Method and device for generating start-end matrix of world cities based on multimodal public data

By using multimodal public data and advanced image processing technology, the target generation model is constructed, and the accuracy and cost of the generation of the starting and end point matrix of global cities is solved, and efficient and economical solutions are achieved around the world, suitable for different cities. type.

CN118761652BActive Publication Date: 2024-11-19TSINGHUA UNIVERSITY +1
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411245262.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-06
Publication Date
2024-11-19
Estimated Expiration
2044-09-06

AI Technical Summary

Technical Problem

Existing technologies are difficult to generate accurate endpoint matrix for all cities around the world, resulting in challenges in urban planning and traffic management, especially in areas where detailed data is lacking. Traditional methods are costly, difficult to obtain data, and cannot be applied to different urban types. .

Method used

Using a multimodal public data method, through satellite image data and population distribution data, image processing technology and spatial data analysis, a target generation model is constructed, and a globally applicable start and end point matrix is ​​generated, combining advanced image processing and machine learning algorithms. , realize the analysis of people's interaction between regions.

Benefits of technology

Provides a cost-effective and easy-to-operate global-level solution for any city, improving the efficiency and accuracy of starting and ending matrix generation, reducing data collection costs, and suitable for developing countries or regions lacking detailed data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118761652B_ABST
    Figure CN118761652B_ABST
Patent Text Reader

Abstract

The present disclosure relates to the field of mobile data generation, and in particular to a method and device for generating a start-end matrix of a world city based on multimodal public data. The method comprises: obtaining multimodal public data of a target area, the multimodal public data comprising satellite image data and target distribution data of the target area, the target distribution data comprising POI distribution data and / or population distribution data of the target area; according to the multimodal public data of the target area, calling a trained target generation model to output and obtain a start-end matrix of the target area, the start-end matrix being used to indicate the movement of people in the target area. The embodiment of the present disclosure uses multimodal public data from a variety of information sources that can be publicly obtained worldwide as model input, and can generate reliable and accurate start-end matrices for all cities in the world.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of mobile data generation, and in particular to a method and device for generating a start-end matrix of a world city based on multimodal public data. Background Art

[0002] The technology of generating mobility data for urban groups involves estimating the flow of people between different areas of the city, which is often expressed as a matrix from the origin to the destination, i.e. the origin-destination matrix (OD matrix) generation technology. This technology is widely used in urban planning, traffic management and digital twin cities. Through the generated mobility data, city managers can plan transportation facilities and urban infrastructure more reasonably, thereby avoiding the high urban transportation costs caused by unreasonable allocation of spatiotemporal resources.

[0003] However, there is currently no reasonable and effective solution that can generate reliable and accurate start-end matrices for all cities in the world to promote the formulation of more scientific development strategies. Summary of the invention

[0004] In view of this, the present disclosure proposes a method and device for generating a start-end matrix of world cities based on multimodal public data.

[0005] According to one aspect of the present disclosure, a method for generating a start-end matrix of world cities based on multimodal public data is provided, the method comprising:

[0006] Acquire multimodal public data of a target area, wherein the multimodal public data includes satellite image data and target distribution data of the target area, wherein the target distribution data includes points of interest (POI) distribution data and / or population distribution data of the target area;

[0007] According to the multimodal public data of the target area, the trained target generation model output is called to obtain the start-end point matrix of the target area, and the start-end point matrix is ​​used to indicate the movement of people in the target area.

[0008] In a possible implementation, the step of calling a trained target generation model output to obtain a start-end matrix of the target area according to the multimodal public data of the target area includes:

[0009] Performing feature extraction on the multimodal public data of the target area to obtain target area features, wherein the target area features include satellite image features and target distribution features of each of the multiple sub-areas of the target area, and the target distribution features include POI distribution features and / or population distribution features of the target area;

[0010] The target area features are input into the target generation model, and the start and end point matrix of the target area is obtained as output.

[0011] In another possible implementation, extracting features from the multimodal public data of the target area to obtain target area features includes:

[0012] According to the satellite image data of the target area, the trained feature extraction model is called to output the satellite image features of the multiple sub-areas of the target area; and according to the target distribution data of the target area, the target distribution features of the target area are statistically obtained.

[0013] In another possible implementation manner, the satellite image data of the target area includes original satellite images of each of the multiple sub-areas;

[0014] The method of calling the trained feature extraction model output to obtain satellite image features of the plurality of sub-regions of the target area according to the satellite image data of the target area includes:

[0015] Preprocessing the original satellite images of the plurality of sub-regions to obtain target satellite images of the plurality of sub-regions, wherein the preprocessing includes splicing and clipping;

[0016] The target satellite images of the respective sub-regions are input into the feature extraction model, and the satellite image features of the respective sub-regions are output.

[0017] In another possible implementation, the target generation model includes a graph denoising diffusion model, which is used to determine a noisy candidate start-end point matrix based on the characteristics of the target area, denoise the noisy candidate start-end point matrix, and output the start-end point matrix of the target area.

[0018] In another possible implementation, the acquiring of multimodal public data of the target area includes:

[0019] The satellite image data of the target area is obtained from a geographic database; and the POI distribution data of the target area is obtained from a global map database, and / or the population distribution data of the target area is obtained from a world population database.

[0020] In another possible implementation, the method further includes:

[0021] Acquire a training data set, the training data set comprising a plurality of sample data sets, each of which comprises sample multimodal public data and a corresponding label start-end matrix;

[0022] According to the training data set, the preset original model is trained to obtain the target generation model.

[0023] According to another aspect of the present disclosure, a device for generating a start-end matrix of world cities based on multimodal public data is provided, the device comprising:

[0024] An acquisition module, used to acquire multimodal public data of a target area, wherein the multimodal public data includes satellite image data and target distribution data of the target area, wherein the target distribution data includes POI distribution data and / or population distribution data of the target area;

[0025] A calling module is used to call the trained target generation model output to obtain the start and end point matrix of the target area according to the multimodal public data of the target area, and the start and end point matrix is ​​used to indicate the movement of people in the target area.

[0026] In a possible implementation, the calling module is further used to:

[0027] Performing feature extraction on the multimodal public data of the target area to obtain target area features, wherein the target area features include satellite image features and target distribution features of each of the multiple sub-areas of the target area, and the target distribution features include POI distribution features and / or population distribution features of the target area;

[0028] The target area features are input into the target generation model, and the start and end point matrix of the target area is obtained as output.

[0029] In another possible implementation, the calling module is further used to:

[0030] According to the satellite image data of the target area, the trained feature extraction model is called to output the satellite image features of the multiple sub-areas of the target area; and according to the target distribution data of the target area, the target distribution features of the target area are statistically obtained.

[0031] In another possible implementation manner, the satellite image data of the target area includes original satellite images of each of the multiple sub-areas;

[0032] The calling module is also used for:

[0033] Preprocessing the original satellite images of the plurality of sub-regions to obtain target satellite images of the plurality of sub-regions, wherein the preprocessing includes splicing and clipping;

[0034] The target satellite images of the respective sub-regions are input into the feature extraction model, and the satellite image features of the respective sub-regions are output.

