River image analysis method, river image analysis device, and river image analysis program

Abstract

To provide a river image analysis method, a river image analysis device, and a river image analysis program that analyze changes in river topography quickly and with high accuracy. [Solution] The river image analysis method includes: a preprocessing step S1 in which a computer's processing unit performs preprocessing on satellite images of a river; a classification step S2 in which a trained model, which has been trained by machine learning to determine the correlation between multiple satellite images of a river and the class to which each region included in the satellite image belongs, classifies each region included in the satellite image into a class; an analysis step S3 in which the time-series change pattern of the region is analyzed based on the classification result in classification step S2; and an extraction step S4 in which regions of interest are extracted from the satellite image based on the time-series change pattern in analysis step S3.

Description

Technical Field

[0001] The present invention relates to river image analysis technology using a computer, and specifically relates to a method, apparatus, and program for analyzing topographic changes of a river using images.

Background Art

[0002] In recent years, in preparation for an increase in disasters due to climate change, maintenance management for ensuring the safe flow of rivers has been demanded. Changes in rivers and their surrounding areas are monitored by regular surveys.

[0003] For example, in Patent Document 1, a technique related to "a river management navigation system characterized in that monitoring target information including a combination of monitoring points and monitoring items necessary for river maintenance management is stored in a portable information terminal with a GPS function, and the GPS is interlocked with the monitoring target information to navigate an inspection worker to the monitoring point, and when the monitoring point is reached, the monitoring item indicating the monitoring work content to be executed is guided to the inspection worker by an image and voice, and the monitoring result inspected by the inspection worker according to this can be input to the portable information terminal on the spot" is disclosed.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, while conventional technologies utilized high-resolution satellite and aerial imagery, the low frequency of acquisition and analysis limited their ability to efficiently monitor vegetation and topographic changes over wide areas. In particular, interpreting vegetation patterns was difficult to perform frequently over large areas due to cost and time constraints. As a result, these technologies failed to adequately address the need for early detection of vegetation and topographic changes in short periods and for more accurate analysis.

[0006] Therefore, the main objective of this invention is to analyze changes in river topography quickly and with high accuracy. [Means for solving the problem]

[0007] The present invention A river image analysis method that uses a computer to analyze the condition of a river based on images taken of the river, A classification step is performed to classify each region in the satellite image into a class, using a pre-trained model that has been trained by machine learning to analyze the correlation between multiple satellite images of rivers and the class to which each region in the satellite image belongs. Based on the classification results in the aforementioned classification stage, an analysis stage is performed to analyze the time-series change pattern of the aforementioned region. The present invention provides a river image analysis method that includes an extraction step of extracting a region of interest in satellite imagery based on the time-series change pattern in the aforementioned analysis step. The satellite images may be images in the near-infrared wavelength band and the visible light wavelength band. The aforementioned time-series change pattern may be a change from trees to gravelly land. The aforementioned machine learning is supervised learning using training data, The classes included in the training data may be classified in more detail than the classes used to classify the trained model. The river image analysis method may further include a detailed classification step in which a trained model, which has been machine-trained on multiple aerial images with a higher resolution than the satellite image, classifies each region included in the aerial image corresponding to the region of interest into a class. The aerial image may be an image taken during aerial laser surveying or UAV laser surveying. The aerial image may be a color image captured in color. The river image analysis method may further include a river cross-section evaluation step in which the river cross-section is evaluated based on the classification results in the detailed classification step and topographic data. The terrain data may be at least one of 3D point cloud data, a digital elevation model (DEM), and a digital surface model (DSM). The river image analysis method may further include a levee breach risk assessment step, which evaluates the level of levee breach risk based on the levee protection line and the time-series change pattern in the analysis step. Furthermore, the present invention is A river image analysis device that uses a computer to analyze the condition of a river based on images taken of the river, A classification unit that classifies each region in a satellite image into a class using a pre-trained model that has been machine-learned to determine the correlation between multiple satellite images of rivers and the class to which each region in the satellite image belongs. An analysis unit analyzes the time-series change pattern of the region based on the classification results from the classification unit, The present invention provides a river image analysis device comprising an extraction unit that extracts regions of interest in satellite images based on time-series change patterns obtained by the aforementioned analysis unit. Furthermore, the present invention is A river image analysis program that uses a computer to analyze the condition of a river based on images taken of the river, A classification step is performed to classify each region in the satellite image into a class, using a pre-trained model that has been trained by machine learning to analyze the correlation between multiple satellite images of rivers and the class to which each region in the satellite image belongs. Based on the classification results in the aforementioned classification stage, an analysis stage is performed to analyze the time-series change pattern of the aforementioned region. The present invention provides a river image analysis program that causes a computer to perform an extraction step, which extracts regions of interest from satellite images based on the time-series change patterns in the aforementioned analysis step.

