Land change simulation method and device based on convolutional neural network and Transformer

Abstract

The present invention relates to the field of data processing technology, and provides a land change simulation method based on a convolutional neural network and a transformer, comprising: extracting time-series spatial features from time-series land use data in a study area through a convolutional neural network, the time-series land use data in the study area includes a land use map of the study area at a preset time point, and the time-series spatial features include spatial features corresponding to the land use map at each time point; inputting the time-series spatial features and the time-series driving factor data of the time-series land use data into a transformer model to obtain suitability data of the time-series land use data; based on the suitability data, domain conditions, conversion restriction conditions, and random disturbance factors of the time-series land use data, obtaining a land use map in a target year. The present invention can provide a long-term high-precision land use change simulation result, and quickly respond to land use coverage and changes.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a land change simulation method and device based on convolutional neural network and Transformer. Background Art

[0002] Land use and land cover changes directly reflect the interaction between human activities and the natural environment, and have important impacts on various aspects such as ecological environment, urban planning, food safety, climate change and natural resource management. With the unprecedented wave of urbanization around the world, accurate modeling of land use and land cover change processes can help land managers conduct land use management and sustainable urban development. Especially in the context of the introduction of sustainable development goals, achieving high-precision characterization and simulation of land use and land cover change processes has become a research hotspot.

[0003] At present, various models have been developed to simulate the dynamic process of land use and change. These models can be divided into the following four types: statistical models, economic models, cellular automaton models and agent-based models.

[0004] However, the above models are difficult to capture the spatial characteristics of land use coverage and change processes, resulting in poor land use change simulation results. Summary of the invention

[0005] The present invention provides a land change simulation method and device based on convolutional neural network and Transformer, so as to solve the defect that it is difficult to capture the spatial characteristics of land use coverage and change process in the prior art.

[0006] The present invention provides a land change simulation method based on a convolutional neural network and a Transformer, the method comprising:

[0007] Proposing temporal spatial features from temporal land use data in a study area by means of a convolutional neural network, wherein the temporal land use data in the study area includes a land use map of the study area at a preset time point, and the temporal spatial features include spatial features corresponding to the land use map at each time point;

[0008] Inputting the temporal spatial characteristics and the temporal driving factor data of the temporal land use data into a Transformer model to obtain suitability data of the temporal land use data;

[0009] Based on the suitability data, domain conditions, conversion restriction conditions and random disturbance factors of the time series land use data, a land use map for the target year is obtained.

[0010] According to a land change simulation method based on a convolutional neural network and a Transformer provided by the present invention, the convolutional neural network is used to extract temporal spatial features from temporal land use data in a study area, including:

[0011] The land use map of the study area at each time point is sequentially input into a convolutional neural network, and the spatial features of the land use map at each time point are obtained through the convolutional neural network;

[0012] Based on the spatial characteristics of the land use map at each time point, temporal spatial characteristics are obtained.

[0013] According to a land change simulation method based on a convolutional neural network and a Transformer provided by the present invention, the temporal spatial characteristics and the temporal driving factor data of the temporal land use data are input into a Transformer model to obtain the suitability data of the temporal land use data, including:

[0014] The temporal spatial characteristics and the temporal driving factor data of the temporal land use data are input into the Transformer model to obtain the conversion adaptability of each cell in the land use map at each time point output by the Transformer model.

[0015] According to a land change simulation method based on a convolutional neural network and a Transformer provided by the present invention, the suitability data, domain conditions, conversion restriction conditions and random disturbance factors based on the time series land use data are used to obtain a land use map in a target year, including:

[0016] Based on the conversion adaptability of each cell at each time point, the domain condition of each cell, the conversion restriction condition and the random disturbance factor, the conversion probability of each cell at each time point is obtained;

[0017] Based on the conversion probability of each cell at each time point, the land use state of each cell at the next time point is obtained;

[0018] Based on the land use status of each cell at the next time point, a land use map for the target year is obtained.

[0019] According to a land change simulation method based on convolutional neural network and Transformer provided by the present invention, the domain condition of each cell is obtained in the following way:

[0020] Based on the land use type of each cell and the domain size of each cell, the domain condition of each cell is obtained.

[0021] According to a land change simulation method based on a convolutional neural network and a Transformer provided by the present invention, the method further includes:

[0022] Obtain time series land use data and time series driving factor data in the study area;

[0023] The time series land use data are clipped with a mask based on the administrative boundary of the study area to obtain the time series land use data in a raster file format with the same number of rows and columns as the coordinate system.