[0035] In another possible implementation, the target generation model includes a graph denoising diffusion model, which is used to determine a noisy candidate start-end point matrix based on the characteristics of the target area, denoise the noisy candidate start-end point matrix, and output the start-end point matrix of the target area.

[0036] In another possible implementation, the acquisition module is further used to:

[0037] The satellite image data of the target area is obtained from a geographic database; and the POI distribution data of the target area is obtained from a global map database, and / or the population distribution data of the target area is obtained from a world population database.

[0038] In another possible implementation, the device further includes: a training module, configured to:

[0039] Acquire a training data set, the training data set comprising a plurality of sample data sets, each of which comprises sample multimodal public data and a corresponding label start-end matrix;

[0040] According to the training data set, the preset original model is trained to obtain the target generation model.

[0041] According to another aspect of the present disclosure, a device for generating a start-end matrix of world cities based on multimodal public data is provided, the device comprising:

[0042] processor;

[0043] a memory for storing processor-executable instructions;

[0044] Wherein, the processor is configured to implement the above method when executing the instructions stored in the memory.

[0045] According to another aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored, and the computer program instructions implement the above method when executed by a processor.

[0046] According to another aspect of the present disclosure, a computer program product is provided, including a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of a computing device, the processor in the computing device executes the above method.

[0047] The disclosed embodiment proposes a method for generating a start-end matrix based on multimodal public data worldwide and universal to all cities in the world. The method obtains multimodal public data of a target area, wherein the multimodal public data includes satellite image data and target distribution data of the target area, and the target distribution data includes POI distribution data and / or population distribution data of the target area; according to the multimodal public data of the target area, the trained target generation model is called to output the start-end matrix of the target area, and the start-end matrix is ​​used to indicate the movement of people in the target area; that is, the multimodal public data from various information sources that can be publicly obtained worldwide is used as the model input, and the start-end matrix generation task of any region in the world can be realized without geographical restrictions. In addition, the method combines advanced image processing technology and spatial data analysis, and is based on satellite image data and combined with target distribution data, wherein the target distribution data includes point of interest data and / or population distribution data of the target area. Through this information, the target generation model can construct a comprehensive regional spatial feature profile, and then deduce the interaction of people flow between regions, thereby improving the efficiency and accuracy of the start-end matrix generation task.

[0048] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0050] Figure 1 A schematic diagram of the structure of a computing device provided by an exemplary embodiment of the present disclosure is shown.

[0051] Figure 2 A flowchart of a method for generating a start-end point matrix of world cities based on multimodal public data provided by an exemplary embodiment of the present disclosure is shown.

[0052] Figure 3 A schematic diagram showing the principle of a method for generating a start-end point matrix of world cities based on multimodal public data provided by an exemplary embodiment of the present disclosure is shown.

[0053] Figure 4A flowchart of a method for generating an origin-destination matrix of world cities based on multimodal public data provided by another exemplary embodiment of the present disclosure is shown.

[0054] Figure 5 A schematic diagram of a process of acquiring satellite image data of a target area and performing preprocessing provided by an exemplary embodiment of the present disclosure is shown.

[0055] Figure 6 A schematic diagram of an image feature extraction process based on a feature extraction model provided by an exemplary embodiment of the present disclosure is shown.

[0056] Figure 7 A schematic diagram of a process of generating a start-end matrix using a target generation model provided by an exemplary embodiment of the present disclosure is shown.

[0057] Figure 8 A schematic diagram of a process of training and evaluating a target generation model provided by an exemplary embodiment of the present disclosure is shown.

[0058] Fig. 9 A schematic diagram of a process of generating past, present and future start-end matrices of a target city using a target generation model provided by an exemplary embodiment of the present disclosure is shown.

[0059] Fig.10 A schematic diagram of a process of inferring a start-end matrix of a target city based on satellite image data and other available data provided by an exemplary embodiment of the present disclosure is shown.

[0060] Fig.11 A schematic diagram showing the principles of a method for generating a start-end point matrix of world cities based on multimodal public data provided by another exemplary embodiment of the present disclosure is shown.

[0061] Fig.12 A schematic structural diagram of a device for generating a start-end point matrix of world cities based on multimodal public data provided by an exemplary embodiment of the present disclosure is shown.

[0062] Fig.13 A block diagram of a device for generating an origin-destination matrix of world cities based on multimodal public data is shown according to an exemplary embodiment. DETAILED DESCRIPTION

[0063] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0064] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0065] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0066] The origin-destination matrix can be viewed as a holistic view composed of multiple origin-destination flow data (i.e., OD flow data). Each OD flow data is a basic element that constitutes the origin-destination matrix, and these elements together describe the group flow between different areas of the city. It is not easy to obtain OD flow data anywhere in the world. Therefore, designing a computational model or method to generate an origin-destination matrix for anywhere in the world is a very valuable task. It should be clear that the generated OD flow data must truly reflect the movement of people within the area of ​​interest. In other words, it needs to be realistic enough to match the actual population flow pattern.

[0067] This is a very challenging task. Current work on origin-destination matrix generation cannot solve this problem well. Current work can be divided into two main categories. The first category includes models that mimic physical laws, such as gravity models and radiation models. These models predict OD flow data for unknown areas by considering limited factors such as population distribution and spatial distance associated with OD flow data. These models can be used to generate origin-destination matrices for any region in the world. However, due to their simplicity, their performance is not ideal. The second category is data-driven methods, such as random forests, graph neural networks, and deep gravity. These methods have become very popular recently. They consider a wider range of factors that may affect OD flow data, including sociodemographics, land use, etc. Then, they use complex models with a large number of learnable parameters to fit the relationship between these factors and OD flow data.

[0068] Related technologies include the following: 1. In related technology 1, a large amount of mobile phone signaling data is relied on to directly perform statistical analysis of group-level mobile data from intermittent point signaling data. This method faces several limitations, including high demand for data costs, data bias problems, and insufficient data reuse capabilities. These limitations are particularly significant in situations where it is challenging to obtain a large amount of continuous and high-cost mobile phone signaling data. 2. In related technology 2, data is directly collected through sensors (such as cameras) installed on the traffic road network to obtain vehicle travel information on the road network. This method has strict requirements on equipment and requires the widespread deployment of advanced sensor networks in the target city, which is not only costly, but also has a long installation and deployment cycle for the equipment. 3. In related technology 3, by collecting probe vehicle trajectory data, a microscopic start-end matrix representing the preferred path of a specific travel node is constructed, and then these matrices are aggregated into a comprehensive matrix to represent all travel nodes, thereby creating a start-end matrix. This method requires the collection of fine-grained trajectory data of a large number of probe vehicles, and also has high requirements on equipment and collection processes. 4. In the related technology 4, by dividing geographical areas and time periods, the flow of physical entities between different times and areas is counted to construct the start and end matrix of each time slot. This method requires very high manpower and time costs to complete.