Advantages of the Invention

[0008] According to the present invention, the topographical changes of a river can be analyzed in a short period with high accuracy. Note that the effects described in this specification are merely examples and are not limited, and there may be other effects.

Brief Description of the Drawings

[0009] [Figure 1] It is a flowchart showing an example of the procedure of a river image analysis method according to an embodiment of the present invention. [Figure 2] It is a schematic diagram showing an example of an image used in the river image analysis method according to an embodiment of the present invention. [Figure 3] It is a schematic diagram showing an example of an image used in the river image analysis method according to an embodiment of the present invention. [Figure 4] It is a schematic diagram showing an example of an image used in the river image analysis method according to an embodiment of the present invention. [Figure 5] It is a schematic diagram showing an example of an image used in the river image analysis method according to an embodiment of the present invention. [Figure 6] It is a block diagram showing a configuration example of a computer 50 used in the river image analysis method according to an embodiment of the present invention. [Figure 7] It is a flowchart showing an example of the procedure of a river image analysis method according to an embodiment of the present invention. [Figure 8] It is a flowchart showing an example of the procedure of a river image analysis method according to an embodiment of the present invention. [Figure 9] It is a flowchart showing an example of the procedure of a river image analysis method according to an embodiment of the present invention. [Figure 10] It is a block diagram showing a configuration example of a river image analysis apparatus 1 according to an embodiment of the present invention.

Modes for Carrying Out the Invention

[0010] Hereinafter, preferred embodiments for implementing the present technology will be described. The embodiments described below show an example of a typical embodiment of the present technology, and thus the scope of the present technology should not be construed narrowly. Unless otherwise specified, in the drawings, "above" means the upward or upper direction in the figure, "below","downward" or in the figure, "left" means the leftward or left side in the figure, and "right" means the rightward or right side in the figure. Also, for the drawings, the same or equivalent elements or members are denoted by the same reference numerals, and redundant descriptions are omitted.

[0011] [[ID=③]] The description of the present invention will be made in the following order. 1. First Embodiment According to the Present Invention (Example 1 of River Image Analysis Method) (1) Overview (2) Pretreatment Stage (3) Classification Stage (4) Analysis Stage (5) Extraction Stage (6) Hardware Configuration 2. Second Embodiment According to the Present Invention (Example 2 of River Image Analysis Method) 3. Third Embodiment According to the Present Invention (Example 3 of River Image Analysis Method) 4. Fourth Embodiment According to the Present Invention (Example 4 of River Image Analysis Method) 5. Fifth Embodiment According to the Present Invention (Example of River Image Analysis Apparatus) 6. Sixth Embodiment According to the Present Invention (Example of River Image Analysis Program)

[0012] <1. First Embodiment According to the Present Invention (Example 1 of River Image Analysis Method)> <(1) Overview> The present invention uses hardware and software to perform image analysis in river maintenance management, and efficiently quantifies the topographic changes of rivers to assist the decision-making of river managers. Specifically, according to the present invention, the surface cover obtained from images such as satellite images and aerial images can be quantified and classified with high accuracy.

[0013] The present invention is a river image analysis method that uses a computer to analyze the condition of a river based on images taken of the river. This river analysis method will be explained with reference to Figure 1. Figure 1 is a flowchart showing an example of the procedure for a river image analysis method according to one embodiment of the present invention.

[0014] As shown in Figure 1, a river image analysis method according to one embodiment of the present invention includes a preprocessing step (step S1), a classification step (step S2), an analysis step (step S3), and an extraction step (step S4).

[0015] <(2) Pre-treatment stage> In the preprocessing stage (step S1), the computer's processing unit performs preprocessing on satellite images of rivers. Specifically, the processing unit converts the satellite images into a format that can be input to a pre-trained model (described later) and adjusts the resolution, color, etc. The processing unit may also optimize the satellite images to suit the characteristics of each region included in the satellite image (for example, water bodies, gravelly areas, grasslands, trees, etc.).

[0016] Satellite imagery is images taken by cameras and sensors mounted on artificial satellites, utilizing electromagnetic waves of various wavelengths, including visible light, infrared, and microwaves. Because artificial satellites take images from high altitudes, they can capture a wide area at once. As artificial satellites orbit in fixed orbits, they can periodically photograph the same location. By analyzing the reflection characteristics of electromagnetic waves of various wavelengths, it is possible to understand more detailed changes in the Earth's surface cover that are difficult to discern with the resolution of visible light images. Satellite imagery is easier and cheaper to obtain than aerial imagery, which will be discussed later.

[0017] To quantify changes in river topography, it is preferable that the ground resolution of the satellite imagery be a predetermined value. Furthermore, even when satellite imagery with different resolutions is used, it is preferable to optimize the learning model to absorb the differences in resolution so as not to affect the analysis results. In this way, it is possible to maintain analysis accuracy while supporting a wide range of resolutions.

