A remote sensing time series image generation method, device and medium

Abstract

The present application discloses a remote sensing time-series image generation method, device and medium, which relates to the field of image video analysis and agricultural information technology. The method comprises: determining remote sensing time-series images of various types of crops in the area to be predicted; dividing the crop distribution map of the current crop at each historical moment into blocks of spatial angles to obtain multiple historical block images; inputting the historical block images of all historical moments of any spatial angle of the current crop into a remote sensing image generation model to obtain the predicted block images of the corresponding spatial angle at the time to be predicted; the remote sensing image generation model is obtained by training the initial model using a training set, and the initial model is constructed based on a Transformer network and a VQ-VAE model; based on the predicted block images of the various types of crops at each spatial angle at the time to be predicted in the area to be predicted, the predicted remote sensing image at the time to be predicted is determined. The prediction accuracy of high-resolution remote sensing images is improved.

Description

Technical Field

[0001] The present application relates to the field of image and video analysis and agricultural information technology, and in particular to a remote sensing time series image generation method, device and medium. Background Art

[0002] Satellite remote sensing images, as a special but important form of multimedia data, are an important data source for quickly obtaining surface information under large-scale continuous monitoring in the agricultural field. With the launch and commissioning of a large number of satellites equipped with high-resolution image acquisition equipment, a large number of high-resolution remote sensing images have been generated. In agricultural production, high-resolution satellite remote sensing images can provide detailed farmland images and accurately extract crop planting areas and location distribution. Large-scale crop growth monitoring and accurate yield prediction must rely on the effective acquisition of high-resolution satellite remote sensing data and meteorological data, such as light, temperature, and precipitation, which are the main factors affecting crop growth. However, due to the long satellite revisit period, as well as the influence of factors such as continuous cloud and rain cover and extremely severe weather, it is impossible to effectively obtain spatiotemporal continuous high-resolution remote sensing images throughout the life cycle of crop growth, making it difficult to meet the needs of agricultural production practice for real-time and accurate monitoring of crop growth.

[0003] Existing work on high-resolution remote sensing time series image generation has made some progress, but there is still a large error between the predicted image and the actual image, which is mainly reflected in the time change between the reference image and the predicted image, as well as the coarse and fine resolution image scales. In particular, continuous cloud and rain cover, extremely bad weather and strong time change conditions will affect the quality of the generated image. This means that there are still some problems and challenges that need to be solved in the prediction and generation of high-resolution remote sensing time series images. Summary of the invention

[0004] The purpose of this application is to provide a remote sensing time series image generation method, device and medium, which improves the prediction accuracy of high-resolution remote sensing images.

[0005] To achieve the above objectives, this application provides the following solutions.

[0006] In a first aspect, the present application provides a remote sensing time series image generation method, comprising:

[0007] Determine the remote sensing time series images of various types of crops in the area to be predicted; the remote sensing time series images include: crop distribution maps at multiple consecutive historical moments;

[0008] Identify any type of crop as the current crop;

[0009] The crop distribution map of the current crop at each historical moment is divided into blocks at a spatial angle to obtain multiple historical block images corresponding to the historical moment;

[0010] Input the historical block images of all historical moments of any spatial angle of the current crop into the remote sensing image generation model to obtain the predicted block images of the corresponding spatial angle at the time to be predicted; the remote sensing image generation model is obtained by training the initial model using the training set, and the initial model is constructed based on the Transformer network and the VQ-VAE model;

[0011] Based on the predicted block images at the time to be predicted of each type of crop in the area to be predicted at each spatial angle, the predicted remote sensing image of the area to be predicted at the time to be predicted is determined.

[0012] Optionally, the training process of the remote sensing image generation model includes:

[0013] Acquire the training set; the training set includes multiple groups of training data, and the training data includes: training block images of the first N historical moments and the corresponding training block images of the N+1th historical moment;

[0014] Constructing the initial model based on the Transformer network and the VQ-VAE model;

[0015] The initial model is trained using the training set to obtain the remote sensing image generation model.