[0069] Taking into account the research of related technologies, it can be found that the above methods have the following limitations, which hinder them from completing the task of generating a global origin-destination matrix: 1. These methods are designed to simulate OD traffic data under a specific urban style, and these methods do not work well when the target city is significantly different from the modeling context. 2. These methods have very high requirements for input data, which means that they require well-structured and information-rich input features. However, obtaining such input data also requires significant costs. Therefore, although these methods perform well, they are actually unusable in underdeveloped areas because these areas cannot obtain the input information required for these generative models. 3. The current patented technology of non-computational models mainly relies on a large amount of mobile phone positioning data or traffic network data to obtain the origin-destination matrix, resulting in extremely high data costs, which limits its widespread application. 4. The above-mentioned OD data acquisition method based on data collection requires very high labor costs and equipment deployment costs, which is very primitive, making it difficult to be widely used.

[0070] The disclosed embodiment proposes a method for generating an origin-destination matrix based on multimodal public data from all over the world, which is universal for all cities in the world. The multimodal public data from various information sources that can be publicly obtained from the world is used as the model input, aiming to solve the challenge and solve the origin-destination matrix generation task of any city type in the world. By utilizing the ability of the recent advanced large-scale multimodal model to extract features from satellite images, a system is designed that can generate an origin-destination matrix for any target area in the world. This system combines advanced image processing technology and spatial data analysis, based on satellite images, combined with POI data in the global map database and world population distribution data from the world population database, to extract key information such as urban structure, distance network and regional functions. With this information, the model can construct a comprehensive overview of urban spatial characteristics, and then deduce the interaction of people flow between regions. This method not only broadens the application scope of traditional origin-destination matrix generation technology, but also provides an economical, efficient and easy-to-operate global solution, which is particularly suitable for those areas that lack detailed origin-destination matrix data.

[0071] First, the application scenarios involved in this disclosure are introduced. Figure 1 , which shows a schematic diagram of the structure of a computing device provided by an exemplary embodiment of the present disclosure.

[0072] The computing device can be a terminal or a server. The terminal includes a mobile terminal or a fixed terminal, such as a mobile phone, a tablet computer, a laptop computer, a desktop computer, etc. The server can be a server, a server cluster consisting of several servers, or a cloud computing service center.

[0073] The computing device includes a processor 10, a memory 20, and a communication interface 30. Those skilled in the art will appreciate that Figure 1 The structure shown in the figure does not constitute a limitation on the computing device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0074] The processor 10 is the control center of the computing device, and uses various interfaces and lines to connect various parts of the entire computing device. By running or executing software programs and / or modules stored in the memory 20, and calling data stored in the memory 20, the processor 10 performs various functions of the computing device and processes data, thereby controlling the computing device as a whole. The processor 10 can be implemented by a central processing unit (CPU) or a graphics processing unit (GPU).

[0075] The memory 20 can be used to store software programs and modules. The processor 10 executes various functional applications and data processing by running the software programs and modules stored in the memory 20. The memory 20 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system 21, an acquisition module 22, a call module 23, and an application program 24 required for at least one function, etc.; the data storage area may store data created according to the use of the computing device, etc. The memory 20 may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk. Accordingly, the memory 20 may further include a memory controller to provide the processor 10 with access to the memory 20 .

[0076] The processor 10 executes the following functions by running the acquisition module 22: acquiring multimodal public data of the target area, the multimodal public data including satellite image data and target distribution data of the target area, the target distribution data including POI distribution data and / or population distribution data of the target area. The processor 10 executes the following functions by running the calling module 23: according to the multimodal public data of the target area, calling the trained target generation model output to obtain the start and end matrix of the target area, the start and end matrix is ​​used to indicate the movement of people in the target area.

[0077] The following uses several exemplary embodiments to introduce the method for generating a start-end matrix of world cities based on multimodal public data provided by the embodiments of the present disclosure.

[0078] Please refer to Figure 2 , which shows a flowchart of a method for generating a start-end matrix of world cities based on multimodal public data provided by an exemplary embodiment of the present disclosure. This embodiment uses this method for Figure 1 The method is illustrated by using the computing device shown in FIG.

[0079] Step 201, obtaining multimodal public data of a target area, wherein the multimodal public data includes satellite image data and target distribution data of the target area, wherein the target distribution data includes POI distribution data and / or population distribution data of the target area.

[0080] Optionally, the computing device obtains multimodal public data of the target area, which may include: obtaining satellite image data of the target area from a geographic database; and obtaining POI distribution data of the target area from a global map database, and / or obtaining population distribution data of the target area from a world population database. For example, the global map database is OpenStreetMap (OSM), and the world population database is the worldpop database.

[0081] Multimodal public data refers to data sets that are available in the public domain and contain multiple data types. Multimodal public data can include different types of data such as images, text, audio, video, etc. Multimodal public data can be freely accessed and used by the public without the need for payment or special permissions.

[0082] The target area is the specific geographic scope of the analysis or observation. The target area can be a city, a neighborhood, a natural area, or any other area with clear boundaries.

[0083] Satellite image data of a target area refers to images covering the target area captured by sensors carried by satellites. These images show features of the earth's surface, including but not limited to natural landforms, urban buildings, transportation networks, vegetation coverage and other surface features. Satellite image data may be remote sensing satellite image data.

[0084] Optionally, the satellite image data of the target area includes the original satellite images of the multiple sub-areas, and the original satellite images of the multiple sub-areas are preprocessed to obtain the target satellite images of the multiple sub-areas, and the preprocessing includes splicing and cropping. This step involves high-precision image processing technology to ensure that the images of each sub-area can be seamlessly connected, thereby providing accurate spatial reference for subsequent feature extraction and analysis. It should be noted that the preprocessing process can refer to the relevant details in the following embodiments, which will not be introduced here.

[0085] The target distribution data of the target area is used to indicate the functional characteristics and / or demographic characteristics of the target area. The target distribution data includes POI distribution data and / or population distribution data of the target area.

[0086] Point of interest is a term in geographic information systems, which refers to a specific location on a map. This point is usually associated with a specific place or landmark, such as a restaurant, store, museum, park entrance, etc. The POI distribution data of the target area is used to indicate the distribution of multiple points of interest in the target area, including the location, category and possible other attributes (such as business hours, contact information, etc.) of multiple points of interest.

[0087] Population distribution data of the target area is used to indicate the number and distribution of the population in the target area. Population distribution data may include total population, population density (population / area), demographic information of different age groups, gender, race, education level, etc. This data is usually obtained through census, sample survey or administrative records, and presented in geographic information systems at different resolutions (such as blocks, administrative areas, etc.).

[0088] Step 202, based on the multimodal public data of the target area, the trained target generation model output is called to obtain the start and end point matrix of the target area, and the start and end point matrix is ​​used to indicate the movement of people in the target area.

[0089] The computing device extracts features from the multimodal public data of the target area to obtain target area features, wherein the target area features include satellite image features and target distribution features of multiple sub-areas of the target area, and the target distribution features include POI distribution features and / or population distribution features of the target area; the target area features are input into a target generation model, and a start-end point matrix of the target area is obtained as output.

