Urban feature quantity analysis system, urban feature quantity analysis method, urban feature quantity analysis program, training device, and training method

Abstract

By using a trained model (301) and using a remote sensing image (321) as input, an inference unit (220) acquires a land use map (302) in which pixels have each been assigned a label indicating the classification of land use for a site corresponding to the pixel. A connectivity analysis unit (230) uses the acquired land use map (302) to calculate an urban feature quantity (303) which is a feature quantity pertaining to a regional welfare level.

Description

Urban feature analysis system, urban feature analysis method, urban feature analysis program, learning device, and learning method

[0001] This disclosure concerns a technology for obtaining town features, which are characteristics related to regional well-being.

[0002] Non-patent document 1 discloses a study that extracted the NDVI (National Land Use Index) from satellite imagery, compared it with happiness scores based on the World Happiness Report, and evaluated the relationship between the two. This study is a benchmark study that investigates the relationship between satellite imagery and happiness scores. In this study, satellite imagery is used from the perspective of land use (more specifically, the greening index).

[0003] In Japan, detailed data collection is being conducted on the Regional Happiness Index based on more than 50 questions. For example, there are questions such as, "In the area where I live, I can feel close to nature." Based on this kind of data collection, it is necessary to clarify how cities and nature are connected and interact with each other, rather than simply looking at land use proportions.

[0004] To clarify the relationship between cities and nature, there are challenges from two perspectives. The first challenge is to obtain more detailed land-use maps from satellite imagery with higher ground resolution. The second challenge is to use the land-use maps to calculate features such as the spatial relationship, connectivity, and entropy between cities and nature.

[0005] Furthermore, Non-Patent Document 1 does not disclose how to simultaneously achieve the following 1) and 2): 1) Generating a high-resolution land-use map with city labels and nature labels using satellite imagery. 2) Analyzing the connectivity between city labels and nature labels.

[0006] Urban green space and happiness in developed countries, KAIST

[0007] This disclosure aims to enable the acquisition of features related to regional well-being using remote sensing images.

[0008] The town feature analysis system disclosed herein comprises: an inference unit that takes remote sensing images as input and uses a trained model to obtain a land use map in which each pixel is labeled to indicate the classification of land use at the location corresponding to that pixel; and a connectivity analysis unit that uses the obtained land use map to calculate town features, which are features related to regional well-being.

[0009] According to this disclosure, features related to regional well-being can be obtained using remote sensing images.

[0010] Configuration diagram of the urban feature analysis system 300 in Embodiment 1. Configuration diagram of the learning device 100 in Embodiment 1. Configuration diagram of the inference device 200 in Embodiment 1. Flowchart of the learning method in Embodiment 1. Diagram showing an example of a land use map in Embodiment 1. Flowchart of the inference method in Embodiment 1. Diagram showing an example of an urban area in Embodiment 1. Diagram showing an example of the configuration of a trained model 301 in Embodiment 1. Hardware configuration diagram of the learning device 100 in Embodiment 1. Hardware configuration diagram of the inference device 200 in Embodiment 1.

[0011] In the embodiments and drawings, the same or corresponding elements are denoted by the same reference numeral. The descriptions of elements denoted by the same reference numeral as the described elements are omitted or simplified as appropriate. The arrows in the figures mainly indicate the flow of data or processing.

[0012] Embodiment 1. The town feature analysis system 300 will be described based on Figures 1 to 8.

[0013] ***Configuration Description*** The configuration of the town feature analysis system 300 will be described based on Figure 1. The town feature analysis system 300 comprises a learning device 100 and an inference device 200. The learning device 100 learns from multiple sets of remote sensing images 311 and land use maps 312 to create a trained model 301. The inference device 200 takes remote sensing images 321 as input, uses the trained model 301 to obtain a land use map 302, and uses the land use map 302 to calculate town features 303. Further details will be described later.

[0014] The configuration of the learning device 100 will be explained based on Figure 2. The learning device 100 is a computer equipped with hardware such as a processor 101, memory 102, auxiliary storage device 103, communication device 104, and input / output interface 105. These hardware components are connected to each other via signal lines.

[0015] Processor 101 is the processor of the learning device 100. A processor is an IC that performs arithmetic processing and controls other hardware. For example, processor 101 is a CPU, DSP, GPU, or a combination of these. IC is an abbreviation for Integrated Circuit. CPU is an abbreviation for Central Processing Unit. DSP is an abbreviation for Digital Signal Processor. GPU is an abbreviation for Graphics Processing Unit.