[0018] Ground resolution is an indicator of how clearly fine details and specific objects on the ground can be seen in images acquired from satellites, aircraft, etc. The higher the ground resolution, the smaller the pixels in the image, and the clearer the display of fine details and small features on the ground.

[0019] In particular, satellite images are in the near-infrared wavelength band and the visible light wavelength band (red, green, and blue). Near-infrared images can capture the characteristics of vegetation and ground cover more clearly. Plants have the property of reflecting near-infrared light, and by utilizing this property, it becomes easier to distinguish between trees, grasslands, and bodies of water. Therefore, by using near-infrared images, changes in vegetation and ground cover around rivers can be effectively detected. On the other hand, visible light wavelength images are useful for general visual classification. By combining these images, changes in vegetation and ground cover around rivers can be detected more accurately and effectively.

[0020] (3) Classification Stages Next, in the classification stage (step S2), the computer's processing unit uses a pre-trained model, which has been trained through machine learning, to classify each region in the satellite images into a class based on the correlation between multiple satellite images of rivers and the class to which each region in those images belongs. This allows for a quantitative understanding of each region and its changes in the ground surface cover.

[0021] This machine learning can be, for example, supervised learning using training data. A trained model can be generated by supervised learning, which takes satellite imagery and corresponding label data as input data, and outputs a segmentation map. Supervised learning is a method that learns the relationship between input data called training data and the desired pairs. Based on the training data, a model can be built that can predict the appropriate output for new input data.

[0022] Label data is a system that labels each pixel in a satellite image, indicating which class it belongs to. Examples of classes include water bodies, gravelly areas, grasslands, and trees. The segmentation map shows which class each pixel in the input satellite image belongs to.

[0023] Examples of classes included in the training data include water bodies, gravelly areas, grasslands, trees, gravelly / grassy areas, and others.

[0024] Alternatively, machine learning can be unsupervised learning. Unsupervised learning uses unlabeled images. A trained model can be generated by training a model on patterns in these images. The trained model can analyze patterns in the images and automatically classify clusters such as bodies of water, gravel, vegetation, and trees.

[0025] Alternatively, machine learning may involve reinforcement learning. In reinforcement learning, a trained model learns how to classify river images based on rewards and penalties. Rewards can be set for each class, such as water bodies, gravel areas, grasslands, and trees, and the trained model can earn rewards for making correct classifications.

[0026] Pre-trained models can be constructed using techniques such as deep neural networks (DNNs), multilayer perceptrons (MLPs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), linear regression, logistic regression, support vector machines, decision trees, and random forests.

[0027] <(4) Analysis Stage> Next, in the analysis stage (step S3), the calculation unit analyzes the time-series change patterns of each region included in the satellite image based on the classification results in the classification stage (step S2). For example, the calculation unit detects increases in trees or changes in water bodies in a particular region. To analyze the time-series change patterns, it is preferable that multiple satellite images taken at regular intervals are input into the trained model.

[0028] This analysis stage will be explained with reference to Figures 2 to 5. Figures 2 to 5 are schematic diagrams showing examples of images used in the river image analysis method according to one embodiment of the present invention. Figures 2 to 5 show images of a river. These images may be satellite images or aerial images. Multiple images of the river at different time periods are taken in order to analyze the time-series change patterns. Although Figures 2 to 5 each show images from two time periods, the number of time periods is not particularly limited. Depending on the nature of the changes to be analyzed, it is preferable to analyze the time-series change patterns using images from three or more time periods rather than two.

[0029] Figure 2A shows images taken in previous years. Figure 2B shows images taken in the most recent year. In each image, a body of water A1, a gravelly area A2, a grassland A3, and trees A4 are shown. To analyze topographic changes, the area is divided into multiple regions in the direction of river flow, at intervals of approximately 10 meters. It is also divided into multiple regions (five in this example) in the direction transverse to the river. Note that there are no particular limitations on the direction of division or the size of each region.

[0030] Focusing on area A enclosed by the dotted line in Figure 2A, we can see that in the low-water channel, the area that was mainly water body A1 and gravelly area A2 in Figure 2A has changed to mainly grassland A3 and trees A4 in Figure 2B. This change in the pattern of ground cover suggests the possibility of a decrease in river cross-section (sediment deposition within the low-water channel) due to topographic changes in the low-water channel. The low-water channel is the lowest part in the center of a river where water flows under normal conditions. A decrease in river cross-section reduces the capacity of the water body, increasing the risk of flooding in the event of heavy rainfall or sudden downpours. Since low-water channels play an important role in ecosystems, such topographic changes can have adverse effects on ecosystems. A decrease in the low-water channel could prevent the local water system from functioning properly.