[0090] Satellite image features include the properties and patterns of the surface and its features extracted from satellite image data. Satellite image features may include one or more of the following: 1. Environmental features: including features such as urban layout, road network, vegetation coverage, and building density; 2. Spectral features: the intensity of spectral signals reflected or emitted by different objects, expressed in multiple bands. 3. Texture features: local patterns that appear repeatedly in satellite image data and their arrangement rules. 4. Shape features: the shape of artificial or natural objects, such as roads, buildings, lakes, etc. 5. Spatial relationship features: the relative position relationship between different objects.

[0091] POI distribution characteristics include the geographic location patterns and category distribution of multiple points of interest (such as businesses, services or public facilities) in the target area. POI distribution characteristics may include one or more of the following: 1. Category distribution: the proportion and distribution of different types of POIs (such as catering, retail, education, etc.) in the target area. 2. Spatial density: the number and degree of aggregation of POIs in the target area. 3. Accessibility: the accessibility of POIs relative to transportation networks or population centers. 4. Association pattern: the co-occurrence relationship or spatial correlation between different types of POIs.

[0092] Population distribution characteristics include the spatial distribution patterns of population size, structure, and social attributes within the target area. Population distribution characteristics may include one or more of the following: 1. Population density: the number of people per unit area. 2. Population structure: the distribution of demographic attributes such as age, gender, race, and education level. 3. Population spatial distribution: the distribution of population in cities, rural areas, or different geographical areas. 4. Population trend: the trend of population growth or decline, and migration patterns. 5. Socioeconomic attributes: socioeconomic characteristics such as income level, employment status, and education level. 6. Housing characteristics: housing type, residential density, home ownership, etc.

[0093] Optionally, the above-mentioned feature extraction step may include: based on the satellite image data of the target area, calling the trained feature extraction model output to obtain the satellite image features of multiple sub-areas of the target area; and based on the target distribution data of the target area, statistically obtaining the target distribution features of the target area.

[0094] Optionally, the target satellite images of each of the multiple sub-regions are input into a feature extraction model, and the satellite image features of each of the multiple sub-regions are output. The feature extraction model can be a visual module in the target generation model, or it can be an independent module different from the target generation model. The feature extraction model is used to extract satellite image features from the satellite image data of the target area. The satellite image features include but are not limited to environmental features such as urban layout, road network, vegetation coverage, and building density. The satellite image features of each of the multiple sub-regions of the target area are combined with the target distribution features (POI distribution features and / or population distribution features of the target area), which are crucial for understanding and predicting the movement of people between regions. It should be noted that the training and application process of the feature extraction model can refer to the relevant description in the following embodiments, which will not be introduced here.

[0095] The target area features obtained by feature extraction (satellite image features and target distribution features of multiple sub-areas of the target area) are used as input parameters of the target generation model, and the start-end matrix generation method is used for training and inference to generate the start-end matrix of the target area. This step involves complex data processing and machine learning algorithms, which aim to capture and learn the pattern of population flow from the characteristics of the target area, and then predict the OD flow between the sub-areas in the target area.

[0096] The computing device obtains the trained target generation model. In one possible implementation, when the computing device is a terminal, the terminal obtains the trained target generation model stored in itself, or obtains the trained target generation model from a server. In another possible implementation, when the computing device is a server, the server obtains the trained target generation model stored in itself.

[0097] The target generation model is a pre-trained model, which is used to generate the start-end matrix of the target area based on the input target area features. The target generation model is a model obtained by training the neural network using the sample multimodal public data and the corresponding label start-end matrix. That is, the target generation model is determined based on the sample multimodal public data and the corresponding label start-end matrix. Among them, the label start-end matrix is ​​the correct start-end matrix corresponding to the sample multimodal public data that is pre-labeled.

[0098] The target generation model is used to transform the target area features of the input multimodal public data into a start-end matrix.

[0099] The target generation model is used to represent the correlation between the target area features of multimodal public data and the start-end matrix.

[0100] The target generation model is a preset mathematical model, which includes model coefficients between the target area features of the multimodal public data and the start and end point matrix. The model coefficients can be fixed values, values ​​that are dynamically modified over time, and values ​​that are dynamically modified according to the usage scenario.

[0101] The target generation model may include a graph denoising diffusion model, which is used to determine the noisy candidate start-end matrix according to the target area characteristics, denoise the noisy candidate start-end matrix, and output the start-end matrix of the target area.

[0102] The graph denoising diffusion model is a special type of graph neural network model used to process and optimize noisy graph structure data. In the task of generating the origin-destination matrix, the graph denoising diffusion model is used to determine an initial candidate origin-destination matrix that may contain noise based on the characteristics of the target area (such as satellite image characteristics, population distribution characteristics, POI distribution characteristics, etc.). Then, by designing a denoising algorithm, the graph denoising diffusion model can reduce or eliminate the noise in the matrix, thereby purifying a more accurate origin-destination matrix to improve the accuracy and reliability of the prediction.

[0103] It should be noted that the training and application process of the target generation model can refer to the relevant description in the following embodiments and will not be introduced here.

[0104] The origin-destination matrix is ​​a tool used in transportation planning and geographic information systems. The origin-destination matrix of a target area is used to represent the flow of people or traffic from one location (origin) to another location (destination) in the target area. The rows of the matrix usually represent the origin, the columns represent the destination, and the values ​​in the matrix represent the flow from the origin to the destination.

[0105] The disclosed embodiment can comprehensively regard the target area and its starting and ending point matrix as a directed weighted graph with attributes, where nodes and edges represent sub-areas in the target area and OD flows between sub-areas, respectively. The features of the nodes may include the following: total population, area of ​​the sub-area, percentage of women in the total population (e.g., a floating point number between 0 and 1), and satellite image features obtained by feature extraction.

[0106] The origin-destination matrix plays a key role in many fields such as traffic planning, travel analysis, and urban planning. The functions of the origin-destination matrix include traffic flow prediction and management, travel behavior analysis and simulation, urban development and land use planning, etc. The generated origin-destination matrix can include the following functions: 1. Traffic flow prediction and management: The origin-destination matrix can be used to estimate the traffic flow of different routes, which is crucial for road traffic management and planning. By analyzing the flow between the origin and the destination, traffic signal scheduling can be optimized and traffic congestion can be reduced. 2. Travel behavior analysis and simulation: The origin-destination matrix provides detailed data on travel behavior, including the starting and ending points of the trip, which is crucial for understanding and simulating people's mobility patterns. For example, in urban planning, by analyzing the travel patterns of residents, public transportation systems can be better designed and optimized. 3. Urban development and land use planning: The origin-destination matrix also plays an important role in urban development and land use planning. By analyzing the distribution and migration paths of people, planners can more reasonably plan the location and scale of commercial areas, residential areas, and industrial areas. 4. Traffic policy evaluation and optimization: Relevant agencies can use the origin-destination matrix to evaluate the effectiveness of current traffic policies, such as impact assessments of traffic restrictions and traffic tax policies. This helps to adjust and formulate more effective policies to improve traffic conditions. 5. Emergency response and disaster management: In emergency situations, such as natural disasters or major accidents, the origin-destination matrix can assist decision makers in understanding the possible distribution and movement trends of the crowd, so as to effectively deploy rescue resources and conduct emergency evacuations. Therefore, the origin-destination matrix is ​​not only a simple tool for recording start and end point data, but also a multifunctional network analysis and planning tool. From traffic planning to disaster response, the origin-destination matrix generated by the disclosed embodiment has a wide range of applications and significant utility.