[0016] Memory 102 is the memory of the learning device 100. Data stored in memory 102 is saved to the auxiliary storage device 103 as needed. Memory is a volatile or non-volatile storage device. Memory is also called main memory. For example, memory is RAM. RAM is an abbreviation for Random Access Memory.

[0017] The auxiliary storage device 103 is an auxiliary storage device of the learning device 100. Data stored in the auxiliary storage device 103 is loaded into the memory 102 as needed. The auxiliary storage device is a non-volatile storage device. For example, the auxiliary storage device is a ROM, HDD, flash memory, or a combination thereof. ROM is an abbreviation for Read Only Memory. HDD is an abbreviation for Hard Disk Drive.

[0018] Communication device 104 is a communication device for learning device 100. Communication of learning device 100 is performed using communication device 104. The communication device consists of a receiver and a transmitter. For example, the communication device is a communication chip or NIC. NIC is an abbreviation for Network Interface Card.

[0019] The input / output interface 105 is the input / output interface of the learning device 100. Input and output of the learning device 100 are performed via the input / output interface 105. The input / output interface is a port to which input devices and output devices are connected. For example, the input / output interface is a USB terminal, the input devices are a keyboard and mouse, and the output device is a display. USB is an abbreviation for Universal Serial Bus.

[0020] The learning device 100 includes elements such as a data acquisition unit 110 and a model generation unit 120. These elements are implemented using software.

[0021] The auxiliary storage device 103 stores a learning program that allows the computer to function as a data acquisition unit 110 and a model generation unit 120. The learning program is loaded into memory 102 and executed by the processor 101. The auxiliary storage device 103 also stores the operating system (OS). At least a portion of the OS is loaded into memory 102 and executed by the processor 101. The processor 101 executes the learning program while executing the OS. OS is an abbreviation for Operating System.

[0022] The learning program data (input data, output data, etc.) is stored in the storage unit 190. Memory 102 functions as the storage unit 190. However, storage devices such as the auxiliary storage device 103, registers in the processor 101, and cache memory in the processor 101 may function as the storage unit 190 instead of, or together with, memory 102.

[0023] The learning program can be recorded (stored) in a computer-readable format on a non-volatile recording medium such as an optical disc or flash memory.

[0024] The configuration of the inference device 200 will be explained based on Figure 3. The inference device 200 is a computer equipped with hardware such as a processor 201, memory 202, auxiliary storage device 203, communication device 204, and input / output interface 205. These hardware components are connected to each other via signal lines.

[0025] Processor 201 is the processor of the inference device 200. Memory 202 is the memory of the inference device 200. Data stored in memory 202 is saved to auxiliary storage device 203 as needed. Auxiliary storage device 203 is the auxiliary storage device of the inference device 200. Data stored in auxiliary storage device 203 is loaded into memory 202 as needed. Communication device 204 is the communication device of the inference device 200. Communication of the inference device 200 is performed using communication device 204. Input / output interface 205 is the input / output interface of the inference device 200. Input and output of the inference device 200 is performed using input / output interface 205.

[0026] The inference device 200 comprises elements such as a data acquisition unit 210, an inference unit 220, and a connectivity analysis unit 230. These elements are implemented in software.

[0027] The auxiliary storage device 203 stores inference programs that enable the computer to function as a data acquisition unit 210, an inference unit 220, and a connectivity analysis unit 230. The inference programs are loaded into memory 202 and executed by the processor 201. The auxiliary storage device 203 also stores the operating system (OS). At least a portion of the OS is loaded into memory 202 and executed by the processor 201. The processor 201 executes the inference programs while executing the OS.

[0028] The data for the inference program is stored in the storage unit 290. Memory 202 functions as the storage unit 290. However, storage devices such as auxiliary storage device 203, registers in the processor 201, and cache memory in the processor 201 may function as the storage unit 290 instead of, or together with, memory 202.

[0029] The inference program can be recorded (stored) in a computer-readable format on a non-volatile recording medium such as an optical disc or flash memory.

[0030] ***Explanation of Operation*** The operation procedure of the town feature analysis system 300 corresponds to the town feature analysis method. Furthermore, the operation procedure of the town feature analysis system 300 corresponds to the processing procedure of the town feature analysis program. The operation procedure of the learning device 100 corresponds to the learning method. Furthermore, the operation procedure of the learning device 100 corresponds to the processing procedure of the learning program. The operation procedure of the inference device 200 corresponds to the inference method. Furthermore, the operation procedure of the inference device 200 corresponds to the processing procedure of the inference program.

[0031] The learning method will be explained based on Figure 4. In step S110, the data acquisition unit 110 acquires learning data.

[0032] The training data is the data used to train the pre-trained model 301.