[0031] Next, focusing on area A enclosed by the dotted line in Figure 3, we see that in Figure 3A, the area that was mainly grassland A3 and trees A4 near the embankment has changed to mainly water area A1 in Figure 3B. This change indicates the possibility that the riverbank has been eroded (or that erosion has progressed). This change suggests that protective facilities and countermeasures will be necessary. Riverbank erosion is a phenomenon in which the riverbank is worn away by the river's water flow and waves. Natural topographic changes and changes in weather conditions may be involved in this. As riverbank erosion progresses, sediment may be washed away, and land may be lost. In addition, as erosion progresses, the river's water area may expand, potentially increasing the risk of embankment breaches during floods. Protective facilities refer to structures and methods used to protect surrounding areas and embankments from riverbank erosion. For example, revetments, foundation reinforcement works, stone masonry, and vegetation installation are used as protective facilities.

[0032] Next, focusing on area A enclosed by the dotted line in Figure 4, in the low-water channel, the area that was mainly water body A1 in Figure 4A has changed to a sandbar mainly composed of gravel A2 and trees A4 in Figure 4B. This change increases the likelihood of the sandbar becoming fixed and the low-water channel shrinking, which in turn makes drift more likely and may also accelerate erosion. Sandbar fixation refers to the reduction in the process by which sandbars are formed and changed by water flow and waves, resulting in a somewhat stable shape and location. Shrinkage of the low-water channel refers to a decrease in the low-water portion of a river or body. This can reduce the capacity of the channel and restrict water flow. Drift refers to the bias of water flow in one direction. As sandbar fixation and low-water channel shrinkage progress, water flow becomes more likely to bias in a particular direction, potentially causing drift. Scouring is the phenomenon of sediment being washed away from the bottom of rivers and bodies of water. As sandbars become more fixed and drift occurs more frequently, erosion is more likely to occur. If these conditions worsen, they could have an impact on the local water environment and ecosystem.

[0033] Next, focusing on area A enclosed by the dotted line in Figure 5, we see that near the embankment, the area that was mainly trees A4 in Figure 5A has changed to mainly sandy gravel A2 in Figure 5B. This time-series change pattern, from trees to sandy gravel, is an important indicator of the progression of riverbank erosion, and as a result, protective facilities and countermeasures are expected to be necessary. Furthermore, this invention considers not only changes in water bodies but also changes from trees to sandy gravel as important indicators of erosion. This change signifies the loss of the soil retention function provided by tree roots, indicating the weakening of the riverbank. A key feature of this invention is that it also focuses on topographic changes other than water bodies, enabling a broader assessment of river erosion risk.

[0034] <(5) Extraction stage> Returning to the explanation of Figure 1, in the extraction stage (step S4), the calculation unit extracts regions of interest in the satellite image based on the time-series change patterns in the analysis stage (step S3). These regions of interest may be areas of change that warrant attention, such as region A shown in Figures 2 to 5. These changes include abnormal changes related to the deterioration of river function, etc.

[0035] Although not shown in the diagram, the calculation unit can notify the user of alerts based on the time-series change pattern in the analysis stage (step S3). For example, if there is a possibility that the low-water channel will decrease in the region of interest, a message indicating this can be displayed, or the region of interest can be displayed on the screen.

[0036] According to the present invention, by quantitatively evaluating changes in river topography using a trained model and extracting areas of change that require attention, topographic changes can be analyzed quickly and with high accuracy.

[0037] In particular, the technology of the present invention is characterized by its ability to estimate the possibility of abnormal changes such as erosion, which are difficult to grasp with conventional methods, even regarding changes in the topography below the water surface.

[0038] By analyzing image data captured from above, the progression of scouring, riverbed erosion, or abnormal changes around river structures can be estimated based on surrounding topographic changes and water flow patterns. This allows for efficient identification of potential risks hidden not only on the ground surface but also underwater and beneath the water surface, enabling early risk assessment. Specifically, based on the captured images, machine learning algorithms can be used to analyze topographic change patterns and automatically identify areas where abnormalities may be occurring.

[0039] <(6) Hardware Configuration> The hardware configuration of the computer used by the river image analysis method according to one embodiment of the present invention will be described with reference to Figure 6. Figure 6 is a block diagram showing an example configuration of the computer 50 used by the river image analysis method according to one embodiment of the present invention. As shown in Figure 6, the computer 50 may include components such as a calculation unit 101, storage 102, memory 103, and display unit 104. Each component is connected, for example, by a bus which serves as a data transmission path.