[0107] For illustration, please refer to Figure 3, which shows a schematic diagram of the principle of a method for generating a start-end matrix of a world city based on multimodal public data provided by an exemplary embodiment of the present disclosure. The data that requires high collection cost in the related technology (such as social demographic structure and interest point distribution data) is replaced with multimodal public data from multiple information sources that can be obtained efficiently and cost-free, and the target area features are extracted from the multimodal public data of the target area. The target area features are input into the trained target generation model, and the start-end matrix of the target area is output. The target area can be any city in the world. This method can comprehensively consider the characteristics and flow patterns of different cities, and process and analyze these data through the target generation model. As a result, this method can generate diversified and specific flow data for each city, overcome the limitations of traditional methods, and capture and reflect the heterogeneity between cities and their specific flow dynamics.

[0108] In summary, the disclosed embodiment proposes a method for generating a start-end matrix based on multimodal public data worldwide, which is universal for all cities in the world. The method obtains multimodal public data of a target area, wherein the multimodal public data includes satellite image data and target distribution data of the target area, and the target distribution data includes POI distribution data and / or population distribution data of the target area; according to the multimodal public data of the target area, the trained target generation model is called to output the start-end matrix of the target area, and the start-end matrix is ​​used to indicate the movement of people in the target area; that is, the multimodal public data from various information sources that can be publicly obtained worldwide is used as the input of the target generation model, and the start-end matrix generation task of any region in the world can be realized without geographical restrictions. In addition, the method combines advanced image processing technology and spatial data analysis, and is based on satellite image data and combined with target distribution data, wherein the target distribution data includes point of interest data and / or population distribution data of the target area. Through this information, the target generation model can construct a comprehensive regional spatial feature profile, and then deduce the interaction of people flow between regions, thereby improving the efficiency and accuracy of the start-end matrix generation task.

[0109] Please refer to Figure 4 , which shows a flowchart of a method for generating a start-end matrix of world cities based on multimodal public data provided by another exemplary embodiment of the present disclosure. This embodiment uses this method for Figure 1 The method is illustrated by using the computing device shown in FIG.

[0110] Step 401 : obtain satellite image data of a target area and perform preprocessing to obtain target satellite images of respective sub-areas of the target area.

[0111] Optionally, the schematic diagram of the computing device acquiring satellite image data of the target area and performing preprocessing is as follows: Figure 5 As shown. The computing device obtains satellite image data of the target area from the geographic database. The satellite image data of the target area includes original satellite images of multiple sub-areas (such as sub-area 1, sub-area 2, sub-area 3 and sub-area 4). The original satellite images of the multiple sub-areas are pre-processed (the pre-processing includes pixel-level stitching and cropping) to obtain target satellite images of the multiple sub-areas, so as to maximize the information related to human activities that can represent each sub-area.

[0112] Optionally, the computing device can perform a preprocessing process on the satellite image data of the target area based on the tile image, including but not limited to the following steps: 1. Determine the sub-region division: At the beginning of the processing, it is first necessary to subdivide the target area into manageable sub-regions based on the overall scope. This step is the basis of the entire processing flow and ensures that the subsequent steps can be carried out in a targeted manner. 2. Locate and download satellite images: Next, determine the corresponding satellite image tile coordinates based on the precise latitude and longitude coordinates of each sub-region. Using high-resolution satellite image services (such as Esri WorldImagery), download the original satellite images covering each sub-region to obtain a set of remote sensing satellite images covering the entire target area. 3. Preprocess (stitching and cropping) images: Once the original satellite images of each sub-region are obtained, a series of meticulous stitching and cropping processes are involved to customize each image to the specific size and shape of each specific sub-region in the region. This involves aligning and stitching multiple image tiles into a continuous image based on geographic location information, and then cropping according to the precise boundaries of the region. In order to accurately mark the boundaries of the target area, a boundary tracking algorithm is used. Based on the classification results of the cells, the algorithm identifies and tracks the boundaries of the regions through specific logical judgments. This process is critical to ensure that the satellite imagery accurately reflects the geographic and physical characteristics of each region, providing an accurate visual representation. In the cropping stage, it is critical to remove images that exceed the predefined boundaries of each region. Cells inside and outside the target region are marked separately, internal cells are marked as "inside", external cells are marked as "outside", and those contained within the boundaries of the target region are marked as "included". 4. Zero padding: In order to maintain the integrity of the image and avoid introducing irrelevant information, the edge areas are filled with zeros. 5. Resize the image: Finally, in order to make the image suitable for subsequent processing and analysis, the cropped image is adjusted to a uniform size. This effectively isolates the area of ​​interest and ensures that the pre-processed target satellite image specifically represents the corresponding sub-region.

[0113] Step 402: extract satellite image features of the plurality of sub-regions from the target satellite images of the plurality of sub-regions.

[0114] In the context of advanced regional function and feature analysis, the disclosed embodiment proposes an advanced feature extraction model, which aims to extract satellite image features of multiple sub-regions from target satellite images of multiple sub-regions. Based on current technology, identifying and interpreting satellite image features is crucial for human mobility analysis. However, there are obvious limitations to directly using raw satellite images as regional features, including but not limited to high noise in the images, data redundancy, and a large amount of information irrelevant to the analysis target, thereby reducing processing efficiency and analysis accuracy.

[0115] To overcome these limitations, the disclosed embodiments use a feature extraction model based on current advanced visual model technology, especially a large visual part specially designed and optimized for multimodal data, called RemoteCLIP (RemoteCLIP) model. The RemoteCLIP model has unique advantages in processing the unique features and complexity of satellite images, and can effectively extract information closely related to human activities and geographical features by combining multimodal learning mechanisms.

[0116] The disclosed embodiment may select the RN50 model in the RemoteCLIP model as the core feature extraction unit. The RN50 model, namely ResNet-50, is a 50-layer convolutional neural network that has been widely verified in visual recognition tasks and has demonstrated excellent generalization and zero-sample recognition capabilities through pre-training supervised by natural language. The RN50 model was selected because it provides an excellent balance between model complexity and performance, making it not only suitable for high-precision visual recognition tasks, but also able to effectively process large-scale satellite image data.

[0117] In the disclosed embodiment, the RN50 model is used to convert the original satellite image of each sub-region into a feature vector (e.g., a 1024-dimensional feature vector). This feature vector is used to indicate the most important information in the original satellite image, removing irrelevant noise and redundant information, so that it can be directly used as the input of the subsequent human mobility analysis and origin-destination matrix generation model. This processing process not only improves the data processing efficiency, but also improves the accuracy and reliability of the final analysis results.

