Land cover change monitoring method, model training method and related device

Abstract

The present disclosure relates to the field of computer and remote sensing technology, and discloses a land cover change monitoring method, a model training method and related devices; the method includes: obtaining original remote sensing time series data of the area to be monitored, and performing vectorization preprocessing on the original remote sensing time series data to obtain processed remote sensing time series data; based on a pre-constructed land cover type prediction model, processing the processed remote sensing time series data to obtain a land cover type prediction result sequence; the land cover type prediction model is used to predict the land cover types of various types of areas; based on the land cover type prediction result sequence, the land cover changes of the area to be monitored are monitored. The present disclosure utilizes a prediction model that can predict the land cover types of various types of areas, and can quickly and accurately generate a land cover type prediction result sequence for the area to be monitored, thereby realizing accurate monitoring of land cover changes in the area to be monitored.

Description

Technical Field

[0001] The present disclosure relates to the field of computer and remote sensing technology, and in particular to a land cover change monitoring method, a model training method and related devices. Background Art

[0002] Land cover change refers to the change in area and conversion between types of surface vegetation cover (forests, grasslands, cultivated vegetation, etc.) and non-vegetation cover (ice and snow, etc.) due to climate change and human activities.

[0003] In recent years, with the intensification of global warming, the continuous growth of population, and the rapid advancement of industrialization and urbanization, the process of land cover change has become more drastic and frequent. The complexity and uncertainty of land cover change have brought unprecedented challenges to natural resource management, ecological environment protection, and sustainable social and economic development. In order to effectively respond to these challenges, it is urgent to develop a method that can accurately predict land cover change. Summary of the invention

[0004] In view of the above situation, the embodiments of the present disclosure provide a land cover change monitoring method, a model training method and related devices, aiming to solve the above problems or at least partially solve the above problems.

[0005] In a first aspect, an embodiment of the present disclosure provides a land cover change monitoring method, the method comprising:

[0006] Acquire original remote sensing time series data of the area to be monitored, and perform vectorization preprocessing on the original remote sensing time series data to obtain processed remote sensing time series data;

[0007] Based on a pre-built land cover type prediction model, the processed remote sensing time series data is processed to obtain a land cover type prediction result sequence; the land cover type prediction model is used to predict the land cover types of various types of regions;

[0008] Based on the land cover type prediction result sequence, the land cover changes of the area to be monitored are monitored.

[0009] In a second aspect, the present disclosure also provides a model training method for training the land cover type prediction model as described in the first aspect, the method comprising:

[0010] Acquire a data set; the samples in the data set include sample remote sensing time series data of various types of regions, and the labels of the samples are corresponding real land cover type sequences;

[0011] The data set is used to train an initial land cover type prediction model with time series data analysis capability to obtain a land cover type prediction model.

[0012] In a third aspect, an embodiment of the present disclosure further provides a land cover change monitoring device, the device comprising:

[0013] An acquisition module is used to acquire original remote sensing time series data of the area to be monitored, and perform vectorization preprocessing on the original remote sensing time series data to obtain processed remote sensing time series data;

[0014] A prediction module is used to process the processed remote sensing time series data based on a pre-built land cover type prediction model to obtain a land cover type prediction result sequence; the land cover type prediction model is used to predict the land cover types of various types of regions;

[0015] A monitoring module is used to monitor the land cover changes in the area to be monitored based on the land cover type prediction result sequence.

[0016] In a fourth aspect, the present disclosure also provides a model training device for training the land cover type prediction model as described in the first aspect, the device comprising:

[0017] An acquisition module is used to acquire a data set; the samples in the data set include sample remote sensing time series data of various types of regions, and the labels of the samples are corresponding real land cover type sequences;

[0018] The training module is used to train an initial land cover type prediction model with time series data analysis capability using the data set to obtain a land cover type prediction model.

[0019] In a fifth aspect, an embodiment of the present disclosure further provides an electronic device, comprising: a processor; and a memory arranged to store computer executable instructions, which, when executed, cause the processor to perform the steps of the above-mentioned land cover change monitoring method or model training method.

[0020] In a sixth aspect, an embodiment of the present disclosure also provides a computer-readable storage medium, which stores one or more programs. When the one or more programs are executed by an electronic device including multiple application programs, the electronic device executes the steps of the above-mentioned land cover change monitoring method or model training method.

[0021] By means of the above technical solution, the land cover change monitoring method, model training method and related devices provided by the embodiments of the present disclosure can pre-process the original remote sensing time series data of the area to be monitored, and input the processed remote sensing time series data into the pre-built land cover type prediction model to obtain a land cover type sequence. Among them, the land cover type prediction model can be used to predict the land cover types of various types of areas, has a strong generalization ability, and can adapt to different geographical and environmental conditions. Therefore, by using the model to predict the processed remote sensing time series data of the area to be monitored, the corresponding land cover type prediction result sequence can be obtained quickly and accurately, thereby realizing accurate monitoring of land cover changes in the area to be monitored based on the land cover type prediction result sequence.