[0024] The present invention also provides a land change simulation device based on a convolutional neural network and a Transformer, the device comprising:

[0025] A temporal feature extraction module, used to extract temporal spatial features from the temporal land use data in the study area through a convolutional neural network, wherein the temporal land use data in the study area includes a land use map of the study area at a preset time point, and the temporal spatial features include spatial features corresponding to the land use map at each time point;

[0026] A location suitability calculation module, used for inputting the time series spatial characteristics and the time series driving factor data of the time series land use data into a Transformer model to obtain suitability data of the time series land use data;

[0027] The cellular automaton simulation module is used to obtain a land use map in a target year based on the suitability data, domain conditions, conversion restriction conditions and random disturbance factors of the time series land use data.

[0028] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a land change simulation method based on a convolutional neural network and a Transformer as described above is implemented.

[0029] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the land change simulation methods based on convolutional neural networks and Transformers as described above.

[0030] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the land change simulation methods based on convolutional neural networks and Transformers as described above.

[0031] The land change simulation method based on convolutional neural network and Transformer provided by the present invention proposes time-series spatial features from the time-series land use data in the study area through the convolutional neural network. The time-series land use data in the study area includes the land use map of the study area at a preset time point, and the time-series spatial features include the spatial features corresponding to the land use map at each time point; the time-series spatial features and the time-series driving factor data of the time-series land use data are input into the Transformer model to obtain the suitability data of the time-series land use data; based on the suitability data, domain conditions, conversion restrictions and random disturbance factors of the time-series land use data, the land use map in the target year is obtained. On the one hand, the present invention introduces a convolutional neural network to process the time-series land use data, extracts the time-series spatial features that reflect the land use coverage and changes; on the other hand, introduces the Transformer model to process the extracted time-series spatial features and time-series driving factors, obtains the conversion suitability of the cellular automaton, and combines the conversion restrictions, neighborhood conditions and random disturbance factors, and obtains the land use map in the target year through reasonable conversion rules, so as to provide a long-term high-precision land use change simulation result and quickly respond to land use coverage and changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 This is one of the flow charts of the land change simulation method based on convolutional neural network and Transformer provided by the present invention;

[0034] Figure 2 It is a schematic diagram of the structure of the convolutional neural network provided by the present invention;

[0035] Figure 3 It is a structural diagram of the Transformer model provided by the present invention;

[0036] Figure 4 This is the second flow chart of the land change simulation method based on convolutional neural network and Transformer provided by the present invention;

[0037] Figure 5 It is a structural schematic diagram of a land change simulation device based on a convolutional neural network and a Transformer provided by the present invention;

[0038] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] In the related art, various models have been developed to simulate the dynamic process of land use and change. These models can be divided into the following four types: statistical models, economic models, cellular automaton models and agent-based models.

[0041] Statistical models predict dynamic land use cover and change processes by establishing relationships between explanatory variables and land use cover and change patterns. These models have advantages in terms of ease of model establishment and effective simulation of quantitative land use changes. However, they often have difficulty capturing the spatial characteristics of land use cover and change. In addition, these models often perform poorly in situations where human decision-making is required.

[0042] Economic models simulate land use coverage and change processes by considering them from a market perspective, but ignore the spatial interactions between different land units.

[0043] Agent-based models simulate complex land use and change processes by considering multi-agent systems and their interactions. However, this model often requires a lot of computing resources and expert experience, and its generalization ability is limited.

[0044] The cellular automaton model uses specific transformation rules, neighborhood effects, and expert knowledge to analyze the spatial dynamics of change. The model can efficiently simulate the dynamic land use and cover change process and has excellent scalability and compatibility. Therefore, the cellular automaton model is often more popular and widely used.

[0045] Cellular automata are discrete dynamic process models that usually include five components: space, state, time step, neighborhood, and transition rules. Typically, cellular automata models divide space into discrete units of equal size, called cells. Each cell can be in different states, representing different properties. For example, in land use, cover, and change simulation, the state of each cell represents the land cover type at a specific time. Each cell has a neighborhood, including other cells adjacent to it. The size and shape of the neighborhood can be adjusted as needed, and in the study of land use, cover, and change simulation, it is usually in the form of a square area centered on the cell. The transition rule defines how the cell updates its state at each time step based on its current state and interaction with cells in the neighborhood. The extraction of transition rules directly determines the effect of land use, cover, and change simulation applied by cellular automata.