[0016] Optionally, the initial model includes: an encoder of a VQ-VAE model, a latent vector flattening module, a meteorological factor information display embedding module, an encoder of a Transformer network and a decoder of a VQ-VAE model.

[0017] Optionally, using the training set to train the initial model to obtain the remote sensing image generation model includes:

[0018] The initial model is trained based on the training set using a total loss function to obtain the remote sensing image generation model; the total loss function includes:

[0019] ;

[0020] in, is the restoration loss of the predicted latent vector; The predicted latent vector is decoded by the decoder of the VQ-VAE model to obtain the similarity measurement loss in the image dimension between the predicted block image and the training block image.

[0021] Optionally, all historical block images of any spatial angle of the current crop at all historical moments are input into the remote sensing image generation model to obtain the predicted block image at the time to be predicted at the corresponding spatial angle, including:

[0022] Determine any spatial angle of the current crop as the current angle;

[0023] The historical block images of all historical moments of the current angle are input into the encoder of the VQ-VAE model in the remote sensing image generation model to obtain a latent vector of size 16×16×4 corresponding to the historical block image;

[0024] The latent vector flattening module in the remote sensing image generation model is used to flatten each latent vector, and the meteorological factor information display embedding module is used to couple the meteorological factor to the latent vector to obtain a Token vector of size 1024 corresponding to the historical block image.

[0025] Based on all the token vectors of the current angle and the encoder of the Transformer network, determine the crop-specific block image symbol of the current angle;

[0026] The current angle is input into the decoder of the VQ-VAE model in the remote sensing image generation model to obtain the predicted block image of the current angle at the time to be predicted.

[0027] Optionally, determining remote sensing time series images of various types of crops in the area to be predicted includes:

[0028] Acquire historical time-series satellite remote sensing data of the area to be predicted; the historical time-series satellite remote sensing data includes: satellite remote sensing data of multiple consecutive historical moments;

[0029] The remote sensing time series images of various types of crops in the area to be predicted are determined according to the historical time series satellite remote sensing data.

[0030] Optionally, determining remote sensing time series images of various types of crops in the area to be predicted based on the historical time series satellite remote sensing data includes:

[0031] Obtain the correlation between the spectral reflectance and leaf area index images of various types of crops;

[0032] Based on the satellite remote sensing data at each historical moment and the association relationship, determining an initial leaf area index data set corresponding to the historical moment;

[0033] Based on the leaf area index images in the initial leaf area index data set at each historical moment, the crop distribution map of each type of crop at the corresponding historical moment is determined, and the remote sensing time series images of each type of crop in the area to be predicted are further obtained.

[0034] In a second aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-described remote sensing time series image generation methods.

[0035] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-mentioned remote sensing time-series image generation methods.

[0036] According to the specific embodiments provided by this application, this application discloses the following technical effects: This application discloses a remote sensing time series image generation method, device and medium. In view of the low utilization rate of massive historical remote sensing images, the complex impact of dynamic changes in meteorological factors, and the continuous lack of large-area high-resolution remote sensing images in actual agricultural production, this application proposes a method for generating high-resolution remote sensing time series images of crops throughout their growth period based on the data characteristics and temporal evolution characteristics of high-resolution remote sensing images, and realizes real-time, low-cost high-resolution remote sensing monitoring of crop growth and yield information on a large regional scale, providing key technical support for accurate perception and prediction of agricultural crop growth. The remote sensing image generation model of this application is constructed based on the Transformer network and the VQ-VAE model, which improves the prediction accuracy of high-resolution remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 A schematic flow chart of a remote sensing time series image generation method provided in one embodiment of the present application.

[0039] Figure 2 Schematic diagram of the architecture of the high-resolution remote sensing time-series image generation method.