[0022] The above description is only an overview of the technical solution of the present disclosure. In order to more clearly understand the technical means of the present disclosure, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present disclosure more obvious and easy to understand, the specific implementation methods of the present disclosure are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The drawings described herein are used to provide a further understanding of the present disclosure and constitute a part of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure. In the drawings:

[0024] Figure 1 A schematic diagram of the process of the land cover change monitoring method provided by the embodiment of the present disclosure is shown;

[0025] Figure 2 A schematic diagram of a process for generating a set of data points to be processed provided by an embodiment of the present disclosure is shown;

[0026] Figure 3 The Transformer model architecture diagram provided by the embodiment of the present disclosure is shown;

[0027] Figure 4 A schematic diagram of a flow chart of a model training method provided by an embodiment of the present disclosure is shown;

[0028] Figure 5 A schematic diagram of the process of the land cover change monitoring method provided by the embodiment of the present disclosure is shown;

[0029] Figure 6 A schematic diagram of the structure of a land cover change monitoring device provided by an embodiment of the present disclosure is shown;

[0030] Figure 7 A schematic diagram of the structure of a model training device provided by an embodiment of the present disclosure is shown;

[0031] Figure 8 A schematic structural diagram of an electronic device provided by an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solutions and advantages of the present disclosure clearer, the technical solutions of the present disclosure will be clearly and completely described below in combination with the specific embodiments of the present disclosure and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.

[0033] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0034] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that such usage is interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the term "including" and its variants are to be interpreted as open-ended terms meaning "including but not limited to".

[0035] As mentioned above, in recent years, with the intensification of global warming, the continuous growth of population, and the rapid advancement of industrialization and urbanization, the land cover change process has become more drastic and frequent. The complexity and uncertainty of land cover change have brought unprecedented challenges to natural resource management, ecological environment protection, and social and economic sustainable development. In order to effectively respond to these challenges, the present invention proposes a land cover change monitoring method, a model training method, and related devices. The present disclosure is described in detail below through specific embodiments.

[0036] It should be noted that the execution subject of the land cover change monitoring method and model training method provided in the embodiments of the present disclosure is generally a computer device with certain computing capabilities, and the computer device includes, for example: a terminal device or a server or other processing device, and the terminal device can be a user equipment (User Equipment, UE), a mobile device, a user terminal, a terminal, a personal digital assistant (Personal Digital Assistant, PDA), a handheld device, etc. In some possible implementations, the land cover change monitoring method and model training method can be implemented by a processor calling a computer-readable instruction stored in a memory.

[0037] Figure 1 The schematic diagram of the process of the land cover change monitoring method provided by the embodiment of the present disclosure is shown. Figure 1 It can be seen that the embodiment of the present disclosure at least includes steps S101-S103:

[0038] S101: Acquire original remote sensing time series data of the area to be monitored, and perform vectorization preprocessing on the original remote sensing time series data to obtain processed remote sensing time series data.

[0039] Here, the original remote sensing time series data includes a plurality of remote sensing data arranged in time sequence. Specifically, each remote sensing data may include at least one of the following items: a reflectance matrix of at least one band, a brightness matrix of at least one band, a temperature matrix of at least one band, a vegetation index of at least one band, etc.

[0040] This embodiment does not limit the type and scope of the area to be monitored. The type of area to be monitored includes but is not limited to: farmland, city, forest, water area, desert, etc.

[0041] This embodiment does not limit the means for obtaining the original remote sensing time series data, for example, it can be obtained from a satellite remote sensing data observation platform. It is also possible to make an aircraft carrying a specific remote sensing device fly according to a preset route. During the flight, the remote sensing device observes and collects data on the ground below, thereby obtaining the original remote sensing time series data.

[0042] In order to facilitate the subsequent processing and analysis of the land cover type prediction model, the original remote sensing time series data can be preprocessed by vectorization to obtain the processed remote sensing time series data.

[0043] S102: Based on a pre-built land cover type prediction model, the processed remote sensing time series data is processed to obtain a land cover type prediction result sequence; the land cover type prediction model is used to predict the land cover types of various types of regions.

[0044] S103: Based on the land cover type prediction result sequence, monitoring the land cover change of the area to be monitored.

[0045] from Figure 1It can be seen from the method shown that the disclosed embodiment can pre-process the original remote sensing time series data of the area to be monitored, and input the processed remote sensing time series data into the pre-built land cover type prediction model to obtain a land cover type sequence. Among them, the land cover type prediction model can be used to predict the land cover types of various types of areas, has a strong generalization ability, and can adapt to different geographical and environmental conditions. Therefore, using the model to predict the processed remote sensing time series data of the area to be monitored, the corresponding land cover type prediction result sequence can be obtained quickly and accurately, thereby based on the land cover type prediction result sequence, accurate monitoring of land cover changes in the area to be monitored can be achieved.

[0046] Furthermore, in order to better illustrate the process of the above-mentioned land cover change monitoring method, as a refinement and extension of the above-mentioned embodiment, the embodiment of the present invention provides several optional embodiments, but is not limited thereto, as shown below.

[0047] In actual application scenarios, if there are many data points in the original remote sensing time series data, the computational cost of the prediction model will be significantly increased, reducing the prediction efficiency of the model. In addition, when there are too many data points in the original remote sensing time series data, the observation time between the data points is too far apart, and the correlation is weak. When the model in the training stage does not learn the patterns contained in the sample data with a long time span, the accuracy of the model prediction will also be low. Based on this, in a possible implementation of the present disclosure, the pre-constructed land cover type prediction model processes the processed remote sensing time series data to obtain a land cover type prediction result sequence, including:

[0048] The land cover type prediction model is used to process the remote sensing data corresponding to the first preset number of data points in the processed remote sensing time series data to obtain an initial land cover type prediction result sequence; a sliding window with the preset number as the size is made to slide backward on the processed remote sensing time series data in chronological order and with a preset sliding step length to obtain a set of data points to be processed; the land cover type prediction model is used to process the remote sensing data corresponding to the new data point in the set of data points to be processed and the target land cover type in the initial land cover type prediction result sequence to obtain a land cover type prediction result for the new data point; the target land cover type is the The land cover type prediction results of the data points in the set of data points to be processed except the new data point; based on the land cover type prediction results of the new data point, an updated land cover type prediction result sequence is obtained; if the updated land cover type prediction result sequence includes the land cover type prediction result of the last data point in the processed remote sensing time series data, the updated land cover type prediction result sequence is used as the land cover type prediction result sequence; otherwise, return to the step of sliding the sliding window with the preset number as the size backward on the processed remote sensing time series data in chronological order and with the preset sliding step size to obtain the set of data points to be processed.