[0046] So far, various methods and models have been used to extract transformation rules, including artificial neural networks, random forests, and convolutional neural networks. However, most of these classic cellular automaton models are limited to describing the relationship between nonlinear factors such as driving factors and land use change, but rarely pay attention to the spatiotemporal dependencies in land use coverage and change processes.

[0047] Therefore, in response to the above problems, the present invention provides a land change simulation method based on convolutional neural network and Transformer. Figure 1 : is a flow chart of the land change simulation method based on convolutional neural network and Transformer provided by the present invention, such as Figure 1 As shown, the method includes:

[0048] Step 110, extracting temporal spatial features from the temporal land use data in the study area by using a convolutional neural network, wherein the temporal land use data in the study area includes a land use map of the study area at a preset time point, and the temporal spatial features include spatial features corresponding to the land use map at each time point;

[0049] Among them, time series land use data refers to the land use data collected at multiple time points (such as different years or quarters). These data are usually presented in the form of land use maps, that is, there is a land use map at each time point, and different colors or symbols on the land use map represent different land use types (such as forests, farmland, cities, etc.).

[0050] Temporal spatial features refer to the spatial characteristics that reflect land use coverage and changes at different time points. By extracting these features, we can better understand the dynamic change process of land use.

[0051] Reference here Figure 2, the convolutional neural network includes an input layer (Input), the first convolutional layer (C1), the first pooling layer (M1), the second convolutional layer (C2), the second pooling layer (M2), a fully connected layer (FC) and an output layer (Output). Through operations such as convolution (Conv), pooling (Pooling) and full connection (Fully connected), the convolutional neural network can extract spatial features from the input land use map.

[0052] Furthermore, in order to ensure that the convolutional neural network can accurately extract spatial features from the land use map, a batch of data is needed to train the convolutional neural network. For example, when the time points of the time series land use data are divided into years, the land use map of the previous year can be used as training data, and the land use map of the next year can be used as label data for training.

[0053] The training data is a tensor data of 1@N×N×T, where 1 represents the number of channels of the land use map, N represents the neighborhood size of the cells in the land use map, T represents the time series length of the land use map, and the label data is the land use classification data of the cells of the land use map of the next year.

[0054] Step 120, inputting the time series spatial characteristics and the time series driving factor data of the time series land use data into a Transformer model to obtain suitability data of the time series land use data;

[0055] Among them, time series driving factor data are those external factor data that can affect land use change, such as traffic driving factor data, location driving factor data, terrain driving factor data, etc. These data usually exist in the form of time series, describing how these data change over time.

[0056] Here, the Transformer model is a deep learning model based on the self-attention mechanism. Through the self-attention mechanism, the Transformer model can capture the dependency between any two positions in the input sequence, no matter how far apart they are. This enables the Transformer to capture long-term dependencies and generate accurate predictions when processing time series data.

[0057] Reference here Figure 3 The Transformer model consists of an input layer, 6 Transformer blocks, a global average pooling layer, a fully connected layer, a softmax layer, and an output layer. The Transformer block is the core component of the Transformer model, and each Transformer block contains multi-head attention and feedforward neural networks.

[0058] After the temporal spatial features and temporal driving factor data are input into the input layer, the features extracted from the input layer and the positional encoding (used to ensure that the model can understand the position of the elements in the sequence) are processed by the jump connection layer and then input into 6 Transformer blocks in turn. In each Transformer block, the complex dependencies between the features are first captured by multi-head attention, then normalized by the jump connection layer, and then input into the feedforward neural network. The features output by the feedforward neural network and the features normalized by the jump connection layer in the previous step are once again normalized by the jump connection layer, which is the output feature of the Transformer block. The features output by the last Transformer block are processed by the global average pooling layer, the fully connected layer, and the softmax layer in turn, and then input into the output layer. Finally, the suitability data of the temporal land use data is output through the output layer.

[0059] Step 130, based on the suitability data, domain conditions, conversion restriction conditions and random disturbance factors of the time series land use data, a land use map for the target year is obtained.

[0060] In this step, suitability data, domain conditions, conversion constraints and random disturbance factors based on time-series land use data are applied to the cellular automaton model to obtain the land use map for the target year.