[0040] The arithmetic unit 101 may be, for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), an FPGA (Field-Programmable Gate Array), an ASIC (Application-Specific Integrated Circuit), a neuroprocessor, or a microcontroller. The calculation unit 101 controls each component of the computer 50. The calculation unit 101 also enables the computer 50 to function by reading and executing programs that implement the preprocessing stage (step S1), classification stage (step S2), analysis stage (step S3), and extraction stage (step S4).

[0041] Storage 102 is a storage device that stores programs used by the arithmetic unit 101, control data such as arithmetic parameters, and image data. Storage 102 can be, for example, a hard disk drive (HDD), a solid-state drive (SSD), or a non-volatile memory such as flash memory.

[0042] Memory 103 is a storage device that temporarily stores, for example, a program executed by the arithmetic unit 101. Memory 103 may be a volatile memory such as DRAM or SRAM.

[0043] The display unit 104 is a device that displays images or characters on a screen, for example. The display unit 104 may be an LCD display, an organic EL display, or the like.

[0044] Although not shown in the diagram, the computer 50 may be equipped with a communication interface. This communication interface may utilize wireless communication technologies such as Wi-Fi, Bluetooth®, or LTE (Long Term Evolution), or it may utilize wired communication technologies. For example, the arithmetic unit 101 can analyze the river conditions based on image data obtained via this communication interface.

[0045] Computer 50 could be, for example, a server, a personal computer (PC), a mainframe computer, a supercomputer, a smartphone, a tablet, a wearable computer, or a head-mounted display.

[0046] The above description of the river image analysis method according to the first embodiment of the present invention can be applied to other embodiments of the present invention, unless there are any particular technical inconsistencies.

[0047] <2. A second embodiment of the present invention (Example 2 of river image analysis method)> The river image analysis method according to one embodiment of the present invention allows for further detailed analysis of the extracted image of the region of interest. This will be explained with reference to Figure 7. Figure 7 is a flowchart showing an example of the procedure for the river image analysis method according to one embodiment of the present invention.

[0048] As shown in Figure 7, the river image analysis method according to one embodiment of the present invention includes a preprocessing step (step S1), a classification step (step S2), an analysis step (step S3), an extraction step (step S4), and a detailed classification step (step S5). It differs from the river image analysis method according to the first embodiment in that it further includes a detailed classification step (step S5).

[0049] In the detailed classification stage (step S5), the computer's processing unit uses a trained model, which has been machine-learned based on multiple aerial images with higher resolution than satellite images, to classify each region included in the aerial images corresponding to the region of interest into a class. This region of interest is the region extracted from satellite images in the extraction stage (step S4). The processing unit classifies each region included in this region of interest into a class in the aerial images with higher resolution than satellite images. Because aerial images have higher resolution than satellite images, they can distinguish even fine surface features such as sand and gravel, which were difficult to distinguish in satellite images, and can classify them into classes with high accuracy.

[0050] Aerial images are images taken during aerial laser surveying or UAV laser surveying. Aerial laser surveying is a method of measuring the shape of the Earth's surface using a laser scanner mounted on an aircraft. The laser scanner measures the time it takes for a laser beam emitted from the Earth's surface to reflect back and calculates the distance to the Earth's surface. In this process, the laser scanner emits laser beams, for example, 50,000 to 100,000 times per second, and can measure the Earth's surface at intervals of 60 cm or less, thus obtaining high-density topographic data.

[0051] Data obtained by aerial laser surveying or UAV laser surveying can be used to create 3D point cloud data and topographic maps called Digital Elevation Models (DEMs). This data can be combined with aerial imagery to understand the features of the Earth's surface. Aerial imagery is high-resolution imagery taken from aircraft, and combining it with other data allows for the acquisition of rich topographic information. Aerial imagery can be orthorectified images. While satellite imagery can be in the near-infrared and visible light wavelength bands, aerial imagery can be color images captured in color.

[0052] To quantify changes in river topography, the ground resolution of aerial imagery is a crucial factor in the analysis. In this technology, the aerial imagery used for analysis is resampled to 10 centimeters for training and analysis, but it is not necessarily limited to 10 centimeters. Depending on the purpose of the analysis and the required accuracy, it is also possible to resample to an appropriate resolution for processing.

[0053] UAV (Unmanned Aerial Vehicle) laser surveying is a method of measuring the shape of the Earth's surface using a laser scanner mounted on an unmanned aerial vehicle. Similar to aerial laser surveying, it calculates the distance to the Earth's surface by measuring the reflection time of the laser beam. Unmanned aerial vehicles also have the capability to capture aerial images.

[0054] In this embodiment, the preprocessing stage (step S1), classification stage (step S2), analysis stage (step S3), and extraction stage (step S4) can also be considered preprocessing for the detailed classification stage (step S5). This preprocessing enables early detection of topographic changes and more accurate analysis, even when satellite image acquisition frequency is low.