[0049] In this embodiment, the preset number and the preset sliding step size can be set according to actual needs, and this embodiment does not limit this. Figure 2 The schematic diagram of the process of generating a set of data points to be processed provided by the embodiment of the present disclosure is shown below. Figure 2 This implementation is described by way of example.

[0050] See also Figure 2 As shown in FIG. 1 , the preset number, i.e., the size of the sliding window mw=12, and the preset step size is 1. The remote sensing data corresponding to the first 12 data points in the processed remote sensing time series data can be first input into the land cover type prediction model to obtain an initial land cover type prediction result sequence consisting of 12 land cover type prediction results. Then, the sliding window is slid on the processed remote sensing time series data to obtain a set of data points to be processed, which includes Figure 2 The 12 data points in the box of (a) in the figure are the 2nd to 12th data points in the processed remote sensing time series data, and the new data point 13. The land cover type prediction model can be used to process the remote sensing data corresponding to the new data point 13 and the land cover type prediction results corresponding to the data points 2-12, and obtain the land cover type prediction result of the data point 13.

[0051] After obtaining the land cover type prediction result of the new data point 13, the prediction result can be added to the initial land cover type prediction result sequence in chronological order to obtain an updated land cover type prediction result sequence. Determine whether the updated land cover type prediction result sequence includes the land cover type prediction result of the last data point in the processed remote sensing time series data. If it does, then use the updated land cover type prediction result sequence at this time as the land cover type prediction result sequence corresponding to the processed remote sensing time series data; if it does not, return to the step of sliding the sliding window on the processed remote sensing time series data to obtain a new set of data points to be processed. At this time, the set of data points to be processed includes Figure 2 The 12 data points in the box of (b) in the figure are the 3rd to 13th data points in the processed remote sensing time series data, and the new data point 14. The land cover type of the new data point 14 is predicted, and it is further determined whether the updated land cover type prediction result sequence includes the land cover type prediction result of the last data point. This is repeated until an updated land cover type prediction result sequence including the land cover type prediction result of the last data point in the processed remote sensing time series data is obtained.

[0052] This embodiment adopts the idea of ​​predicting the land cover type for the next data point through autoregressive iterative identification: the processed remote sensing time series data is processed in stages, first obtaining the initial land cover type prediction result sequence, and then using the sliding window to process the local data points. Combined with historical information, it can better capture the subtle patterns of land cover changes and greatly improve the land cover classification accuracy of a single data point. In addition, this solution avoids unnecessary data traversal, thereby significantly reducing the computational cost of the model and improving the prediction efficiency of land cover types, thereby achieving near real-time monitoring of long-term land cover change types.

[0053] In a possible implementation of the present disclosure, the land cover type prediction model includes an input embedding layer module, an encoder, an output embedding layer module and a decoder; using the land cover type prediction model to process the remote sensing data corresponding to the new data point in the set of data points to be processed and the target land cover type in the initial land cover type prediction result sequence to obtain the land cover type prediction result of the new data point, including:

[0054] The input embedding layer module is used to encode the remote sensing data corresponding to the new data point to obtain an initial feature vector of the remote sensing data; the output embedding layer module is used to encode each target land cover type to obtain a land cover type feature vector; the initial feature vector of the remote sensing data is input into the encoder to obtain a remote sensing data deep feature vector; the remote sensing data deep feature vector and the land cover type feature vector are input into the decoder to obtain a land cover type prediction result of the new data point.

[0055] Figure 3 The Transformer model architecture diagram provided by the embodiment of the present disclosure is shown. Figure 3 As shown in the figure, the Transformer model mainly includes an encoder, a decoder, an input embedding layer module and an output embedding layer module. The encoder is composed of Nx identical layers stacked together, and each layer includes two key parts: a multi-head attention (layer) and a feedforward layer. The decoder is composed of Nx identical layers stacked together, and each layer includes three key parts: a masked multi-head attention (layer), a multi-head attention (layer) and a feedforward layer. For the input embedding layer module and the output embedding layer module, for example, a token embedding layer and a positional embedding layer may be included, which are respectively used to convert the input raw data into a vector representation, and to generate a position vector for each position of the input raw data, so that the model can understand the position information in the sequence. The following is combined with Figure 3 This embodiment is described as an example.

[0056] In this embodiment, the input embedding layer module can be used to embed the remote sensing data corresponding to the new data point, that is, Figure 3 The remote sensing time series data in the encoder is encoded to obtain the initial feature vector of the remote sensing data; then the initial feature vector of the remote sensing data is input into the encoder. As one of the key mechanisms of the Transformer model, the multi-head self-attention mechanism contained in the encoder structure can capture the complex relationship between the data at each time point in the remote sensing time series data through multiple self-attention heads, thereby improving the modeling effect of the model on the time series. This self-attention mechanism also allows the model to more flexibly focus on the moments that have a greater impact on the current land cover status when processing uneven time series data. The output of each self-attention head is fused, and the fusion result is subjected to residual connection processing and normalization processing. Finally, the temporal feature representation extracted by the original input sequence through sequence modeling is output, that is, the deep feature vector of the remote sensing data, which is used in the land cover type prediction task of the subsequent decoder.