[0061] The suitability data, domain conditions, and conversion constraints of time-series land use data are integrated into the cellular automaton model, and random disturbance factors are set. At initialization, the state of each cell is set to the current land use type. Through iterations of multiple time steps, the cellular automaton model can simulate the changing trend of land use types in the future. Based on these simulation results, the land use classification in the target year can be predicted.

[0062] In this embodiment, a convolutional neural network is used to extract temporal spatial features from the temporal land use data in the study area. The temporal land use data in the study area includes a land use map of the study area at a preset time point, and the temporal spatial features include spatial features corresponding to the land use map at each time point; the temporal spatial features and the temporal driving factor data of the temporal land use data are input into the Transformer model to obtain the suitability data of the temporal land use data; based on the suitability data, domain conditions, conversion constraints and random disturbance factors of the temporal land use data, a land use map in the target year is obtained. On the one hand, the present invention introduces a convolutional neural network to process the temporal land use data to extract temporal spatial features that reflect land use coverage and changes; on the other hand, a Transformer model is introduced to process the extracted temporal spatial features and temporal driving factors to obtain the conversion suitability of the cellular automaton, and together with the conversion constraints, neighborhood conditions and random disturbance factors, the land use map in the target year is obtained through reasonable conversion rules, thereby providing a long-term high-precision land use change simulation result and quickly responding to land use coverage and changes.

[0063] Based on the above embodiment, the method of extracting temporal spatial features from the temporal land use data in the study area by using a convolutional neural network includes:

[0064] The land use map of the study area at each time point is sequentially input into a convolutional neural network, and the spatial features of the land use map at each time point are obtained through the convolutional neural network;

[0065] Based on the spatial characteristics of the land use map at each time point, temporal spatial characteristics are obtained.

[0066] In this step, the land use map at each time point is input into the convolutional neural network as an input data. After being processed by the convolutional neural network, the land use map at each time point will obtain the corresponding spatial features.

[0067] After obtaining the spatial features of the land use map at each time point, these features can be combined into temporal spatial features. Temporal spatial features not only contain the spatial information at each time point, but also reflect the changing trends and patterns of these spatial information over time.

[0068] Based on the above embodiment, the time series spatial characteristics and the time series driving factor data of the time series land use data are input into the Transformer model to obtain the suitability data of the time series land use data, including:

[0069] The temporal spatial characteristics and the temporal driving factor data of the temporal land use data are input into the Transformer model to obtain the conversion adaptability of each cell in the land use map at each time point output by the Transformer model.

[0070] It can be understood here that in this embodiment, the space of the land use map is divided into discrete units of equal size, namely cells. Therefore, in this embodiment, cells are used as basic units for processing.

[0071] In this step, the above-mentioned temporal spatial characteristics and temporal driving factor data are used as input to the Transformer model. After processing, the Transformer model outputs the conversion adaptability of each cell in the land use map (i.e., each discrete unit after the land use map is divided) at each time point. Among them, the conversion adaptability refers to the possibility of the cell changing from one land use type to another land use type.

[0072] Refer to the following formula to get the conversion adaptability P of each cell: trans,i :

[0073] P trans,i =g tran,i (X spatial ,X driving );

[0074] Among them, X spatial is the temporal spatial feature, X driving It is the time series driving factor data.

[0075] Furthermore, in order to ensure that the Transformer model can accurately output the conversion adaptability of each cell in the land use map, a batch of data is needed to train the Transformer model. For example, when the time points are divided into years, sampling can be performed in the training set according to the set time span, and the land use type of the central cell from the starting year T+1 is used as the label data. The data is input into the Transformer model together, and the parameters are adjusted until the training accuracy reaches the standard.

[0076] Based on the above embodiment, the suitability data, domain conditions, conversion restriction conditions and random disturbance factors of the time series land use data are used to obtain a land use map in the target year, including:

[0077] Based on the conversion adaptability of each cell at each time point, the domain condition of each cell, the conversion restriction condition and the random disturbance factor, the conversion probability of each cell at each time point is obtained;

[0078] Based on the conversion probability of each cell at each time point, the land use state of each cell at the next time point is obtained;

[0079] Based on the land use status of each cell at the next time point, a land use map for the target year is obtained.

[0080] In this step, the conversion suitability P of the cell obtained in the previous step is trans,i , combined with the neighborhood condition Ω of each cell i,m , multiply the conversion restriction condition CONS and the random disturbance factor R together to calculate the final conversion probability of each cell. Then, by setting the conversion rules reasonably, the land use status of each cell at the next moment is obtained through the final conversion probability of each cell. The final land use change prediction results are summarized to obtain the land use map in the target year.