[0033] For example, a user inputs learning data into the learning device 100, and the data acquisition unit 110 receives the input learning data.

[0034] The training data includes a plurality of sets of remote sensing images 311, land use maps 312, and GIS data 319. However, the GIS data 319 may be absent.

[0035] The remote sensing image 311 is a remote sensing image for training. The remote sensing image is an image obtained by remote sensing. For example, the remote sensing image is a high-resolution optical image (optical satellite image) obtained by photographing from an artificial satellite equipped with an optical camera.

[0036] The land use map 312 is a land use map for training corresponding to the remote sensing image 311. In other words, the land use map 312 is a land use map of the area shown in the remote sensing image 311.

[0037] The land use map is an image in which each pixel is labeled with a label indicating the classification of the land use of the location shown in the pixel. Each pixel of the land use map is mainly labeled with a natural label or an urban label. The natural label is a label indicating a classification related to nature. For example, the classification related to nature includes bare land, grassland, trees, water, farmland, forest, etc. The urban label is a label indicating a classification related to the city. For example, the classification related to the city includes pavement, roads, buildings, etc.

[0038] Fig. 5 shows an example of a land use map. In Fig. 5, the land use map is labeled with classifications such as "urban area", "road", "grassland" or "forest" for each pixel. And each pixel of the land use map is color-coded according to the classification indicated by the attached label.

[0039] Returning to Fig. 4, the description of step S110 is continued. The land use map 312 is created manually using a computer, for example. A publicly available land use map may be used as the land use map 312.

[0040] The GIS data 319 is learning GIS data corresponding to the remote sensing image 311. In other words, the GIS data 319 is the GIS data of the area shown in the remote sensing image 311. GIS is an abbreviation for Geographic Information System.

[0041] In step S120, the model generation unit 120 learns the learning data to create a learned model 301.

[0042] The learned model 301 is a model that takes a remote sensing image as input and generates a land use map corresponding to the remote sensing image.

[0043] The learned model 301 is used as a semantic segmentation model. The semantic segmentation model obtains a class label corresponding to each pixel. Examples of semantic segmentation models are U-Net, DeepLabV3, etc.

[0044] In step S130, the model generation unit 120 saves the learned model 301.

[0045] For example, the model generation unit 120 saves the learned model 301 in the inference device 200 by communicating with the inference device 200.

[0046] However, the learned model 301 may be saved in the inference device 200 by the user. Also, the learned model 301 may be managed outside the inference device 200 and accessed from the inference device 200.

[0047] Based on FIG. 6, the inference method will be described. In step S210, the data acquisition unit 210 acquires input data.

[0048] The input data is data that serves as the input of the learned model 301.

[0049] For example, the user inputs the input data into the inference device 200, and the data acquisition unit 210 receives the input input data.

[0050] The input data includes remote sensing images 321 and GIS data 329. However, the GIS data 329 is optional.

[0051] The remote sensing image 321 is a remote sensing image for inference.

[0052] GIS data 329 is GIS data for inference that corresponds to the remote sensing image 321.

[0053] In step S220, the inference unit 220 uses the input data as input to obtain a land use map 302 using the trained model 301.

[0054] Land use map 302 is a land use map corresponding to remote sensing image 321.

[0055] The inference unit 220 may output a land use map 302. For example, the inference unit 220 may display the land use map 302 on a display.

[0056] In step S230, the connectivity analysis unit 230 calculates town features 303 using the land use map 302.

[0057] Town feature 303 is a feature related to regional well-being. Specifically, town feature 303 is a feature related to predefined nature connectivity.

[0058] Natural connectivity describes the proximity between urban and natural spaces. It can also be described as the connection to nature within one's mind. Natural connectivity reflects not only the objective quantity of nature, but also how close residents feel to nature.

[0059] Details of step S230 will be described later.

[0060] In step S240, the connectivity analysis unit 230 outputs the town feature quantity 303.

[0061] For example, the connectivity analysis unit 230 displays the town features 303 on the display.

[0062] ***Detailed Explanation*** The details of the calculation of town features 303 (step S230) are explained below. Natural connectivity is defined in advance as "proximity of natural areas," "visual proximity of natural areas," "proportion of natural areas," "diversity of natural areas," "green coverage rate," "accessibility to nature," or "visual complexity of nature." Connectivity analysis unit 230 calculates features for the predefined natural connectivity as town features 303.

[0063] This section explains the definition of "proximity to natural areas." The feature for "proximity to natural areas" is the statistical distance (average, minimum, maximum, etc.) from each urban area pixel to the nearest natural area pixel. Pixels with urban labels are called urban pixels. Areas mainly composed of urban pixels are called urban areas (see Figure 7). Urban area pixels are each individual pixel within an urban area. Pixels with natural labels are called natural pixels. Areas mainly composed of natural pixels are called natural areas. Natural area pixels are each individual pixel within a natural area.