[0040] Figure 3 Schematic diagram of crop type identification and planting distribution extraction at a plot scale.

[0041] Figure 4 Schematic diagram of local area compression encoding of high-resolution remote sensing images by the VQ-VAE model.

[0042] Figure 5 Schematic diagram of Transformer encoder Token training.

[0043] Figure 6 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0045] The purpose of this application is to provide a method for generating high-resolution remote sensing time-series images. In order to solve the problems of low utilization rate of massive historical remote sensing images, complex impact of dynamic changes in meteorological factors, and continuous lack of high-resolution remote sensing images over large areas (county level) in actual agricultural production, this application proposes a method for generating high-resolution remote sensing time-series images of crops throughout their entire growth period based on the data characteristics and temporal evolution characteristics of high-resolution remote sensing images. This method can realize real-time, low-cost high-resolution remote sensing monitoring of crop growth and yield information on a large regional scale, provide key technical support for accurate perception and prediction of agricultural crop growth, and provide strong support for agricultural management and food security.

[0046] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0047] In an exemplary embodiment, Figure 1 and Figure 2 As shown, the remote sensing time series image generation method in this embodiment includes the following steps.

[0048] Step 1: Determine the remote sensing time series images of each type of crops in the area to be predicted.

[0049] Among them, remote sensing time series images include: crop distribution maps at multiple consecutive historical moments.

[0050] As an optional implementation, step 1 includes steps 11 and 12.

[0051] Step 11: Acquire historical time series satellite remote sensing data of the area to be predicted; the historical time series satellite remote sensing data includes: satellite remote sensing data of multiple consecutive historical moments.

[0052] Step 12: Determine the remote sensing time series images of various types of crops in the area to be predicted based on the historical time series satellite remote sensing data.

[0053] As an optional implementation, step 12 includes steps 121 to 123.

[0054] Step 121: Obtain the correlation between the spectral reflectance and leaf area index image of each type of crop.

[0055] Step 122: Based on the satellite remote sensing data and the association relationship at each historical moment, determine the initial leaf area index data set corresponding to the historical moment.

[0056] Step 123: Based on the leaf area index images in the initial leaf area index data set at each historical moment, the crop distribution map of each type of crop at the corresponding historical moment is determined, and further the remote sensing time series image of each type of crop in the area to be predicted is obtained.

[0057] Specifically, step 1 can realize the extraction of high-resolution remote sensing leaf area index (LAI) images and crop planting distribution at the county level. The specific steps include: first, selecting crop samples. This application takes Nenjiang City, where the terrain is relatively flat and the plots are relatively regular, as an example, but it is not limited to one demonstration county and can be extended to the whole country. The corn and soybean planting areas in the demonstration county of Nenjiang City account for more than 90%, and the growing seasons of corn and soybeans are almost the same. Therefore, the crop types are determined to be corn and soybeans. Then, the Sentinel-2 satellite high-resolution remote sensing data is selected as the historical time series satellite remote sensing data, and based on the existing field LAI collection data, the remote sensing leaf area index image of Nenjiang City is inverted by establishing the quantitative relationship between the spectral reflectance of corn and LAI, and the quantitative relationship between the spectral reflectance of soybeans and LAI, thereby constructing a crop high-resolution remote sensing time series LAI image dataset with a time resolution of 5 days. Finally, the crop type recognition and planting distribution extraction methods at the plot scale are studied, such as Figure 3 As shown in the figure, the specific steps include: (1) Collecting complete high-resolution time series LAI images (i.e., median sequence (SAR / OPT) data) of remote sensing in the study area, and using multi-temporal processing to perform data encoding, noise removal, multi-temporal registration, geocoding, radiation correction and cropping on the data of each phase. (2) Extracting the crop planting area as a spatial constraint, using LSTM to calculate the short-term characteristics of crop LAI at the plot level. (3) Selecting the plot vector (Geo-parcel) for encoding, constructing the multi-dimensional attribute information at the crop plot level, and regularizing the unit layer to process the sequence values. (4) Based on the sample characteristics of the crop type to be classified, a time series model with different numbers of neurons and hierarchical structures is constructed, and the random forest algorithm is used to discriminate the crop type information of each plot, and then invert the crop distribution map of each crop type.