[0055] Although not shown in the diagram, the river image analysis method may include a preprocessing step between the extraction step (step S4) and the detailed classification step (step S5). This preprocessing step may be the same as the preprocessing step (step S1) performed before the classification step (step S2), or it may be a different process.

[0056] The above description of the river image analysis method according to the second embodiment of the present invention can be applied to other embodiments of the present invention, unless there are any particular technical inconsistencies.

[0057] <3. A third embodiment of the present invention (Example 3 of river image analysis method)> The river image analysis method according to one embodiment of the present invention can evaluate the river crosslength based on the classification results at the detailed classification stage and topographic data. This will be explained with reference to Figure 8. Figure 8 is a flowchart showing an example of the procedure for the river image analysis method according to one embodiment of the present invention.

[0058] As shown in Figure 8, the river image analysis method according to one embodiment of the present invention includes a preprocessing stage (step S1), a classification stage (step S2), an analysis stage (step S3), an extraction stage (step S4), a detailed classification stage (step S5), and a river cross-sectional area evaluation stage (step S6). It differs from the river image analysis method according to the second embodiment in that it further includes a river cross-sectional area evaluation stage (step S6).

[0059] In the river cross-section evaluation stage (step S6), the computer's calculation unit evaluates the river cross-section based on the classification results from the detailed classification stage (step S5) and topographic data. River cross-section refers to the area of ​​water occupied in the cross-section of a river. Generally, it refers to the cross-sectional area below the planned high water level. River cross-section is one of the indicators that represent the flow capacity of a river. The larger the river cross-section, the more water can be carried. River cross-section can be calculated, for example, by multiplying the average water depth by the river width.

[0060] The topographic data consists of at least one of the following: 3D point cloud data, a digital elevation model (DEM), and a digital surface model (DSM). As mentioned above, this 3D point cloud data is acquired using aerial laser surveying or UAV laser surveying. Height information can be obtained based on the 3D point cloud data. Therefore, by combining the classification results from a trained model with the 3D point cloud data, it is possible to quantitatively evaluate, for example, the changes in river crosslength for each classification. This enables evaluation with very high spatial resolution, allowing for the understanding of subtle changes in rivers and topographic fluctuations.

[0061] The river image analysis method according to one embodiment of the present invention can also evaluate river crosslength using a DEM created from this 3D point cloud data and other elevation data. Laser surveying can acquire accurate height information of the ground surface using laser light and generate a DEM that removes the influence of buildings, vegetation, etc. This makes it possible to understand the changes in the ground surface itself in detail.

[0062] Alternatively, the river image analysis method according to one embodiment of the present invention can also evaluate river crosslength using a digital surface model (DSM) obtained by photogrammetry (SfM: Structure from Motion). SfM is a technique that reconstructs the three-dimensional shape of an object based on multiple images taken from different angles and positions. DSM is three-dimensional surface data that includes not only topographic surface data but also all objects present on the ground surface, such as buildings, trees, and structures. This data is particularly useful for analyses that include the influence of the presence of ground objects and is suitable for grasping the overall picture of the site.

[0063] The above description of the river image analysis method according to the third embodiment of the present invention can be applied to other embodiments of the present invention, unless there are any particular technical inconsistencies.

[0064] <4. A fourth embodiment of the present invention (Example 4 of river image analysis method)> In the explanation of Figure 3, it was stated that as erosion progresses, the river body expands, potentially increasing the risk of levee breaches during floods. This technology can assess the level of this risk of levee breach. This will be explained with reference to Figure 9. Figure 9 is a flowchart showing an example of the procedure for a river image analysis method according to one embodiment of the present invention.

[0065] As shown in Figure 9, a river image analysis method according to one embodiment of the present invention includes a preprocessing step (step S1), a classification step (step S2), an analysis step (step S3), an extraction step (step S4), and a levee breach risk assessment step (step S7). This differs from the river image analysis method according to the first embodiment in that it further includes a levee breach risk assessment step (step S7). The stage in which the levee breach risk assessment step (step S7) is performed is not particularly limited. For example, the risk assessment step (step S7) may be performed after the detailed classification step (see Figure 7) or the river evaluation step (see Figure 8).

[0066] In the levee breach risk assessment stage (step S7), the computer's calculation unit evaluates the level of levee breach risk based on the levee protection line and the time-series change pattern during the analysis stage.

[0067] Specifically, first, data for pre-defined levee protection lines is input into the calculation unit, and these lines are saved as the basis for analysis. These protection lines indicate the boundary that serves as the standard when evaluating the risk of levee breach in a specific area of ​​the river.

[0068] Next, the time-series change patterns analyzed in the analysis phase (step S3) are evaluated by overlaying them with the levee protection line. At this stage, it is confirmed whether the progress of riverbank erosion is approaching or reaching the levee protection line, and high-risk areas are identified.