[0057] The output embedding layer module can be used to embed each target land cover type, namely Figure 3The predicted land cover type sequence in is encoded to obtain the land cover type feature vector; then, the remote sensing data depth feature vector and the land cover type feature vector are input into the decoder together to obtain the decoder output, that is, the land cover type prediction result of the new data point.

[0058] In actual application scenarios, factors such as cloud cover and strips may cause uneven intervals of processed remote sensing time series data, and irregular time series data will make it difficult for the model to capture stable time dependencies, thereby reducing the prediction accuracy of the model. Based on this, in a possible implementation of the present disclosure, the input embedding layer module includes a first token embedding layer, a first position embedding layer, and a first time embedding layer; the input embedding layer module is used to encode the remote sensing data corresponding to the new data point to obtain the initial feature vector of the remote sensing data, including:

[0059] Using the first token embedding layer, the remote sensing data corresponding to the new data point is encoded to obtain a first semantic feature vector; using the first position embedding layer, the position information corresponding to the new data point is encoded to obtain a first position feature vector; using the first time embedding layer, the time information corresponding to the new data point is encoded to obtain a first time feature vector; the first semantic feature vector, the first position feature vector, and the first time feature vector are fused to obtain the initial feature vector of the remote sensing data; the output embedding layer module includes a second token embedding layer, a second position embedding layer, and a second time embedding layer; using the output embedding layer module to encode each of the target land cover types to obtain a land cover type feature vector, including: using the second token embedding layer to encode each of the target land cover types to obtain a second semantic feature vector; using the second position embedding layer to encode the position information corresponding to each of the target land cover types to obtain a second position feature vector; using the second time embedding layer to encode the time information corresponding to each of the target land cover types to obtain a second time feature vector; fusing the second semantic feature vector, the second position feature vector, and the second time feature vector to obtain the land cover type feature vector.

[0060] In implementation, for the first token embedding layer or the second token embedding layer, for example, a one-dimensional convolution method can be used to map the input sequence to a dimension of Specifically, the projection is performed using the following formula:

[0061]

[0062] in, represents the semantic feature vector, represents a one-dimensional convolution operation, Represents the object to be encoded, that is, the remote sensing data corresponding to the new data point, or the target land cover type. Indicates that the dimension of the object to be encoded is C, The dimension of the semantic feature vector is .

[0063] For the first position embedding layer or the second position embedding layer, for example, the position encoding value of each element in the object to be encoded can be calculated using the following formulas:

[0064]

[0065]

[0066] Among them, i represents the dimension index of the vector, represents the position of an element in the input sequence, Represents the positional encoding value of the element at position pos in the 2ith dimension. Represents the positional encoding value of the element at position pos in the 2i+1th dimension.

[0067] For the first time embedding layer or the second time embedding layer, for example, each time encoding value in the time feature vector can be calculated according to the following formula:

[0068]

[0069] in, Represents the time encoding value of the element at position pos in the i-th dimension, represents the temporal embedding layer, Indicates the day of the current year on which the date corresponding to a data point in the object to be encoded falls. This can be obtained by encoding and adding the month and day information corresponding to each data point.

[0070] Finally, the semantic feature vector, position feature vector, and time feature vector are fused. For example, the corresponding elements of the semantic feature vector, position feature vector, and time feature vector can be added or weighted to obtain the initial feature vector of remote sensing data or the land cover type feature vector.

[0071] This embodiment uses three embedding layers to encode the inputs of the encoder and decoder. Among them, the time embedding layer can capture the time information in the sequence, ensure that the model can understand the time order and intervals of the elements in the input sequence, improve the model's perception of uneven time points, and thus significantly improve the model's prediction accuracy.

[0072] In actual application scenarios, predicting the change time of land cover types also has a profound impact on policy making, environmental protection, disaster warning, agricultural sustainable development and scientific research. Based on this, in a possible implementation of the present disclosure, the method also includes:

[0073] At least one target land cover type pair is selected from the land cover type prediction result sequence; the target land cover type pair includes two adjacent but different land cover types; and based on the land cover type prediction result sequence, the land cover changes of the area to be monitored are monitored, including: monitoring the land cover changes of the area to be monitored according to the time information corresponding to the previous land cover type in the at least one target land cover type pair and the land cover type prediction result sequence.

[0074] Exemplarily, suppose the sequence of land cover type prediction results is: forest, forest, forest, forest, bare land, bare land, bare land, cultivated land, cultivated land. Then the target land cover type pairs can be screened out: (forest, bare land) and (bare land, cultivated land). If the time information corresponding to the 4th land cover type prediction result is time 1, and the time information corresponding to the 7th land cover type prediction result is time 2, then the land cover changes in the monitored area can be monitored based on the land cover type prediction result sequence (forest, forest, forest, forest, bare land, bare land, bare land, cultivated land, cultivated land), and the time 1 and time 2 corresponding to the 4th and 7th land cover type prediction results in the sequence, respectively.

[0075] Based on the same concept, the embodiment of the present disclosure also proposes a model training method for training the land cover type prediction model in any of the aforementioned solutions. Figure 4 A schematic diagram of the flow chart of the model training method provided by the embodiment of the present disclosure is shown, from Figure 4 It can be seen that the embodiment of the present disclosure at least includes steps S401-S402:

[0076] S401: Acquire a data set; the samples in the data set include sample remote sensing time series data of various types of regions, and the labels of the samples are corresponding real land cover type sequences.