[0058] Step 2: Determine any type of crop as the current crop.

[0059] Step 3: Divide the crop distribution map of the current crop at each historical moment into blocks at a spatial angle to obtain multiple historical block images corresponding to the historical moments.

[0060] Specifically, the size of the history block image is 256×256.

[0061] Step 4: Input the historical block images of all historical moments of any spatial angle of the current crop into the remote sensing image generation model to obtain the predicted block image of the corresponding spatial angle at the time to be predicted.

[0062] Among them, the remote sensing image generation model is obtained by training the initial model using the training set, and the initial model is built based on the Transformer network and VQ-VAE model.

[0063] As an optional implementation, the training process of the remote sensing image generation model in step 4 includes steps 401 to 403.

[0064] Step 401: Obtain a training set; the training set includes multiple groups of training data, and the training data includes: training block images of the first N historical moments and the corresponding training block images of the N+1th historical moment.

[0065] Step 402: Build an initial model based on the Transformer network and the VQ-VAE model.

[0066] Step 403: Use the training set to train the initial model to obtain a remote sensing image generation model.

[0067] As an optional implementation, the initial model includes: an encoder of a VQ-VAE model, a latent vector flattening module, a meteorological factor information display embedding module, an encoder of a Transformer network, and a decoder of a VQ-VAE model.

[0068] As an optional implementation, step 403 includes:

[0069] The total loss function is used to train the initial model based on the training set to obtain the remote sensing image generation model; the total loss function includes:

[0070] .

[0071] in, is the restoration loss of the predicted latent vector; The predicted latent vector is decoded by the decoder of the VQ-VAE model to obtain the similarity measurement loss in the image dimension between the predicted block image and the training block image.

[0072] As an optional implementation, step 4 includes steps 411 to 415.

[0073] Step 411: Determine any spatial angle of the current crop as the current angle.

[0074] Step 412: Input the historical block images of all historical moments of the current angle into the encoder of the VQ-VAE model in the remote sensing image generation model to obtain a latent vector of size 16×16×4 corresponding to the historical block image.

[0075] Step 413: Use the latent vector flattening module in the remote sensing image generation model to flatten each latent vector, and use the meteorological factor information display embedding module to couple the meteorological factor to the latent vector to obtain a Token vector of size 1024 corresponding to the historical block image.

[0076] Step 414: Based on all Token vectors of the current angle and the encoder of the Transformer network, determine the crop-specific block image symbol of the current angle.

[0077] Step 415: Input the current angle into the decoder of the VQ-VAE model in the remote sensing image generation model to obtain the predicted block image of the current angle at the time to be predicted.

[0078] Step 5: Determine the predicted remote sensing image of the area to be predicted at the time to be predicted based on the predicted block images of each type of crop in the area to be predicted at each spatial angle at the time to be predicted.

[0079] Specifically, the crop LAI images from October 2017 to December 2023 from Sentinel-2 satellite remote sensing are used as the data source, and a database of high-resolution remote sensing time series and LAI corresponding meteorological factors for Nenjiang City is constructed for a period of 6 years. Due to the extremely high resolution and huge amount of information of remote sensing images, the commonly used image coding methods are generally only learned at low resolution, and it is difficult to fully explore the intrinsic correlation between complex meteorological change factors and crop growth evolution. Therefore, a special local spatiotemporal coding method for remote sensing images with explicit embedding of coupled meteorological factor matching information is designed, while considering reducing the computational overhead and reducing the information loss caused by image compression coding. Combined with the key information of meteorological factors that affect the prediction of high-resolution remote sensing images (temperature, wind speed, relative humidity, shortwave radiation, etc.), the VQ-VAE model is used to compress and represent the high-resolution remote sensing image sequence from the spatial and semantic dimensions to obtain high-quality block image token representation. The methods involved include the following.