[0118] In an illustrative example, a schematic diagram of an image feature extraction process based on a feature extraction model is shown in FIG. Figure 6 After obtaining the target satellite images of the multiple sub-regions (for example, the multiple sub-regions include sub-regions A to sub-region E), the target satellite images of the multiple sub-regions are input into the feature extraction model, and the satellite image features of the multiple sub-regions are output.

[0119] Step 403: input the satellite image features and target distribution features of the multiple sub-regions into the target generation model, and output the start and end point matrix of the target region.

[0120] Optionally, based on the satellite image features of each of the multiple sub-regions obtained, combined with the target distribution features of the target region (POI distribution features and / or population distribution features of the target region), a target generation model is used to generate a start-end matrix of the target region. The schematic diagram of the process of using the target generation model to generate a start-end matrix is ​​as follows: Figure 7 shown.

[0121] Schematically, the target generation model includes a graph denoising diffusion model, which is a graph conditional diffusion model on the edge of the node. The graph denoising diffusion model consists of two parts: the forward diffusion process q and the inverse denoising process Both processes are at the edge E and their corresponding weights The forward diffusion process is carried out in the attributed directed weighted graph space. The forward diffusion process can be described by the following calculation formula at each OD flow level.

[0122]

[0123] Among them, q is a conditional probability distribution, For the given conditions Next The probability distribution of is the flow rate from sub-region i to sub-region j at diffusion step t, is the flow rate from sub-region i to sub-region j at diffusion step t-1, represents a Gaussian distribution, is the default value, Used to control the noise size, I is a diagonal matrix. For the given conditions Next The probability distribution of is the flow rate from sub-area i to sub-area j at diffusion step 0, T is a preset value, and t, i, and j are all positive integers.

[0124] The inverse denoising process is shown in the following formula:

[0125]

[0126] in,

[0127]

[0128] . It is the intermediate state of diffusion step t-1 in the reverse denoising process. It is the intermediate state of diffusion step t in the reverse denoising process. It is the environmental feature that characterizes the target area and is used to control the diffusion process to obtain the output corresponding to the task and scene. is a given denoising network , intermediate state and the environmental characteristics of the target area The probability distribution of . is the default value, is the control parameter of the noise level in each diffusion step, is the control parameter of the accumulated noise level, the variable Based on the denoising network The predicted noise that needs to be removed.

[0129] Among them, the denoising network It can be a graph transformer (English: Transformer) network, which can accept inputs of nodes and edges at the same time, capture graph features, and then output noise predictions at the edge level. That is, in the graph transformer network, the information of each node and edge is aggregated through the self-attention mechanism, and the dependency between nodes and edges is simulated through the feature linear modulation (FiLM) layer. This structure enables the network to effectively process graph data, capture complex graph features, and output noise predictions at the edge level, thereby achieving efficient denoising.

[0130] The workflow of the graph transformer includes but is not limited to: 1. Input processing: Satellite image features of each sub-region are used as node inputs. The noisy start and end matrices provide edge inputs. These inputs are processed separately by their respective multilayer perceptrons (MLPs) and then fed into the graph transformer. 2. Self-attention mechanism: The graph transformer contains multiple layers. In each layer of the graph transformer, each node interacts with all other nodes to calculate attention weights through the self-attention mechanism and aggregates information from all other nodes based on these weights. 3. Feature-wise Linear Modulation (FiLM): In order to simulate the dependency between nodes and edges, the weights calculated by self-attention are combined with edge features using FiLM to obtain the final attention weights. 4. Information aggregation: The calculated attention weights are combined with the original edge features as new edge features in the calculation of the next layer of denoising network. 5. Feedforward network: After the node and edge information is aggregated, the data is passed through the feedforward network. 6. Noise prediction output: The network outputs noise predictions at the edge level for the inverse denoising process.

[0131] The computation within each graph transformer layer can be described by the following formula.

[0132]

[0133] in, and is the node feature of the lth layer, is the edge feature of the lth layer, They are The query, key, and value matrices of the k-th attention head at the layer, d k is the dimension of the query and key vectors, k is the number of attention heads, k is a positive integer, K is a preset value, is the attention score of the kth attention head in layer l, is the weight matrix of the k-th attention head in layer l, yes The attention weight obtained after normalization by the softmax function, is the edge feature of the l+1th layer, is the output multilayer perceptron of the lth layer edge feature, is the node feature of the l+1th layer, is the output multilayer perceptron of the layer node features, is related to node r i A set of connected neighbor nodes.

[0134] After multiple layers of processing, the final edge features are fed into the fully connected layer to predict the noise. In this way, the graph transformer network can not only process complex graph structure data, but also perform well in the denoising task, thereby outputting the start and end point matrix of the target area.

[0135] Optionally, before the computing device obtains the target generation model, it is necessary to train the training sample set to obtain the target generation model. The training process of the target generation model may include the following steps: obtaining a training data set, the training data set includes multiple groups of sample data groups, each sample data group includes sample multimodal public data and corresponding label start and end matrices; according to the training data set, training the preset original model to obtain the target generation model. By learning these training data sets, the target generation model can capture the complex relationships between different modal data, and use these relationships to predict the start and end matrices of new areas. The target generation model can be a large model specially designed to process multimodal public data, integrating information from multiple data sources such as satellite image data, POI distribution data, and population distribution data.

[0136] In a possible implementation, according to the training data set, the preset original model is trained to obtain the target generation model, which may include: for each sample data group in the training data set, extracting sample area features from the sample multimodal public data, the sample area features include satellite image features and sample distribution features of multiple sub-regions of the sample area, and the sample distribution features include POI distribution features and / or population distribution features of the sample area. Input the sample area features into the preset original model to obtain the training result. Compare the training result with the label start and end matrix to obtain the calculation loss, and the calculation loss is used to indicate the error between the training result and the label start and end matrix. According to the calculation loss corresponding to each of the multiple sample data groups, the target generation model is obtained by training using the error back propagation algorithm. Schematically, the gradient direction of the target generation model is determined according to the calculation loss by the back propagation algorithm, and the model parameters in the target generation model are updated layer by layer from the output layer of the target generation model. It should be noted that the embodiment of the present disclosure does not limit the training method of the target generation model.

[0137] In a possible application scenario, the target generation model provided by the embodiment of the present disclosure is trained and evaluated based on the attributes of all cities in country A, the group mobile flow data, and the start-end matrix. For example, the training set, the validation set, and the test set are divided into 8:1:1 to verify the effectiveness of the embodiment of the present disclosure. The schematic diagram of the process is shown in FIG. Figure 8 As shown. The data set includes regional demographic characteristic data corresponding to the spatial division of the census area, and POI distribution data from a public map service. Multimodal public data includes satellite image data, point of interest distribution data, and population distribution data. The scheme of the embodiment of the present disclosure is compared with the method in the related art to verify the advancement and effectiveness of the embodiment of the present disclosure. The evaluation indicators include Root Mean Square Error (RMSE) and Common Part of Commute (CPC). Schematically, the calculation method of the evaluation indicators is as follows:

[0138]

[0139] RMSE is the square root of the average of the squares of the differences between the predicted values ​​and the actual values. RMSE is used to measure the start and end point matrix of the prediction. The actual start and end matrix The difference between is the number of elements in the origin and destination matrix, R is the set of all sub-regions in the city, r i and r j represent the i-th and j-th sub-regions respectively, is the actual value from sub-region r iTo sub-region r j The starting and ending matrix of is the start and end matrix predicted by the model.