[0081] Specifically, refer to the following formula to convert the final conversion probability of each cell into the land use state S of each cell at the next time (t+1): i,t+1 :

[0082] S i,t+1 =f(S i,t ,f ann,i (X 0 ,X fac ,X st ),Ω i,m ,CONS,R);

[0083] Among them, S i,t represents the land use status of the ith cell at the current time (t), X 0 represents the initial state of the cell, X fac It represents the time series driving factor data, X st represents the temporal spatial characteristics, f ann,i represents the ANN function acting on the i-th cell.

[0084] Specifically, the domain condition Ω of each cell i,m It is obtained by:

[0085] Based on the land use type of each cell and the domain size of each cell, the domain condition of each cell is obtained.

[0086] Specifically, refer to the following formula:

[0087]

[0088] Among them, S i,kIt means that the i-th pixel is the k-th land use type, and m represents the neighborhood size of the cell.

[0089] Specifically, the random perturbation factor R is obtained in the following way:

[0090] R=1+(-lnγ) α ;

[0091] Among them, γ and α are adjustment parameters of random disturbance factors, the value range of γ is 0-1, and the value range of α is 1-10.

[0092] Based on the above embodiment, it also includes:

[0093] Obtain time series land use data and time series driving factor data in the study area;

[0094] The time series land use data are clipped with a mask based on the administrative boundary of the study area to obtain the time series land use data in a raster file format with the same number of rows and columns as the coordinate system.

[0095] First, determine the research area and the research scope of the research area, and collect historical data of the research scope of the research area, including historical time series land use data and historical time series driving factor data. The time series data should be kept at equal intervals as much as possible, and the interval years should be kept above a certain number of years, such as 5 years.

[0096] Since most of the time series land use data come from remote sensing images, the time series land use data are generally rectangular, while the administrative boundaries of the study area are generally irregular. In addition, some land use classification data are isosurface projections, which are different from the vector data used in normal situations. Therefore, in this embodiment, in order to ensure the processing of subsequent data, the time series land use data are clipped with the administrative boundaries of the study area as a mask to obtain the time series land use data in the raster file format with the same number of rows and columns as the coordinate system.

[0097] For ease of understanding, refer to Figure 4 This embodiment also proposes a land change simulation method based on a convolutional neural network and a Transformer, the method comprising:

[0098] Step 410: Collect time series land use data and time series driving factor data;

[0099] The method proposed in the present invention requires a series of input data, which mainly include two categories, time series land use data and time series driving factor data.

[0100] The source of time series land use data can be the land classification map data of 2005, 2010, 2015 and 2020 in the CLCD dataset. The initial data contains 8 land cover types with a spatial resolution of 30m. The time series driving factor data mainly considers three categories: traffic driving factors, based on POI data and road network data; location driving factors, based on administrative boundary zoning maps; terrain driving factors, based on terrain and slope data of the geospatial data cloud.

[0101] Step 420: Prepare a time series land use dataset and a time series driving factor dataset;

[0102] In this step, the collected data are clipped and processed using the administrative boundaries of the study area as a mask to prepare the input data for the subsequent model.

[0103] Step 430: extracting spatial features from the land use map through a convolutional neural network;

[0104] The land use map is used as the input of the convolutional neural network, and the changes in each cell in the land use map are used as label data. If a cell changes from one time step to the next time step (5 years later), it is labeled as 1; if a cell remains unchanged from one year to the next, it is labeled as 0.

[0105] Among them, the land use maps from 2005 to 2010 were used as training sets, and 5% of the pixels were randomly selected as training data. Finally, the spatial features of 2010, 2015 and 2020 were output.

[0106] Step 440, preparing transformation suitability by using the spatial features extracted from the time series driving factor data and the time series land use data through the Transformer structure;

[0107] The obtained spatial features are used as a driving factor and merged with the time series driving factor data. In the training phase, the time series driving factor data are from 2010 to 2015, where the land use map of 2015 is used as the label data. In the prediction phase, the input data is the time series driving factor data from 2015 to 2020, and the conversion suitability of each land use type in 2020 is output.

[0108] Step 450: Comprehensively consider conversion suitability, random factors, conversion restriction conditions, and neighborhood conditions, and simulate a land use map for the target year through cellular automation.