[0080] (1) Local spatiotemporal coding method for high-resolution remote sensing images.

[0081] The first step is to divide the high-resolution remote sensing time series images into blocks according to spatial angles to obtain multiple block images of size 256×256, and use these block images to train a deep neural network, namely the VQ-VAE model, to achieve the effect of local image coding compression, such as Figure 4 shown.

[0082] In the training phase, a high-resolution remote sensing time series LAI image is randomly sampled at each iteration, and the marked patch is , put it into the VQ-VAE model for iterative training. The encoder-decoder structure of VQ-VAE takes the original high-resolution LAI labeled plot image as input. The encoder (CNN Encoder) downsamples the image to a latent vector of 16×16×4 dimensions, keeping the image synchronously compressed from the spatial and channel dimensions to low dimensions, which is convenient for subsequent vectorization processing. The latent vector is input to the decoder (CNN Decoder), which upsamples it to the original block image and keeps it as consistent as possible with the input LAI labeled plot image. During the entire training process, the encoder is learning how to losslessly compress the original LAI labeled plot image into the latent space, while the decoder is learning how to restore the given latent vector to the true original LAI labeled plot image, that is, .

[0083] In the second step, for the high-resolution time series LAI image data, a divide-and-conquer approach is used to model and predict the time series relationship of each divided plot. The latent code vector corresponding to the above plot is flattened into a 1024 Token vector, and the length of the block sequence corresponding to a period of time in the time dimension is 1024×N.

[0084] The third step is to transform the high-resolution time series LAI generation problem into a time series image prediction problem. Given the Token vectors corresponding to the local high-resolution time series LAI images of the previous N moments, the time series prediction model is used to predict the Token vector of the N+1th time point. In order for the time series prediction model to explicitly obtain time information, an additional spatial encoding mechanism is introduced. The dates corresponding to the N moments are mapped to a time coding vector with the same length as the block Token vector through a time embedding module, and then the block Tokens are summed and the results are input into the time series prediction model. After obtaining the predicted Token vector, the VQ-VAE decoder is used to restore the image to achieve the time series prediction of the local high-resolution LAI image.

[0085] (2) Explicit embedding algorithm of meteorological factor coupling matching information

[0086] High-resolution remote sensing LAI image data has many noise points, complex forms and highly mixed information. Simply relying on the encoding algorithm for time series prediction will cause the network model to fall into local optimality and ignore the impact of some key factors on high-resolution image prediction. Based on this, an explicit embedding mechanism for meteorological factor information affecting crop high-resolution remote sensing LAI is designed to decouple crop information from different regions and current meteorological factors for independent modeling, and maintain a conditional Token vector for each crop and current factor.

[0087] Assume that the same block has the same crop type at different time points. In the process of predicting the time series characteristics of the same crop block, several different meteorological element information Token vectors corresponding to the block are selected, and the high-dimensional vectors of the region are interacted with through the cross-attention mechanism at each network level, so as to achieve the purpose of explicitly embedding the meteorological factor information into the high-dimensional representation of the block. The above technical solution is expected to enhance the robustness and interpretability of the high-dimensional representation of the block, thereby providing a reliable guarantee for the subsequent prediction of the time series characteristics of the block.

[0088] (3) Temporal modeling of high-resolution remote sensing images based on self-attention image generation.

[0089] For the high-resolution remote sensing time series LAI image database, a time series encoder based on the Transformer structure is constructed to model the time series relationship of remote sensing images of different time lengths, and fully integrate and utilize the visual information of historical remote sensing observation information. Here, only the Transformer encoder is used for training. Figure 5 As shown in Figure 2, the encoder is mainly composed of a multi-head attention layer and a feedforward network. The multi-head attention layer allows the model to focus on multiple local features at the same time. Figure 5 In the equation, h1, h2 and h3 are model parameters.

