimation device that performs at least an estimation stage of estimating the movement area based on the difference information.

Explanation of symbols

[0155] S1 Difference generation stage S2 Estimation stage S21 First estimation stage S22 Second estimation stage 50 Computer (earth and sand movement area estimation device) 101 Arithmetic unit 102 Storage 103 Memory 104 Display unit

Claims

1. A method for estimating a sediment movement area using a computer, comprising: a difference generating step in which a calculation unit of the computer generates difference information based on the difference between the digital elevation model and the digital surface model measured before the landslide disaster and the digital surface model measured after the landslide disaster; The method for estimating a soil movement area includes at least an estimation step in which the calculation unit estimates the movement area based on the difference information.

2. Each of the digital elevation model and the digital surface model is obtained using airborne laser surveying or UAV laser surveying. The method for estimating a sediment movement area according to claim 1.

3. In the estimation step, the calculation unit performs machine learning using a trained model generated by using the difference information as an input and a soil movement area as an output, The method for estimating a sediment movement area according to claim 1 or 2.

4. The trained model includes a deep neural network (DNN) and a convolutional neural network (CNN); the estimation step includes a first estimation step and a second estimation step; In the first estimation step, the calculation unit inputs the difference information to the DNN and outputs the movement range; In the second estimation step, the calculation unit inputs the difference information and the output result of the DNN to the CNN to output the movement range. The method for estimating an area of ​​sediment movement according to claim 3.

5. The DNN outputs the movement area in units of meshes, The CNN outputs the motion area in pixels. The method for estimating an area of ​​sediment movement according to claim 4.

6. the difference information includes first difference information and second difference information, The first difference information is generated by a difference between a digital elevation model measured before the landslide disaster and a digital surface model measured after the landslide disaster; The second difference information is generated based on a difference between a digital surface model measured before the landslide disaster and a digital surface model measured after the landslide disaster. The method for estimating an area of ​​sediment movement according to claim 4.

7. In the second estimation step, the calculation unit further inputs a Morphometric Protection Index (MPI) to the CNN to output the movement range. The method for estimating an area of ​​sediment movement according to claim 6.

8. The neural network includes an input layer, at least one intermediate layer, and an output layer. a first difference information generated by subtracting a digital elevation model measured before the landslide disaster from a digital surface model measured after the landslide disaster, and a second difference information generated by subtracting a digital surface model measured before the landslide disaster from a digital surface model measured after the landslide disaster are input to the input layer, and the presence or absence of sediment movement is output from the output layer, thereby learning weighting coefficients between the nodes in each layer; A trained model for causing a computer to function to perform calculations based on the trained weighting coefficients based on differential information generated by subtracting the digital elevation model and digital surface model measured before the landslide disaster from the digital surface model measured after the landslide disaster, and output the area of ​​sediment movement.

9. The system includes at least a calculation unit for estimating a soil movement area, The calculation unit, a difference generation stage for generating difference information by differences between the digital elevation model and the digital surface model measured before the landslide disaster and the digital surface model measured after the landslide disaster; and an estimation step of estimating the movement area based on the difference information.

Images