[0109] After calculating the conversion suitability, multiply the conversion suitability, random factor, conversion restriction condition, and neighborhood factor together to obtain the probability of each cell being converted into each land use type, and set the land use type with the highest probability as the land cover type of the cell at the next moment. In this way, the land use map of the target year can be simulated.

[0110] Based on any of the above embodiments, the present invention also provides a land change simulation device based on a convolutional neural network and a Transformer. Figure 5 : is a schematic diagram of the structure of a land change simulation device based on a convolutional neural network and a Transformer provided by the present invention, such as Figure 5 As shown, the device includes: a time series feature extraction module, a location suitability calculation module and a cellular automaton simulation module.

[0111] Specifically, the temporal feature extraction module is used to extract temporal spatial features from the temporal land use data in the study area through a convolutional neural network, wherein the temporal land use data in the study area includes a land use map of the study area at a preset time point, and the temporal spatial features include spatial features corresponding to the land use map at each time point;

[0112] A location suitability calculation module, used for inputting the time series spatial characteristics and the time series driving factor data of the time series land use data into a Transformer model to obtain suitability data of the time series land use data;

[0113] The cellular automaton simulation module is used to obtain a land use map in a target year based on the suitability data, domain conditions, conversion restriction conditions and random disturbance factors of the time series land use data.

[0114] The device provided by the embodiment of the present invention proposes time-series spatial features from the time-series land use data in the study area through a convolutional neural network. The time-series land use data in the study area includes a land use map of the study area at a preset time point, and the time-series spatial features include spatial features corresponding to the land use map at each time point; the time-series spatial features and the time-series driving factor data of the time-series land use data are input into the Transformer model to obtain the suitability data of the time-series land use data; based on the suitability data, domain conditions, conversion restrictions and random disturbance factors of the time-series land use data, the land use map in the target year is obtained. On the one hand, the present invention introduces a convolutional neural network to process the time-series land use data to extract the time-series spatial features that reflect the land use coverage and changes; on the other hand, the Transformer model is introduced to process the extracted time-series spatial features and time-series driving factors to obtain the conversion suitability of the cellular automaton, and in combination with the conversion restrictions, neighborhood conditions and random disturbance factors, the land use map in the target year is obtained through reasonable conversion rules, so as to provide a long-term high-precision land use change simulation result and quickly respond to land use coverage and changes.

[0115] The land change simulation device based on convolutional neural network and Transformer described in this embodiment and the land change simulation method based on convolutional neural network and Transformer provided by the present invention described above can be referred to each other, and will not be described in detail here.

[0116] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6 As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the land change simulation method based on the convolutional neural network and the Transformer, and the method includes:

[0117] Proposing temporal spatial features from temporal land use data in a study area by means of a convolutional neural network, wherein the temporal land use data in the study area includes a land use map of the study area at a preset time point, and the temporal spatial features include spatial features corresponding to the land use map at each time point;

[0118] Inputting the temporal spatial characteristics and the temporal driving factor data of the temporal land use data into a Transformer model to obtain suitability data of the temporal land use data;

[0119] Based on the suitability data, domain conditions, conversion restriction conditions and random disturbance factors of the time series land use data, a land use map for the target year is obtained.

[0120] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0121] On the other hand, the present invention further provides a computer program product, the computer program product comprising a computer program, the computer program can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer can execute the land change simulation method based on a convolutional neural network and a Transformer provided by the above methods, the method comprising:

[0122] Proposing temporal spatial features from temporal land use data in a study area by means of a convolutional neural network, wherein the temporal land use data in the study area includes a land use map of the study area at a preset time point, and the temporal spatial features include spatial features corresponding to the land use map at each time point;

[0123] Inputting the temporal spatial characteristics and the temporal driving factor data of the temporal land use data into a Transformer model to obtain suitability data of the temporal land use data;

[0124] Based on the suitability data, domain conditions, conversion restriction conditions and random disturbance factors of the time series land use data, a land use map for the target year is obtained.

[0125] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the land change simulation method based on a convolutional neural network and a Transformer provided by the above methods, the method comprising:

[0126] Proposing temporal spatial features from temporal land use data in a study area by means of a convolutional neural network, wherein the temporal land use data in the study area includes a land use map of the study area at a preset time point, and the temporal spatial features include spatial features corresponding to the land use map at each time point;

[0127] Inputting the temporal spatial characteristics and the temporal driving factor data of the temporal land use data into a Transformer model to obtain suitability data of the temporal land use data;

[0128] Based on the suitability data, domain conditions, conversion restriction conditions and random disturbance factors of the time series land use data, a land use map for the target year is obtained.