[0069] Finally, the degree of topographical change relative to the levee protection line is evaluated, and the level of levee breach risk is quantitatively calculated. If the risk exceeds a certain threshold, an alert is issued, and relevant parties are notified, enabling prompt countermeasures.

[0070] The above description of the river image analysis method according to the fourth embodiment of the present invention can be applied to other embodiments of the present invention unless there is a particular technical contradiction.

[0071] <5. A fifth embodiment of the present invention (an example of a river image analysis device)> The present invention provides a river image analysis device that uses a computer to analyze the condition of a river based on images of the river, comprising: a classification unit that classifies each region included in a satellite image into a class using a trained model that has been trained by machine learning to determine the correlation between multiple satellite images of a river and the class to which each region included in the satellite image belongs; an analysis unit that analyzes the time-series change pattern of the region based on the classification result by the classification unit; and an extraction unit that extracts a region of interest in the satellite image based on the time-series change pattern by the analysis unit.

[0072] A river image analysis device according to one embodiment of the present invention will be described with reference to Figure 10. Figure 10 is a block diagram showing an example of the configuration of the river image analysis device 1 according to one embodiment of the present invention.

[0073] As shown in Figure 10, the river image analysis device 1 comprises a classification unit 11, an analysis unit 12, and an extraction unit 13. The classification unit 11, the analysis unit 12, and the extraction unit 13 may be included in the calculation unit 101 described above. The river image analysis device 1 may be the computer 50 described above.

[0074] The classification unit 11 uses a pre-trained model, which has been trained using machine learning to analyze the correlation between multiple satellite images of rivers and the class to which each region included in the satellite images belongs, to classify each region included in the satellite images into a class.

[0075] The analysis unit 12 analyzes the time-series change pattern of the region based on the classification results from the classification unit 11.

[0076] The extraction unit 13 extracts regions of interest in the satellite image based on the time-series change patterns obtained by the analysis unit 12.

[0077] The above description of the river image analysis device according to the fifth embodiment of the present invention can be applied to other embodiments of the present invention, unless there is a particular technical contradiction.

[0078] <6. A sixth embodiment of the present invention (an example of a river image analysis program)> The present invention provides a river image analysis program that uses a computer to analyze the condition of a river based on images of the river, and causes the computer to perform the following steps: a classification step in which multiple satellite images of a river are classified into classes using a trained model that has been trained by machine learning to determine the correlation between multiple satellite images of a river and the class to which each region included in the satellite images belongs; an analysis step in which the time-series change pattern of the region is analyzed based on the classification results in the classification step; and an extraction step in which regions of interest are extracted from the satellite images based on the time-series change pattern in the analysis step.

[0079] For example, the aforementioned arithmetic unit 101 can read and execute this program, thereby enabling the computer to function.

[0080] This program can be stored on a storage medium. This storage medium can be a volatile storage medium or a non-volatile storage medium. Examples of volatile storage media include CPUs and GPUs. Examples of non-volatile storage media include hard disk drives (HDDs), solid-state drives (SSDs), optical discs, and flash memory.

[0081] Alternatively, this program can be stored in a storage medium such as ROM (Read-Only Memory), PROM (Programmable Read-Only Memory), or EEPROM (Electrically Erasable Programmable Read-Only Memory).

[0082] The above description of the river image analysis program according to the sixth embodiment of the present invention can be applied to other embodiments of the present invention, unless there is a particular technical inconsistency.

[0083] Furthermore, the present invention can also take the following configuration. [1] A river image analysis method that uses a computer to analyze the condition of a river based on images taken of the river, A classification step is performed to classify each region in the satellite image into a class, using a pre-trained model that has been trained by machine learning to analyze the correlation between multiple satellite images of rivers and the class to which each region in the satellite image belongs. Based on the classification results in the aforementioned classification stage, an analysis stage is performed to analyze the time-series change pattern of the aforementioned region. A river image analysis method comprising: an extraction step of extracting a region of interest in a satellite image based on the time-series change pattern in the aforementioned analysis step. [2] The aforementioned satellite images are images in the near-infrared wavelength band and the visible light wavelength band. [1] The river image analysis method described. [3] The aforementioned time-series change pattern is a change from trees to gravelly land. The river image analysis method described in [1] or [2]. [4] This further includes a detailed classification step in which a trained model, which has been machine-trained on multiple aerial images with a higher resolution than the aforementioned satellite image, classifies each region included in the aerial image corresponding to the region of interest into a class. A river image analysis method described in any one of [1] to [3]. [5] The aerial image is an image taken during aerial laser surveying or UAV laser surveying. The river image analysis method described in [4]. [6] The aforementioned aerial image is a color image captured in color. The river image analysis method described in [4] or [5]. [7] This further includes a river cross-section evaluation stage, in which the river cross-section is evaluated based on the classification results from the detailed classification stage and the topographic data. A river image analysis method described in any one of [4] to [6]. [8] The aforementioned topographic data is at least one of 3D point cloud data, a digital elevation model (DEM), and a digital surface model (DSM). The river image analysis method described in [7]. [9] The process further includes a levee breach risk assessment step, which evaluates the level of levee breach risk based on the levee protection line and the time-series change pattern in the analysis step, [1] The river image analysis method described.