[0140] CPC is a ratio indicator used to measure the start and end matrix of the forecast The actual start and end matrix The degree of overlap between them. It calculates the proportion of the common parts between the two. This indicator reflects the degree of match between the predicted value and the actual value. The closer the CPC value is to 1, the more accurate the prediction is.

[0141] Finally, it can be concluded that, compared with the methods provided in the related art, the solution provided by the embodiments of the present disclosure can achieve considerably superior performance, as shown in Table 1.

[0142] Table 1

[0143]

[0144] In another possible application scenario, the solution provided by the embodiment of the present disclosure has specific applications in more fine-grained scenarios when generating start-end matrices for different regions of the world. For example, the generation of start-end matrices under counterfactual reasoning for long-term urban planning and transformation changes. In specific implementation, it is necessary to first collect historical urban satellite image data and population data, and also to collect current urban satellite image data. If possible, a target generation model can also be used to generate future urban satellite images or future images under given planning measures. On the basis of these data, the target generation model provided by the embodiment of the present disclosure is used to generate the past, present and future start-end matrices of the target city, and then the obtained data is analyzed. The schematic diagram of the process is shown in the figure. Fig. 9 When analyzing the data obtained, we can use spatial data analysis and causal inference techniques. Furthermore, we can use advanced statistical analysis methods, such as multivariate regression analysis, cluster analysis, and principal component analysis, to find out the key factors affecting urban development and population mobility. These analyses can reveal the advantages and disadvantages of urban infrastructure construction, transportation planning, economic development strategies, etc.

[0145] In another possible application scenario, the embodiments of the present disclosure can generate data for places where it is difficult to obtain traditional survey data. Specifically, the satellite image data of the target city is first collected, and then the model and algorithm proposed by the embodiments of the present disclosure are applied. In the absence of traditional survey data, the start and end matrix of the target city can be inferred based on the satellite image data and other available data. The schematic diagram of the process is shown in FIG. Fig.10This includes, but is not limited to, the movement of people from residence to work in different time periods, commuting patterns between cities and rural areas, and traffic hotspots and congestion points within the region. Next, using the generated origin-destination matrix, we can conduct an in-depth analysis of the traffic and population flow characteristics of the target city to help identify deficiencies in transportation infrastructure and public service facilities. This information is particularly important for regions with limited resources because it can guide policymakers and planners to make more effective decisions in transportation planning, urban development, and public service layout.

[0146] In an illustrative example, the schematic diagram of the principle of the method for generating the origin and destination matrix of world cities based on multimodal public data is as follows: Fig.11 As shown. The method involves the following steps: preprocessing of satellite image data, image feature extraction based on a feature extraction model, generation of start and end matrix based on a graph denoising diffusion model, and model training based on a training sample set. The details of each step can be referred to the relevant description in the above embodiment, which will not be repeated here.

[0147] The disclosed embodiment proposes a method for generating a start-end matrix based on multimodal public data from all over the world, which is universal for all cities in the world and can bring the following technical effects: 1. Global applicability: This method is designed to be applicable to all cities in the world without geographical restrictions, which makes it a universally applicable tool that can serve urban planning and traffic analysis on a global scale. 2. Multimodal data fusion: By combining multiple information sources such as satellite image data, POI distribution data, and population distribution data, this method can more comprehensively capture and analyze urban characteristics and personnel flow patterns. 3. Improve data acquisition efficiency: Using publicly available multimodal data as input reduces the cost and difficulty of data collection, which is particularly important for developing countries or regions that lack detailed start-end matrix data. 4. Model universality: Using a trained target generation model means that once the model is trained and verified, it can be widely applied to different cities and scenarios without the need to retrain the model for each new target area. 5. Advanced technology application: Using the latest large-scale multimodal model to extract features from satellite images, combined with advanced image processing technology and spatial data analysis, improves the accuracy and efficiency of analysis. 6. Cost-effective: Compared with traditional data collection and analysis methods, this method provides a cost-effective solution because it relies on public data and trained models, reducing the investment of human and material resources. 7. Easy to operate: Since the system is designed with user-friendliness in mind, even non-professionals can operate the system to generate and analyze the origin-destination matrix. 8. Support decision-making: The generated origin-destination matrix can provide valuable information for urban planners to help them better understand the movement of people within the city, so as to make more informed planning and decisions. 9. Strong adaptability: As the city grows and changes, the method can be easily updated and adjusted to reflect the latest urban conditions and population mobility trends. 10. Promote research and development: This method provides a powerful tool for academia and industry to study topics such as urban dynamics, traffic patterns and population migration, and promotes research and development in related fields. In summary, this solution provides an innovative and cost-effective solution for cities around the world to generate origin-destination matrices, which helps improve urban management and planning, while promoting a deeper understanding of urban dynamics.

[0148] The following is an apparatus embodiment of the present disclosure. For parts not described in detail in the apparatus embodiment, reference may be made to the technical details disclosed in the above method embodiment.

[0149] Please refer to Fig.12, which shows a schematic diagram of the structure of a device for generating a start-end matrix of world cities based on multimodal public data provided by an exemplary embodiment of the present disclosure. The device can be implemented in whole or in part by software, hardware, or a combination of the two, which is called a computing device. The device includes: an acquisition module 22 and a calling module 23.

[0150] An acquisition module 22 is used to acquire multimodal public data of a target area, wherein the multimodal public data includes satellite image data and target distribution data of the target area, wherein the target distribution data includes POI distribution data and / or population distribution data of the target area;

[0151] The calling module 23 is used to call the trained target generation model output to obtain the start and end matrix of the target area according to the multimodal public data of the target area. The start and end matrix is ​​used to indicate the movement of people in the target area.

[0152] In a possible implementation, the calling module 23 is further used to:

[0153] Extracting features from the multimodal public data of the target area to obtain target area features, where the target area features include satellite image features and target distribution features of multiple sub-areas of the target area, where the target distribution features include POI distribution features and / or population distribution features of the target area;

[0154] The target area features are input into the target generation model, and the start and end point matrix of the target area is output.

[0155] In another possible implementation, the calling module 23 is further used to:

[0156] According to the satellite image data of the target area, the trained feature extraction model is called to output the satellite image features of the multiple sub-areas of the target area; and according to the target distribution data of the target area, the target distribution features of the target area are statistically obtained.

[0157] In another possible implementation, the satellite image data of the target area includes original satellite images of each of the plurality of sub-areas;

[0158] The calling module 23 is also used for:

[0159] Preprocessing the original satellite images of the multiple sub-regions to obtain target satellite images of the multiple sub-regions, wherein the preprocessing includes stitching and clipping;

[0160] The target satellite images of the multiple sub-regions are respectively input into the feature extraction model, and the satellite image features of the multiple sub-regions are respectively obtained as output.