[0129] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0130] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A land change simulation method based on convolutional neural network and Transformer, characterized in that: The method comprises: Proposing temporal spatial features from temporal land use data in a study area by means of a convolutional neural network, wherein the temporal land use data in the study area includes a land use map of the study area at a preset time point, and the temporal spatial features include spatial features corresponding to the land use map at each time point; Inputting the temporal spatial characteristics and temporal driving factor data into the Transformer model to obtain the suitability data of the temporal land use data; Based on the suitability data, domain conditions, conversion restriction conditions and random disturbance factors of the time series land use data, a land use map in the target year is obtained; The step of inputting the temporal spatial characteristics and temporal driving factor data into the Transformer model to obtain the suitability data of the temporal land use data includes: Inputting the temporal spatial characteristics and temporal driving factor data into the Transformer model, and obtaining the conversion adaptability of each cell in the land use map at each time point output by the Transformer model; The land use map for the target year is obtained based on the suitability data, domain conditions, conversion restriction conditions and random disturbance factors of the time series land use data, including: Based on the conversion adaptability of each cell at each time point, the domain condition of each cell, the conversion restriction condition and the random disturbance factor, the conversion probability of each cell at each time point is obtained; Based on the conversion probability of each cell at each time point, the land use state of each cell at the next time point is obtained; Based on the land use status of each cell at the next time point, a land use map for the target year is obtained.

2. The land change simulation method based on convolutional neural network and Transformer according to claim 1, characterized in that: The method of extracting temporal spatial features from temporal land use data in the study area through convolutional neural network includes: The land use map of the study area at each time point is sequentially input into a convolutional neural network, and the spatial features of the land use map at each time point are obtained through the convolutional neural network; Based on the spatial characteristics of the land use map at each time point, the temporal spatial characteristics are obtained.

3. The land change simulation method based on convolutional neural network and Transformer according to claim 1, characterized in that: The domain condition of each cell is obtained in the following way: Based on the land use type of each cell and the domain size of each cell, the domain condition of each cell is obtained.

4. The land change simulation method based on convolutional neural network and Transformer according to any one of claims 1 to 3, characterized in that: The method further comprises: Obtain time series land use data and time series driving factor data in the study area; The time series land use data are clipped with a mask based on the administrative boundary of the study area to obtain the time series land use data in a raster file format with the same number of rows and columns as the coordinate system.

5. A land change simulation device based on convolutional neural network and Transformer, characterized in that: The device comprises: A temporal feature extraction module, used to extract temporal spatial features from the temporal land use data in the study area through a convolutional neural network, wherein the temporal land use data in the study area includes a land use map of the study area at a preset time point, and the temporal spatial features include spatial features corresponding to the land use map at each time point; A location suitability calculation module, used for inputting the temporal spatial characteristics and temporal driving factor data into the Transformer model to obtain suitability data of the temporal land use data; A cellular automaton simulation module is used to obtain a land use map in a target year based on suitability data, domain conditions, conversion restriction conditions, and random disturbance factors of the time series land use data; The step of inputting the temporal spatial characteristics and temporal driving factor data into the Transformer model to obtain the suitability data of the temporal land use data includes: Inputting the temporal spatial characteristics and temporal driving factor data into the Transformer model, and obtaining the conversion adaptability of each cell in the land use map at each time point output by the Transformer model; The land use map for the target year is obtained based on the suitability data, domain conditions, conversion restriction conditions and random disturbance factors of the time series land use data, including: Based on the conversion adaptability of each cell at each time point, the domain condition of each cell, the conversion restriction condition and the random disturbance factor, the conversion probability of each cell at each time point is obtained; Based on the conversion probability of each cell at each time point, the land use state of each cell at the next time point is obtained; Based on the land use status of each cell at the next time point, a land use map for the target year is obtained.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the land change simulation method based on convolutional neural network and Transformer as described in any one of claims 1 to 4 is implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the land change simulation method based on a convolutional neural network and a Transformer as described in any one of claims 1 to 4 is implemented.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the land change simulation method based on a convolutional neural network and a Transformer as described in any one of claims 1 to 4 is implemented.

Images