[0077] S402: Using the data set to train an initial land cover type prediction model with time series data analysis capability to obtain a land cover type prediction model.

[0078] In this embodiment, the sample remote sensing time series data may include at least one of the following items: reflectance matrix of at least one band, brightness matrix of at least one band, temperature matrix of at least one band, vegetation index of at least one band, etc. The types of regions include, but are not limited to, farmland, city, forest, water area, desert, etc. Exemplarily, a sample in the data set is a sequence: reflectance matrix A1 of band A and reflectance matrix B1 of band B (corresponding to observation point 1), reflectance matrix A2 of band A and reflectance matrix B2 of band B (corresponding to observation point 2), reflectance matrix A3 of band A and reflectance matrix B3 of band B (corresponding to observation point 3), and the label corresponding to the sample is a sequence: farmland, farmland, city.

[0079] In a possible implementation, the initial land cover type prediction model is constructed based on a Transformer model, an LSTM model or a TCN model.

[0080] During specific training, the sample remote sensing time series data in the data set can be preprocessed, including data cleaning, normalization and other operations, to improve data quality and model training effect. Then, the preprocessed time series data is segmented, and the long time series data is divided into multiple shorter time series data so that the model can learn and process better. The segmented remote sensing time series data is input into the initial land cover type prediction model. During the training process, a suitable loss function such as the cross entropy loss function can be used to calculate the difference between the land cover type sequence predicted by the model and the real land cover type sequence. Optimization algorithms such as stochastic gradient descent are used to update parameters. During the training process, the data set is divided into a training set, a validation set and a test set according to a certain ratio. After each training cycle, the validation set is used to evaluate the model, and the training parameters such as the learning rate and the dropout probability are adjusted according to the evaluation results. After multiple iterations of training, when the performance of the model on the validation set reaches the preset standard or no longer improves significantly, the training ends and the final land cover type prediction model is obtained.

[0081] The present disclosure also provides a land cover change monitoring method. Figure 5 The flow chart of the land cover change monitoring method provided by the embodiment of the present disclosure is shown below. Figure 5 The embodiment of the present disclosure is described. The embodiment of the present disclosure includes the following steps:

[0082] Step S1: Obtain a data set; the samples in the data set include sample remote sensing time series data of various types of regions, and the labels of the samples are the corresponding real land cover type sequences.

[0083] Step S2: Use the data set to train the initial land cover type prediction model obtained based on the Transformer model to obtain a land cover type prediction model. The land cover type prediction model includes an input embedding layer module, an encoder, an output embedding layer module and a decoder. Among them, the input embedding layer module includes a first token embedding layer, a first position embedding layer and a first time embedding layer, and the output embedding layer module includes a second token embedding layer, a second position embedding layer and a second time embedding layer.

[0084] Step S3: Obtaining original remote sensing time series data of the area to be monitored, and performing vectorization preprocessing on the original remote sensing time series data to obtain processed remote sensing time series data. In this embodiment, the land cover types of the area to be monitored include forest land, cultivated land and impervious surface.

[0085] Step S4: Using the land cover type prediction model, the remote sensing data corresponding to the first 12 data points in the processed remote sensing time series data are processed to obtain an initial land cover type prediction result sequence.

[0086] See also Figure 5 In Figure (a), in this step, the first token embedding layer can be used to encode the first 12 data points to obtain the semantic feature vector, that is, Figure 5 The token sequence in . Then, the first position embedding layer is used to positionally encode the position information of each element in the token sequence to obtain the corresponding position feature vector. The first time embedding layer is used to temporally encode the time information of each element in the token sequence to obtain the corresponding time feature vector. The tokens (B and E) of begin of sequence (BOS) and end of sequence (EOS) can also be spliced ​​at the head and tail of the token sequence, respectively, to facilitate the control of the output length of the model in tasks where the input and output are sequences of unequal lengths. The three feature vectors are added together to obtain a fusion vector, which is input into the decoder. The decoder receives the BOS token representing the beginning of the sequence, processes its own hidden state and the fusion vector from the encoder, and obtains the decoder output. The decoder output passes through the linear layer and the softmax layer to finally obtain the initial land cover type prediction result sequence, that is, Figure 5 Land cover sequence in .

[0087] Step S5: Let the sliding window with a size of 12 slide backward by 1 data point on the processed remote sensing time series data in chronological order to obtain a set of data points to be processed. Figure 5 In Figure (b), the Transformer model and the window sliding mechanism are used together as Transformer change detectors to perform sliding monitoring on the processed remote sensing time series data.

[0088] Step S6: Use the first token embedding layer to encode the remote sensing data corresponding to the new data point to obtain a first semantic feature vector; use the first position embedding layer to encode the position information corresponding to the new data point to obtain a first position feature vector; use the first time embedding layer to encode the time information corresponding to the new data point to obtain a first time feature vector.

[0089] Step S7: Fusing the first semantic feature vector, the first position feature vector, and the first time feature vector to obtain an initial feature vector of remote sensing data.

[0090] Step S8: Encode each target land cover type using the second token embedding layer to obtain a second semantic feature vector; Encode the position information corresponding to each target land cover type using the second position embedding layer to obtain a second position feature vector; Encode the time information corresponding to each target land cover type using the second time embedding layer to obtain a second time feature vector. Here, the target land cover type is the land cover type prediction result of the data points in the set of data points to be processed except the new data points.