[10] A river image analysis device that uses a computer to analyze the condition of a river based on images taken of the river, A classification unit that classifies each region in a satellite image into a class using a pre-trained model that has been machine-learned to determine the correlation between multiple satellite images of rivers and the class to which each region in the satellite image belongs. An analysis unit analyzes the time-series change pattern of the region based on the classification results from the classification unit, A river image analysis device comprising: an extraction unit that extracts regions of interest in satellite images based on time-series change patterns obtained by the aforementioned analysis unit.

[11] A river image analysis program that uses a computer to analyze the condition of a river based on images taken of the river, A classification step is performed to classify each region in the satellite image into a class, using a pre-trained model that has been trained by machine learning to analyze the correlation between multiple satellite images of rivers and the class to which each region in the satellite image belongs. Based on the classification results in the aforementioned classification stage, an analysis stage is performed to analyze the time-series change pattern of the aforementioned region. A river image analysis program that causes a computer to perform an extraction step to extract regions of interest in satellite images based on the time-series change patterns in the aforementioned analysis step. [Explanation of Symbols]

[0084] S1 Pretreatment stage S2 classification stage S3 Analysis Phase S4 Extraction Stage S5 Detailed Classification Stage S6 Kawatsu Evaluation Stage S7 Levee Breach Risk Assessment Stage 1. River image analysis device 11 Classification section 12 Analysis Department 13 Extraction part 50 Computers 101 Arithmetic section 102 storage 103 memory 104 Display section

Claims

1. A river image analysis method that uses a computer to analyze the condition of a river based on images taken of the river, A classification stage is performed by classifying each region in the satellite image into a class, using a pre-trained model that has been trained by machine learning to analyze the correlation between multiple satellite images of rivers and the class to which each region contained in those satellite images belongs. Based on the classification results in the aforementioned classification stage, an analysis stage is performed to analyze the time-series change pattern of the aforementioned region. A river image analysis method comprising: an extraction step of extracting a region of interest in a satellite image based on the time-series change pattern in the aforementioned analysis step.

2. The aforementioned satellite images are images in the near-infrared wavelength band and the visible light wavelength band. The river image analysis method according to claim 1.

3. The aforementioned time-series change pattern is a change from trees to gravelly land. The river image analysis method according to claim 1.

4. This further includes a detailed classification step in which a trained model, which has been machine-trained on multiple aerial images with a higher resolution than the aforementioned satellite image, classifies each region included in the aerial image corresponding to the region of interest into a class. The river image analysis method according to claim 1.

5. The aerial image is an image taken during aerial laser surveying or UAV laser surveying. The river image analysis method according to claim 4.

6. The aforementioned aerial image is a color image captured in color. The river image analysis method according to claim 4.

7. This further includes a river cross-section evaluation stage, in which the river cross-section is evaluated based on the classification results from the detailed classification stage and the topographic data. The river image analysis method according to claim 4.

8. The aforementioned topographic data is at least one of 3D point cloud data, a digital elevation model (DEM), and a digital surface model (DSM). The river image analysis method according to claim 7.

9. The process further includes a levee breach risk assessment step, which evaluates the level of levee breach risk based on the levee protection line and the time-series change pattern in the analysis step, The river image analysis method according to claim 1.

10. A river image analysis device that uses a computer to analyze the condition of a river based on images taken of the river, A classification unit that classifies each region in a satellite image into a class using a pre-trained model that has been machine-learned to determine the correlation between multiple satellite images of rivers and the class to which each region in the satellite image belongs. An analysis unit analyzes the time-series change pattern of the region based on the classification results from the classification unit, A river image analysis device comprising: an extraction unit that extracts regions of interest in satellite images based on time-series change patterns obtained by the aforementioned analysis unit.

11. A river image analysis program that uses a computer to analyze the condition of a river based on images taken of the river, A classification stage is performed by classifying each region in the satellite image into a class, using a pre-trained model that has been trained by machine learning to analyze the correlation between multiple satellite images of rivers and the class to which each region contained in those satellite images belongs. Based on the classification results in the aforementioned classification stage, an analysis stage is performed to analyze the time-series change pattern of the aforementioned region. A river image analysis program that causes a computer to perform an extraction step to extract regions of interest in satellite images based on the time-series change patterns in the aforementioned analysis step.

Images