[0064] Figure 7 shows an example of an urban area. The urban area enclosed by the dashed line is mainly composed of pixels with urban labels (urban area, roads).

[0065] The feature quantity for "proximity to natural areas" is calculated as follows. First, the connectivity analysis unit 230 sets urban areas and natural areas in the land use map 302. Urban areas are areas mainly composed of urban pixels with urban labels. Each pixel within an urban area is called an urban area pixel. Natural areas are areas mainly composed of natural pixels with natural labels. Each pixel within a natural area is called a natural area pixel. Next, for each urban area pixel, the connectivity analysis unit 230 selects a natural area pixel that is close to the urban area pixel and calculates the distance to the selected natural area pixel. Then, the connectivity analysis unit 230 calculates statistics (for example, average, minimum, or maximum) of the calculated distance. The calculated statistics are the feature quantity for "proximity to natural areas".

[0066] For example, the feature variable for "proximity of natural areas" can be expressed by the following formula.

[0067]

[0068] avg_distance is a feature that describes the "proximity of natural areas". "N" is the number of pixels in urban areas. ｉ " represents the coordinates of the i-th city area pixel. ｊ " represents the coordinates of the j-th natural area pixel.

[0069] This section explains the definition of "visual proximity to natural areas." "Visual proximity to natural areas" is an evaluation of how close residents feel to nature. "Closeness to nature" is evaluated based on how much of the natural area is included within their field of vision. For example, "closeness to nature" can be calculated as the percentage of natural areas within a certain radius centered on the urban area. This radius could be, for example, 500 meters or 1 kilometer.

[0070] The feature quantity for "visual proximity to natural areas" is calculated as follows. First, the connectivity analysis unit 230 sets urban areas and natural areas within the land use map 302. Next, the connectivity analysis unit 230 sets a certain range of area that includes the urban area. The set area is called the reference area. For example, the connectivity analysis unit 230 sets an area with a certain radius (for example, 500 meters or 1 kilometer) centered on the center of the urban area as the reference area. Then, the connectivity analysis unit 230 calculates the proportion of natural areas within the reference area. Specifically, the connectivity analysis unit 230 counts the number of pixels of natural areas within the reference area and the number of pixels of the reference area, and divides the number of pixels of natural areas included in the reference area by the number of pixels of the reference area. The calculated proportion is the feature quantity for "visual proximity to natural areas".

[0071] This section explains the definition of "Proportion of Nature." "Proportion of Nature" is the percentage of pixels in an image that represent natural areas (forests, farmland, bodies of water, etc.).

[0072] The feature quantity for "proportion of natural areas" is calculated as follows: The connectivity analysis unit 230 calculates the proportion of natural pixels in the land use map 302. Specifically, the connectivity analysis unit 230 counts the number of natural pixels in the land use map 302 and divides the number of natural pixels in the land use map 302 by the total number of pixels in the land use map 302. The calculated proportion is the feature quantity for "proportion of natural areas".

[0073] If there are three types of natural areas, the feature quantity for "the proportion of natural areas" can be expressed by the following formula, for example: P_nature = (N1 + N2 + N3) / PA P_nature is the feature quantity for "the proportion of natural areas". "N1" is the number of pixels for the first type of natural area (e.g., forest). "N2" is the number of pixels for the second type of natural area (e.g., farmland). "N3" is the number of pixels for the third type of natural area (e.g., water body). "PA" is the total number of pixels in the land use map 302.

[0074] If the types of natural areas are bare land, green space, forest, and water, the characteristic quantity for "proportion of natural areas" can be expressed by, for example, the following formula.

[0075]

[0076] `proposition_nature` is a feature variable that describes the "proportion of natural areas". ｂａｒｅｌａｎｄ This is the number of pixels in the bare area. ｇｒａｓｓ This is the number of pixels in the green area. ｗａｔｅｒ This is the number of pixels in the water area. ｉｍａｇｅ This represents the total number of pixels in land use map 302.

[0077] This section explains the definition of "diversity in natural areas." "Diversity in natural areas" is an assessment of the diversity of species within a natural area. It expresses the richness of nature as perceived by residents.

[0078] The feature quantity for "diversity of natural areas" is calculated as follows. First, the connectivity analysis unit 230 sets a certain range of area (reference area) that includes each city pixel in the land use map 302. Then, the connectivity analysis unit 230 calculates the diversity index of natural areas within the reference area. Specifically, the connectivity analysis unit 230 counts the number of pixels of natural areas within the reference area for each type of natural label, and calculates the diversity index using the counted number of pixels for each type of natural area. Examples of diversity indices are the Simpson index or the Shannon index. Examples of types of natural labels are forests, farmland, water bodies, etc.