[0091] Step S9: Fusing the second semantic feature vector, the second position feature vector, and the second time feature vector to obtain a land cover type feature vector.

[0092] Step S10: Inputting the initial feature vector of remote sensing data into the encoder to obtain the deep feature vector of remote sensing data.

[0093] Step S11: input the remote sensing data depth feature vector and the land cover type feature vector into the decoder to obtain the land cover type prediction result of the new data point.

[0094] Step S12: Based on the land cover type prediction results of the new data points, an updated land cover type prediction result sequence is obtained.

[0095] Step S13: Determine whether the updated land cover type prediction result sequence includes the land cover type prediction result of the last data point in the processed remote sensing time series data.

[0096] Step S14: if the updated land cover type prediction result sequence includes the land cover type prediction result of the last data point in the processed remote sensing time series data, the updated land cover type prediction result sequence is used as the land cover type prediction result sequence;

[0097] Step S15: If the updated land cover type prediction result sequence does not include the land cover type prediction result of the last data point in the processed remote sensing time series data, return to step S5.

[0098] Step S16: selecting at least one target land cover type pair from the land cover type prediction result sequence; the target land cover type pair includes two adjacent but different land cover types.

[0099] Step S17: monitoring the land cover changes in the monitored area according to the time information corresponding to the previous land cover type in at least one target land cover type pair and the land cover type prediction result sequence.

[0100] Those skilled in the art will appreciate that, in the above method of the specific implementation manner, the writing order of the steps does not mean a strict execution order, but does not constitute any limitation on the implementation process. The specific execution order of the steps should be determined by their functions and possible internal logic.

[0101] It should be noted that, in practical applications, all possible implementations described above can be combined in any manner to form possible embodiments of the present disclosure, which will not be described one by one here.

[0102] Based on the same concept, the embodiment of the present disclosure also provides a land cover change monitoring device. Figure 6 The structure diagram of the land cover change monitoring device provided by the embodiment of the present disclosure is shown in FIG. Figure 6 As shown, the land cover change monitoring device 600 provided in the embodiment of the present disclosure includes:

[0103] The acquisition module 601 is used to acquire the original remote sensing time series data of the area to be monitored, and perform vectorization preprocessing on the original remote sensing time series data to obtain processed remote sensing time series data.

[0104] The prediction module 602 is used to process the processed remote sensing time series data based on a pre-built land cover type prediction model to obtain a land cover type prediction result sequence; the land cover type prediction model is used to predict the land cover types of various types of regions.

[0105] The monitoring module 603 is used to monitor the land cover change of the area to be monitored based on the land cover type prediction result sequence.

[0106] In a possible implementation, the device also includes a training module, which is used to: obtain a data set; the samples in the data set include sample remote sensing time series data of various types of regions, and the labels of the samples are corresponding real land cover type sequences; use the data set to train an initial land cover type prediction model with time series data analysis capabilities to obtain the land cover type prediction model.

[0107] In a possible implementation manner, in the above-mentioned device, the prediction module is used to: use the land cover type prediction model to process the remote sensing data corresponding to the first preset number of data points in the processed remote sensing time series data to obtain an initial land cover type prediction result sequence; let the sliding window with the preset number as the size slide backward on the processed remote sensing time series data in chronological order and a preset sliding step length to obtain a set of data points to be processed; use the land cover type prediction model to process the remote sensing data corresponding to the new data point in the set of data points to be processed and the target land cover type in the initial land cover type prediction result sequence to obtain the land cover type prediction result of the new data point result; the target land cover type is the land cover type prediction result of the data points in the set of data points to be processed except the new data point; based on the land cover type prediction result of the new data point, an updated land cover type prediction result sequence is obtained; if the updated land cover type prediction result sequence includes the land cover type prediction result of the last data point in the processed remote sensing time series data, the updated land cover type prediction result sequence is used as the land cover type prediction result sequence; otherwise, return to the step of sliding the sliding window with the preset number as the size backward on the processed remote sensing time series data in chronological order and with the preset sliding step size to obtain the set of data points to be processed.

[0108] In a possible implementation, in the above-mentioned device, the land cover type prediction model includes an input embedding layer module, an encoder, an output embedding layer module and a decoder; the prediction module is used to: use the input embedding layer module to encode the remote sensing data corresponding to the new data point to obtain an initial feature vector of the remote sensing data; use the output embedding layer module to encode each of the target land cover types to obtain a land cover type feature vector; input the initial feature vector of the remote sensing data into the encoder to obtain a deep feature vector of the remote sensing data; input the deep feature vector of the remote sensing data and the land cover type feature vector into the decoder to obtain a land cover type prediction result of the new data point.

[0109] In a possible implementation, in the above-mentioned device, the input embedding layer module includes a first token embedding layer, a first position embedding layer and a first time embedding layer; the prediction module is used to: use the first token embedding layer to encode the remote sensing data corresponding to the new data point to obtain a first semantic feature vector; use the first position embedding layer to encode the position information corresponding to the new data point to obtain a first position feature vector; use the first time embedding layer to encode the time information corresponding to the new data point to obtain a first time feature vector; fuse the first semantic feature vector, the first position feature vector and the first time feature vector to obtain the initial remote sensing data. feature vector; the output embedding layer module includes a second token embedding layer, a second position embedding layer and a second time embedding layer; the prediction module is used to: use the second token embedding layer to encode each of the target land cover types to obtain a second semantic feature vector; use the second position embedding layer to encode the position information corresponding to each of the target land cover types to obtain a second position feature vector; use the second time embedding layer to encode the time information corresponding to each of the target land cover types to obtain a second time feature vector; fuse the second semantic feature vector, the second position feature vector and the second time feature vector to obtain the land cover type feature vector.