[0161] In another possible implementation, the target generation model includes a graph denoising diffusion model, which is used to determine the noisy candidate start-end point matrix according to the characteristics of the target area, denoise the noisy candidate start-end point matrix, and output the start-end point matrix of the target area.

[0162] In another possible implementation, the acquisition module 22 is further configured to:

[0163] Obtain satellite image data of the target area from a geographic database; and obtain POI distribution data of the target area from a global map database, and / or obtain population distribution data of the target area from a world population database.

[0164] In another possible implementation, the apparatus further includes: a training module, configured to:

[0165] Obtain a training data set, the training data set includes multiple groups of sample data sets, each group of sample data sets includes sample multimodal public data and corresponding label start and end matrices;

[0166] According to the training data set, the preset original model is trained to obtain the target generation model.

[0167] It should be noted that the device provided in the above embodiment only uses the division of the above-mentioned functional modules as an example to implement its functions. In practical applications, the above-mentioned functions can be assigned to different functional modules according to actual needs, that is, the content structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0168] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0169] The embodiment of the present disclosure also provides a device for generating a start-end point matrix of world cities based on multimodal public data, the device comprising: a processor; a memory for storing processor executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0170] The embodiment of the present disclosure further provides a non-volatile computer-readable storage medium on which computer program instructions are stored. When the computer program instructions are executed by a processor, the above method is implemented.

[0171] The embodiment of the present disclosure also provides a computer program product, including a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of a computing device, the processor in the computing device executes the above method.

[0172] Fig.13 A block diagram of a device 1900 for generating a start-end matrix of world cities based on multimodal public data is shown according to an exemplary embodiment. For example, the device 1900 can be provided as a server or a terminal device. Fig.13 , the apparatus 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932 for storing instructions, such as an application, that can be executed by the processing component 1922. The application stored in the memory 1932 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute instructions to perform the above method.

[0173] The device 1900 may also include a power supply component 1926 configured to perform power management of the device 1900, a wired or wireless network interface 1950 configured to connect the device 1900 to a network, and an input / output interface 1958 (I / O interface). The device 1900 may operate based on an operating system stored in the memory 1932, such as Windows ServerTM, MacOS XTM, UnixTM, LinuxTM, FreeBSDTM or the like.

[0174] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions, which can be executed by the processing component 1922 of the device 1900 to perform the above method.

[0175] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0176] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the above. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the above. The computer-readable storage medium used herein is not to be interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0177] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0178] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect through the Internet). In some embodiments, by using the state information of the computer-readable program instructions to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit may execute the computer-readable program instructions, thereby implementing various aspects of the present disclosure.

[0179] Various aspects of the present disclosure are described herein with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of boxes in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0180] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0181] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0182] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0183] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for generating the origin and destination matrix of world cities based on multimodal public data, It is characterized in that The method comprises: Acquire multimodal public data of a target area, wherein the multimodal public data includes satellite image data and target distribution data of the target area, wherein the target distribution data includes point of interest (POI) distribution data and / or population distribution data of the target area; According to the multimodal public data of the target area, the trained target generation model is called to output a start-end matrix of the target area, where the start-end matrix is ​​used to indicate the movement of people in the target area; The step of calling the trained target generation model output to obtain the start and end point matrix of the target area according to the multimodal public data of the target area includes: Performing feature extraction on the multimodal public data of the target area to obtain target area features, wherein the target area features include satellite image features and target distribution features of each of the multiple sub-areas of the target area, and the target distribution features include POI distribution features and / or population distribution features of the target area; Input the target area features into the target generation model, and output the start and end point matrix of the target area; The extracting features of the multimodal public data of the target area to obtain features of the target area includes: According to the satellite image data of the target area, calling the trained feature extraction model output to obtain the satellite image features of the multiple sub-areas of the target area; and according to the target distribution data of the target area, obtaining the target distribution features of the target area by statistics; Among them, the target generation model includes a graph denoising diffusion model, which is used to determine the noisy candidate start-end point matrix according to the characteristics of the target area, denoise the noisy candidate start-end point matrix, and output the start-end point matrix of the target area.

2. The method according to claim 1, It is characterized in that The satellite image data of the target area includes original satellite images of each of the plurality of sub-areas; The method of calling the trained feature extraction model output to obtain satellite image features of the plurality of sub-regions of the target area according to the satellite image data of the target area includes: Preprocessing the original satellite images of the plurality of sub-regions to obtain target satellite images of the plurality of sub-regions, wherein the preprocessing includes splicing and clipping; The target satellite images of the respective sub-regions are input into the feature extraction model, and the satellite image features of the respective sub-regions are output.

3. The method according to claim 1 or 2, It is characterized in that The obtaining of multimodal public data of the target area includes: The satellite image data of the target area is obtained from a geographic database; and the POI distribution data of the target area is obtained from a global map database, and / or the population distribution data of the target area is obtained from a world population database.

4. The method according to claim 1 or 2, It is characterized in that The method further comprises: Acquire a training data set, the training data set comprising a plurality of sample data sets, each of which comprises sample multimodal public data and a corresponding label start-end matrix; According to the training data set, the preset original model is trained to obtain the target generation model.

5. A device for generating the origin and destination matrix of world cities based on multimodal public data, It is characterized in that The device comprises: An acquisition module, used to acquire multimodal public data of a target area, wherein the multimodal public data includes satellite image data and target distribution data of the target area, wherein the target distribution data includes POI distribution data and / or population distribution data of the target area; A calling module, used to call the trained target generation model output to obtain the start-end matrix of the target area according to the multimodal public data of the target area, and the start-end matrix is ​​used to indicate the movement of people in the target area; The calling module is also used for: Performing feature extraction on the multimodal public data of the target area to obtain target area features, wherein the target area features include satellite image features and target distribution features of each of the multiple sub-areas of the target area, and the target distribution features include POI distribution features and / or population distribution features of the target area; Input the target area features into the target generation model, and output the start and end point matrix of the target area; The calling module is also used for: According to the satellite image data of the target area, calling the trained feature extraction model output to obtain the satellite image features of the multiple sub-areas of the target area; and according to the target distribution data of the target area, obtaining the target distribution features of the target area by statistics; Among them, the target generation model includes a graph denoising diffusion model, which is used to determine the noisy candidate start-end point matrix according to the characteristics of the target area, denoise the noisy candidate start-end point matrix, and output the start-end point matrix of the target area.

6. A device for generating the origin and destination matrix of world cities based on multimodal public data, It is characterized in that The device comprises: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to implement the method described in any one of claims 1 to 4 when executing the instructions stored in the memory.

7. A non-volatile computer-readable storage medium having computer program instructions stored thereon, It is characterized in that When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Citation Information

Patent Citations

  • Method and device for determining type of urban functional area

    CN113392859A

  • Method and device for generating population flow data at starting point and ending point

    CN113536865A

  • Characterization learning method and device of satellite image

    CN115620157A

  • Urban group mobile traffic network generation method and device

    CN118139097A

  • Method and apparatus for providing an indoor pedestrian origin-destination matrix and flow analytics

    US10921130B1