[0079] This section explains the definition of "Green Coverage Index." "Green Coverage Index" is the percentage of green space within a field of view for a given urban area pixel.

[0080] The feature quantity for "green coverage ratio" is calculated as follows. First, the connectivity analysis unit 230 sets the urban area in the land use map 302. Next, the connectivity analysis unit 230 sets a certain range of area (reference area) that includes the urban area. Then, the connectivity analysis unit 230 calculates the proportion of green space pixels in the reference area. Green space pixels are pixels with green space labels. Examples of green space labels are forests and parks. Specifically, the connectivity analysis unit 230 counts the number of green space pixels in the reference area and the number of pixels in the reference area, and divides the number of green space pixels in the reference area by the number of pixels in the reference area. The calculated proportion is the feature quantity for "green coverage ratio".

[0081] This section explains the definition of "accessibility to nature." Accessibility to nature is an assessment of how easily urban residents can access natural areas.

[0082] The feature vector for "accessibility to nature" is calculated as follows: First, the connectivity analysis unit 230 defines the natural areas within the land use map 302. Next, for each city pixel in the land use map 302, the connectivity analysis unit 230 calculates the (shortest) access route from the city pixel to the natural area. Then, for each city pixel in the land use map 302, the connectivity analysis unit 230 calculates the amount of travel along the access route (e.g., travel time or travel distance). Finally, the connectivity analysis unit 230 calculates statistics (e.g., average) of the calculated amount of travel. The calculated statistics are the feature vector for "accessibility to nature".

[0083] This section explains the definition of "Visual Complexity of Nature." Visual complexity of nature is an assessment of the visual complexity of a natural area. Visual complexity reflects the richness and diversity of nature as perceived by residents.

[0084] The feature quantity regarding "the visual complexity of nature" is calculated as follows. First, the connectivity analysis unit 230 sets a natural area in the land use map 302. Then, the connectivity analysis unit 230 calculates the visual complexity of the natural area pixels. Examples of visual complexity are fractal dimension or edge density.

[0085] For example, the feature quantity regarding "the visual complexity of nature" is represented by the following formula.

[0086]

[0087] avg_visual_complexity represents the feature quantity regarding "the visual complexity of nature". "N" is the number of pixels in the urban area. "R" is a certain radius. "u ｉ " is the coordinate of the i-th urban area pixel. "r" is the radius. S(u ｉ , r) is the area with radius r centered on pixel ui. S ｎａｔｕｒｅ (u ｉ (u

[0088] The connectivity analysis unit 230 may calculate the feature quantity for each of two or more types of predefined natural connectivities, and integrate the calculated feature quantities to calculate the town feature quantity 303. For example, the connectivity analysis unit 230 calculates the statistics of the calculated feature quantities. The calculated statistics become the town feature quantity 303.

[0089] For example, the connectivity analysis unit 230 calculates feature quantities such as proportion_nature, avg_distance, and avg_visual_complexity, and integrates the calculated feature quantities to calculate the town feature quantity 303. For example, the town feature quantity 303 is represented by the following formula.

[0090]

[0091] NatureConnectedness is the town feature quantity 303.

[0092] ***Features of Embodiment 1*** Embodiment 1 discloses urban feature analysis.

[0093] The urban feature analysis system 300 is a system that generates urban features for evaluating a regional happiness index aggregated in a certain area, using remote sensing images 311 observed in a wide area of ​​interest. The urban feature analysis system 300 includes a learning device 100. The learning device 100 includes a data acquisition unit 110 and a model generation unit 120. The data acquisition unit 110 acquires training data including remote sensing images 311 observed in the area of ​​interest and a land use map 312 (ground truth) that maps the land use situation in the area of ​​interest at high resolution (sub-meter resolution). The model generation unit 120 uses the training data to generate a trained model 301 for inferring land use from remote sensing images. The high-resolution land use map 312 includes at least urban labels with a resolution capable of discriminating houses and buildings, and natural labels. The urban feature analysis system 300 also includes an inference device 200. The inference device 200 comprises a data acquisition unit 210, an inference unit 220, and a connectivity analysis unit 230. The data acquisition unit 210 acquires remote sensing images 321 of the region of interest. The inference unit 220 maps a high-resolution land use map 302 of the region of interest using a trained model 301. The connectivity analysis unit 230 extracts city labels from the land use map 302, analyzes the connectivity between city labels and nature labels, and calculates the results as urban features 303.