[0110] In a possible implementation, the device further includes a time prediction module, which is used to: screen out at least one target land cover type pair from the land cover type prediction result sequence; the target land cover type pair includes two adjacent but different land cover types; and monitoring the land cover changes in the area to be monitored based on the land cover type prediction result sequence includes: monitoring the land cover changes in the area to be monitored based on the time information corresponding to the previous land cover type in the at least one target land cover type pair and the land cover type prediction result sequence.

[0111] It should be noted that any of the above-mentioned land cover change monitoring devices can implement the above-mentioned land cover change monitoring method one by one, which will not be repeated here.

[0112] Based on the same concept, the embodiment of the present disclosure also provides a model training device for training the land cover type prediction model in any of the above solutions. Figure 7 A schematic diagram of the structure of the model training device provided by the embodiment of the present disclosure is shown. Figure 7 As shown, the model training device 700 provided in the embodiment of the present disclosure includes:

[0113] The acquisition module 701 is used to acquire a data set; the samples in the data set include sample remote sensing time series data of various types of regions, and the labels of the samples are corresponding real land cover type sequences.

[0114] The training module 702 is used to train an initial land cover type prediction model with time series data analysis capability using the data set to obtain a land cover type prediction model.

[0115] In a possible implementation, in the above-mentioned device, the initial land cover type prediction model is constructed based on a Transformer model, a LSTM model or a TCN model.

[0116] It should be noted that any of the above-mentioned model training devices can implement the aforementioned model training method one-to-one, which will not be repeated here.

[0117] Figure 8 FIG. 1 shows a schematic diagram of the structure of an electronic device provided by an embodiment of the present disclosure. Figure 8 As shown, at the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. The memory may include a memory, such as a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage, etc. Of course, the electronic device may also include hardware required for other services.

[0118] The processor, network interface and memory can be interconnected through an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0119] The memory is used to store programs. Specifically, the program may include program code, and the program code includes computer operation instructions. The memory may include internal memory and non-volatile memory, and provides instructions and data to the processor.

[0120] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a land cover change monitoring device or a model training device at the logical level. The processor executes the program stored in the memory and is specifically used to execute the above method.

[0121] The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present disclosure may be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present disclosure may be directly embodied as being executed by a hardware decoding processor, or may be executed by a combination of hardware and software modules in a decoding processor. The software module may be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0122] The electronic device can execute the land cover change monitoring method provided by the multiple embodiments of the present disclosure, and realize the land cover change monitoring device in Figure 6 The functions of the embodiments shown in the figure, or the execution of the model training method, and the implementation of the model training device in Figure 7 The functions of the illustrated embodiments will not be described in detail in the embodiments of the present disclosure.

[0123] The embodiments of the present disclosure also propose a computer-readable storage medium, which stores one or more programs, and the one or more programs include instructions. When the instructions are executed by an electronic device including multiple application programs, the electronic device can execute the land cover change monitoring method or model training method provided by multiple embodiments of the present disclosure.

[0124] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0125] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0126] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0127] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0128] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0129] Memory may include non-permanent storage in a computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0130] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape, disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0131] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity or device including the element.

[0132] It should be understood by those skilled in the art that the embodiments of the present disclosure may be provided as methods, systems or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0133] The above are only embodiments of the present disclosure and are not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the scope of the claims of the present disclosure.

Claims

1. A land cover change monitoring method, characterized in that: The method comprises: Acquire original remote sensing time series data of the area to be monitored, and perform vectorization preprocessing on the original remote sensing time series data to obtain processed remote sensing time series data; Based on a pre-built land cover type prediction model, the processed remote sensing time series data is processed to obtain a land cover type prediction result sequence; the land cover type prediction model is used to predict the land cover types of various types of regions; Based on the land cover type prediction result sequence, monitoring the land cover change of the area to be monitored; The pre-built land cover type prediction model is used to process the processed remote sensing time series data to obtain a land cover type prediction result sequence, including: Using the land cover type prediction model, the remote sensing data corresponding to the first preset number of data points in the processed remote sensing time series data are processed to obtain an initial land cover type prediction result sequence; Let the sliding window with the preset number as the size slide backward on the processed remote sensing time series data in chronological order and with a preset sliding step length to obtain a set of data points to be processed; Using the land cover type prediction model, the remote sensing data corresponding to the new data point in the set of data points to be processed and the target land cover type in the initial land cover type prediction result sequence are processed to obtain the land cover type prediction result of the new data point; the target land cover type is the land cover type prediction result of the data points in the set of data points to be processed except the new data point; Based on the land cover type prediction results of the new data points, an updated land cover type prediction result sequence is obtained; If the updated land cover type prediction result sequence includes the land cover type prediction result of the last data point in the processed remote sensing time series data, the updated land cover type prediction result sequence is used as the land cover type prediction result sequence; otherwise, return to the step of sliding the sliding window with the preset number as the size backward on the processed remote sensing time series data in chronological order and with the preset sliding step size to obtain a set of data points to be processed.

2. The method according to claim 1, characterized in that The pre-built land cover type prediction model is trained according to the following method: Acquire a data set; the samples in the data set include sample remote sensing time series data of various types of regions, and the labels of the samples are corresponding real land cover type sequences; The data set is used to train an initial land cover type prediction model with time series data analysis capability to obtain the land cover type prediction model.

3. The method according to claim 1, characterized in that The land cover type prediction model comprises an input embedding layer module, an encoder, an output embedding layer module and a decoder; the land cover type prediction model is used to process the remote sensing data corresponding to the new data point in the set of data points to be processed and the target land cover type in the initial land cover type prediction result sequence to obtain the land cover type prediction result of the new data point, including: Using the input embedding layer module, encoding the remote sensing data corresponding to the new data point to obtain an initial feature vector of the remote sensing data; Using the output embedding layer module, encoding each of the target land cover types to obtain a land cover type feature vector; Inputting the initial feature vector of the remote sensing data into the encoder to obtain a deep feature vector of the remote sensing data; The remote sensing data depth feature vector and the land cover type feature vector are input into the decoder to obtain the land cover type prediction result of the new data point.

4. The method according to claim 3, characterized in that The input embedding layer module includes a first token embedding layer, a first position embedding layer and a first time embedding layer; the remote sensing data corresponding to the new data point is encoded by using the input embedding layer module to obtain an initial feature vector of the remote sensing data, including: Using the first token embedding layer, the remote sensing data corresponding to the new data point is encoded to obtain a first semantic feature vector; using the first position embedding layer, the position information corresponding to the new data point is encoded to obtain a first position feature vector; using the first time embedding layer, the time information corresponding to the new data point is encoded to obtain a first time feature vector; fusing the first semantic feature vector, the first position feature vector, and the first time feature vector to obtain an initial feature vector of the remote sensing data; The output embedding layer module includes a second token embedding layer, a second position embedding layer and a second time embedding layer; the output embedding layer module is used to encode each of the target land cover types to obtain a land cover type feature vector, including: Using the second token embedding layer, each of the target land cover types is encoded to obtain a second semantic feature vector; using the second position embedding layer, the position information corresponding to each of the target land cover types is encoded to obtain a second position feature vector; using the second time embedding layer, the time information corresponding to each of the target land cover types is encoded to obtain a second time feature vector; The second semantic feature vector, the second position feature vector, and the second time feature vector are fused to obtain the land cover type feature vector.

5. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: Selecting at least one target land cover type pair from the land cover type prediction result sequence; the target land cover type pair includes two adjacent but different land cover types; The monitoring of the land cover change of the area to be monitored based on the land cover type prediction result sequence includes: The land cover changes of the area to be monitored are monitored according to the time information corresponding to the previous land cover type in the at least one target land cover type pair and the land cover type prediction result sequence.

6. A model training method, characterized in that: For training the land cover type prediction model according to any one of claims 1 to 5, the method comprises: Acquire a data set; the samples in the data set include sample remote sensing time series data of various types of regions, and the labels of the samples are corresponding real land cover type sequences; The data set is used to train an initial land cover type prediction model with time series data analysis capability to obtain the land cover type prediction model.

7. The method according to claim 6, characterized in that The initial land cover type prediction model is constructed based on a Transformer model, an LSTM model or a TCN model.

8. A land cover change monitoring device, characterized in that: The device comprises: An acquisition module is used to acquire original remote sensing time series data of the area to be monitored, and perform vectorization preprocessing on the original remote sensing time series data to obtain processed remote sensing time series data; A prediction module is used to process the processed remote sensing time series data based on a pre-built land cover type prediction model to obtain a land cover type prediction result sequence; the land cover type prediction model is used to predict the land cover types of various types of regions; A monitoring module, used for monitoring the land cover change of the area to be monitored based on the land cover type prediction result sequence; The prediction module is specifically used to: use the land cover type prediction model to process the remote sensing data corresponding to the first preset number of data points in the processed remote sensing time series data to obtain an initial land cover type prediction result sequence; let the sliding window with the preset number as the size slide backward on the processed remote sensing time series data in chronological order and with a preset sliding step length to obtain a set of data points to be processed; use the land cover type prediction model to process the remote sensing data corresponding to the new data point in the set of data points to be processed and the target land cover type in the initial land cover type prediction result sequence to obtain the land cover type prediction result of the new data point; the target land The coverage type is the land cover type prediction result of the data points other than the new data point in the set of data points to be processed; based on the land cover type prediction result of the new data point, an updated land cover type prediction result sequence is obtained; if the updated land cover type prediction result sequence includes the land cover type prediction result of the last data point in the processed remote sensing time series data, the updated land cover type prediction result sequence is used as the land cover type prediction result sequence; otherwise, return to the step of sliding the sliding window with the preset number as the size, sliding backward on the processed remote sensing time series data in chronological order and with a preset sliding step size to obtain a set of data points to be processed.

9. A model training device, characterized in that: Used for training the land cover type prediction model according to any one of claims 1 to 5, the device comprises: An acquisition module is used to acquire a data set; the samples in the data set include sample remote sensing time series data of various types of regions, and the labels of the samples are corresponding real land cover type sequences; The training module is used to train an initial land cover type prediction model with time series data analysis capability using the data set to obtain a land cover type prediction model.

10. An electronic device comprising: processor; as well as A memory arranged to store computer executable instructions, characterized in that when the executable instructions are executed, the processor performs the method according to any one of claims 1 to 5, or the steps of the method according to any one of claims 6 to 7.

11. A computer-readable storage medium storing one or more programs, characterized in that: When the one or more programs are executed by an electronic device including a plurality of application programs, the electronic device executes the method as claimed in any one of claims 1 to 5, or the steps of the method as claimed in any one of claims 6 to 7.

Images