[0094] ***Effects of Embodiment 1*** The urban feature analysis system 300 evaluates the connectivity between urban labels and natural labels from the land use map 302 obtained from the analysis results of the remote sensing image 321, and creates urban feature quantities 303 for the area to be analyzed. Embodiment 1 makes it possible to comprehensively obtain feature quantities related to regional well-being from satellite images, and promotes the use of regional well-being in urban planning.

[0095] ***Supplement to Embodiment 1*** Embodiment 1 has the following features: (1) A land use map is generated that has "high resolution" urban labels and land cover classification labels including nature labels using remote sensing images. Specifically, mapping is performed by AI learning and inference from remote sensing images using a trained model. Remote sensing images include satellite images. Examples of urban labels include pavement, roads, buildings, etc. Examples of nature labels include trees, water, farmland, etc. Examples of land cover classification labels include bare land, grassland, pavement, roads, trees, water, farmland, buildings. A land use map is also called a land use diagram. AI is an abbreviation for artificial intelligence. (2) Urban features are calculated by connectivity analysis of urban labels and nature labels using the land use map. Urban features include features for objectively evaluating how close residents feel to nature.

[0096] The survey results revealed that regional well-being is associated with subjective opinions about the living environment. Examples of subjective opinions about the living environment include: "There are places where you can walk comfortably, such as parks and along rivers," "There are natural landscapes to be proud of," "You can feel close to nature," and "The air and water in the area are clear and clean." Based on these findings, in Embodiment 1, a land use map generated by inference is used to calculate town features as a more objective evaluation index indicating well-being based on the natural environment. Town features are calculated by connectivity analysis between urban labels and nature labels.

[0097] Specifically, the inference unit calculates town features using a trained model generated by the learning unit. The learning unit comprises a data acquisition unit and a model generation unit. The data acquisition unit acquires training data including remote sensing images obtained by remote sensing observing the region of interest, and a land use map (ground truth) that maps the land use status of the region of interest at high resolution (sub-meter resolution). The model generation unit uses the training data to generate a trained model for inferring land use status from remote sensing images. The inference unit comprises a data acquisition unit, an inference unit, and a connectivity analysis unit. The data acquisition unit acquires remote sensing images observing the region of interest. The inference unit uses the trained model to map the high-resolution land use map of the region of interest. The connectivity analysis unit extracts city labels from the land use map, analyzes the connectivity between city labels and nature labels, and calculates the results as town features. In this connectivity analysis, town features are calculated based on an index of proximity to natural areas. Examples of indicators include the distance from each city pixel to the nearest natural area pixel, the proportion of natural areas, the number of pixels in urban areas, and statistics on the coordinates of city pixels and natural areas. City features are calculated by integrating the proximity of natural areas with other indicators. Examples of other indicators include the proportion of natural areas and the visual proximity of natural areas. The proportion of natural areas refers to the proportion of pixels in the image that are natural areas such as forests, farmland, and bodies of water. The visual proximity of natural areas refers to an evaluation of how much of the field of view contains natural areas.

[0098] GIS data is data used to manage, analyze, and visualize geographic location information (Spatial Data) and attribute information (Attribute Data). GIS data is used in various fields to handle information about geographical phenomena or objects.

[0099] This section explains the basic concepts of GIS data (location information and attribute information). (1) Location information is information that indicates a location on Earth. Typically, a coordinate system is used to specify the location. Location information can be represented in the form of points, lines, or surfaces. (2) Attribute information is detailed information related to a geographical location. For example, road names, building uses, and land types are included in attribute information. Attribute information is used in combination with location information for analysis.

[0100] This section describes the main types of GIS data (vector data and raster data). (1) Vector data represents "points," "lines," and "polygons." A "point" is information that indicates a specific location. For example, a "point" indicates the location of a post office, school, etc. A "line" is information that connects multiple points in a continuous sequence. For example, a "line" represents a road, river, railway, etc. A "polygon" is an area enclosed by a closed line. For example, a "polygon" represents a lake, land parcel, administrative boundary, etc. (2) Raster data is data in grid format. In grid format, each cell arranged in a grid, like the pixels of an image, has a specific value. For example, land elevation, temperature, and satellite imagery can be represented in grid format. The value of each cell is numerical data or color information, making it suitable for dealing with continuous geographic variables.

[0101] Let's add some details about the trained model 301. The trained model 301 can be constructed using a neural network. A neural network consists of an input layer made up of multiple neurons, an intermediate layer (hidden layer) made up of multiple neurons, and an output layer made up of multiple neurons. The intermediate layer may be one layer or two or more layers. An example of a three-layer neural network will be explained based on Figure 8. Multiple input values ​​are input to the input layer (X1 to X3). In the input layer (X1 to X3), each input value is multiplied by a first weight (w11 to w16) to calculate multiple values. The multiple values ​​calculated in the input layer are input to the intermediate layer (Y1, Y2). In the intermediate layer (Y1, Y2), each value calculated in the input layer is multiplied by a second weight (w21 to w26) to calculate multiple output values. The multiple output values ​​are output from the output layer (Z1 to Z3). Multiple output values ​​change when either the first weights (w11-w16) or the second weights (w21-w26) are changed. The three-layer neural network takes remote sensing images 311 and GIS data 319 as input to the input layer and adjusts the first weights (w11-w16) and the second weights (w21-w26) so that the results output from the output layer approximate the land use map 312. This trains the pre-trained model 301.

[0102] The hardware configuration of the learning device 100 will be described based on Figure 9. The learning device 100 includes a processing circuit 109. The processing circuit 109 is the hardware that implements the data acquisition unit 110 and the model generation unit 120. The processing circuit 109 may be dedicated hardware, or it may be a processor 101 that executes a program stored in the memory 102.

[0103] If the processing circuit 109 is dedicated hardware, the processing circuit 109 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC, an FPGA, or a combination thereof. ASIC is an abbreviation for Application Specific Integrated Circuit. FPGA is an abbreviation for Field Programmable Gate Array.

[0104] The learning device 100 may include multiple processing circuits that replace the processing circuit 109.

[0105] In the processing circuit 109, some functions may be implemented by dedicated hardware, while the remaining functions may be implemented by software or firmware.

[0106] Thus, the functions of the learning device 100 can be realized by hardware, software, firmware, or a combination thereof.

[0107] The hardware configuration of the inference device 200 will be described based on Figure 10. The inference device 200 includes a processing circuit 209. The processing circuit 209 is the hardware that implements the data acquisition unit 210, the inference unit 220, and the connectivity analysis unit 230. The processing circuit 209 may be dedicated hardware, or it may be a processor 201 that executes a program stored in the memory 202.

[0108] If the processing circuit 209 is dedicated hardware, the processing circuit 209 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC, an FPGA, or a combination thereof.

[0109] The inference device 200 may include multiple processing circuits that replace the processing circuit 209.

[0110] In the processing circuit 209, some functions may be implemented by dedicated hardware, while the remaining functions may be implemented by software or firmware.

[0111] Thus, the functions of the inference device 200 can be realized by hardware, software, firmware, or a combination thereof.

[0112] Embodiment 1 is an example of a preferred embodiment and is not intended to limit the technical scope of this disclosure. Embodiment 1 may be implemented in part or in combination with other embodiments. The procedure described using flowcharts, etc., may be modified as appropriate.

[0113] The town feature analysis system 300 may be implemented with one device or three or more devices. Each element of the learning device 100 and the inference device 200 may be implemented with software, hardware, firmware, or a combination thereof. The word "part" in each element of the learning device 100 and the inference device 200 may be read as "processing," "process," "circuit," or "circuit."

[0114] 100 Learning device, 101 Processor, 102 Memory, 103 Auxiliary storage device, 104 Communication device, 105 Input / Output interface, 109 Processing circuit, 110 Data acquisition unit, 120 Model generation unit, 190 Storage unit, 200 Inference device, 201 Processor, 202 Memory, 203 Auxiliary storage device, 204 Communication device, 205 Input / Output interface, 209 Processing circuit, 210 Data acquisition unit, 220 Inference unit, 230 Connectivity analysis unit, 290 Storage unit, 300 City feature analysis system, 301 Trained model, 302 Land use map, 303 City features, 311 Remote sensing image, 312 Land use map, 319 GIS data, 321 Remote sensing image, 329 GIS data.

Claims

1. A town feature analysis system comprising: an inference unit that takes remote sensing images as input and uses a trained model to obtain a land use map in which each pixel is labeled to indicate the land use classification of the location corresponding to that pixel; and a connectivity analysis unit that uses the obtained land use map to calculate town features, which are features related to regional well-being.

2. The connectivity analysis unit sets up an urban area mainly composed of urban pixels with urban labels and a natural area mainly composed of natural pixels with natural labels, calculates the distance from each urban pixel in the urban area to a natural area pixel in the natural area, and calculates the statistics of the calculated distances as the urban features, according to claim 1.

3. The connectivity analysis unit sets an urban area mainly composed of urban pixels with urban labels and a natural area mainly composed of natural pixels with natural labels, and calculates the proportion of the natural area within a certain range including the urban area as the urban feature, according to claim 1.

4. The town feature analysis system according to claim 1, wherein the connectivity analysis unit calculates the proportion of natural pixels with natural labels as the town feature.

5. The urban feature analysis system according to claim 1, wherein the connectivity analysis unit sets a natural area mainly composed of natural pixels with natural labels, and calculates the diversity index of the natural area within a certain range including each city pixel with a city label as the urban feature.

6. The urban feature analysis system according to claim 1, wherein the connectivity analysis unit sets an urban area mainly composed of urban pixels with urban labels, and calculates the proportion of green space pixels with green space labels within a certain range including the urban area as the urban feature.

7. The urban feature analysis system according to claim 1, wherein the connectivity analysis unit sets a natural area mainly composed of natural pixels with natural labels, calculates the amount of movement to the natural area for each urban pixel with an urban label, and calculates the statistics of the calculated amount of movement as the urban feature.

8. The urban feature analysis system according to claim 1, wherein the connectivity analysis unit sets a natural area mainly composed of natural pixels with natural labels, and calculates the visual complexity of the natural area pixels, which are pixels within the natural area, as the urban feature.

9. A town feature analysis system according to claim 1, wherein two or more types of natural connectivity are defined in advance, the connectivity analysis unit calculates feature quantities for each of the two or more types of natural connectivity defined in advance, and integrates the calculated feature quantities to calculate the town feature quantity.

10. Two or more types of natural connectivity are defined in advance, at least two of which are "proximity to natural areas," "visual proximity to natural areas," "proportion of natural areas," "diversity of natural areas," "green coverage rate," "accessibility to nature," and "visual complexity of nature." The connectivity analysis unit sets up an urban area mainly composed of urban pixels with urban labels and a natural area mainly composed of natural pixels with natural labels, calculates the distance from each urban area pixel (a pixel within the urban area) to a natural area pixel (a pixel within the natural area), calculates the statistics of the calculated distance as a feature for "proximity to natural areas," sets up the urban area and the natural area, calculates the proportion of the natural area within a certain range including the urban area as a feature for "visual proximity to natural areas," calculates the proportion of natural pixels with natural labels as a feature for "proportion of natural areas," sets up the natural area, and calculates the diversity index of the natural area within a certain range including each urban pixel as a feature for "diversity of natural areas." A city feature analysis system according to claim 9, comprising: setting the urban area; calculating the proportion of green space pixels with green space labels within a certain range including the urban area as a feature for "green coverage"; setting the natural area; calculating the amount of travel to the natural area for each urban pixel with an urban label; calculating the statistics of the calculated amount of travel as a feature for "accessibility to nature"; and setting the natural area and calculating the visual complexity of the natural area pixels as a feature for "visual complexity of nature".

11. The urban feature analysis system according to any one of claims 1 to 10, wherein the inference unit takes the remote sensing image and geographic information system data as input and uses the trained model to obtain the land use map.

12. A city feature analysis system according to any one of claims 1 to 10, comprising a model generation unit that generates the trained model by learning multiple sets of remote sensing images and land use maps for training.

13. The urban feature analysis system according to claim 12, wherein the model generation unit generates a trained model by learning a plurality of sets of remote sensing images for training, land use maps for training, and geographic information system data for training, and the inference unit takes the remote sensing images for inference and geographic information system data for inference as input and uses the trained model to obtain the land use map.

14. A method for analyzing urban features, which involves using a trained model with remote sensing images as input to obtain a land use map in which each pixel is labeled to indicate the land use classification of the area corresponding to that pixel, and then using the obtained land use map to calculate urban features, which are features related to regional well-being.

15. An inference process that uses a pre-trained model with remote sensing images as input to obtain a land use map in which each pixel is labeled to indicate the land use classification of the location corresponding to that pixel; a connectivity analysis process that uses the obtained land use map to calculate urban features, which are features related to regional well-being; and an urban feature analysis program to cause a computer to perform the above.

16. A learning device comprising a model generation unit that creates a trained model by learning multiple sets of a remote sensing image for training and a land use map for training, which is a land use map in which each pixel in the remote sensing image for training is labeled to indicate the land use classification of the location corresponding to that pixel, wherein the trained model is a model that takes a remote sensing image as input and generates a land use map corresponding to the remote sensing image.

17. A learning method for creating a trained model by training multiple sets of training remote sensing images and training land use maps, which are land use maps in which each pixel in the training remote sensing image is labeled to indicate the land use classification of the area corresponding to that pixel, wherein the trained model is a model that takes a remote sensing image as input and generates a land use map corresponding to the remote sensing image.

Images