[0090] During the training process, each data sampling extracts equal-length time series data of different blocks from the high-resolution remote sensing time series LAI, and uses the local spatiotemporal coding algorithm to convert it into data of dimension B×1024×N, while embedding time annotation information, where B is the batch size and 1024 is the Token vector dimension. After obtaining a series of batches of data, the predicted symbol vector is returned to the trained VQ-VAE decoder to restore it to the original block image size.

[0091] The autoregressive method is used to train the symbolic representation of the crop-specific block image at a time point. The total loss function of the model is: ,in Restore the loss for the predicted latent vector; The predicted latent vector is decoded into the image dimension by the VQ-VAE decoder and the similarity measurement loss of the image dimension is performed with the original image. and The consistency between the predicted Token and the real high-resolution LAI image Token is measured from the perspectives of low dimension and high dimension respectively to accelerate the convergence of the network model.

[0092] In order to prevent the leakage of information of previous and subsequent time series, the self-attention mechanism uses mask operations to construct reasonable causal relationships, that is, the current prediction result is only affected by the previous time series information. In the inference stage, following the decoding mechanism of the language model, for each specific high-resolution LAI block, the Token corresponding to the block image N moments before the time point of interest is used as the conditional input, and the Token result of the predicted block image encoding is output, and the predicted Token vector is returned to the VQ-VAE decoder to restore it to the original high-resolution LAI block image size. Finally, the predicted images corresponding to each block are spliced ​​together to obtain the predicted high-resolution remote sensing LAI image generation result, that is, the predicted remote sensing image.

[0093] In an exemplary embodiment, a computer device is provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a remote sensing time series image generation method.

[0094] In an exemplary embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a remote sensing time-series image generation method is implemented.

[0095] In an exemplary embodiment, a computer program product is provided, including a computer program, which implements a remote sensing time series image generation method when executed by a processor.

[0096] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a remote sensing time series image generation method is implemented.

[0097] Those skilled in the art will understand that Figure 6The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0098] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0099] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0100] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0101] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0102] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A remote sensing time series image generation method, characterized in that: The remote sensing time series image generation method comprises: Determine the remote sensing time series images of various types of crops in the area to be predicted; the remote sensing time series images include: crop distribution maps at multiple consecutive historical moments; Identify any type of crop as the current crop; The crop distribution map of the current crop at each historical moment is divided into blocks at a spatial angle to obtain multiple historical block images corresponding to the historical moment; Input the historical block images of all historical moments of any spatial angle of the current crop into the remote sensing image generation model to obtain the predicted block images of the corresponding spatial angle at the time to be predicted; the remote sensing image generation model is obtained by training the initial model using the training set, and the initial model is constructed based on the Transformer network and the VQ-VAE model; Based on the predicted block images at the time to be predicted of each type of crops in the area to be predicted at each spatial angle, the predicted remote sensing image of the area to be predicted at the time to be predicted is determined.

2. The remote sensing time series image generation method according to claim 1, characterized in that: The training process of the remote sensing image generation model includes: Acquire the training set; the training set includes multiple groups of training data, and the training data includes: training block images of the first N historical moments and the corresponding training block images of the N+1th historical moment; Constructing the initial model based on the Transformer network and the VQ-VAE model; The initial model is trained using the training set to obtain the remote sensing image generation model.

3. The remote sensing time series image generation method according to claim 2, characterized in that: The initial model includes: an encoder of a VQ-VAE model, a latent vector flattening module, a meteorological factor information display embedding module, an encoder of a Transformer network and a decoder of a VQ-VAE model.

4. The remote sensing time series image generation method according to claim 3, characterized in that: Using the training set to train the initial model to obtain the remote sensing image generation model includes: The initial model is trained based on the training set using a total loss function to obtain the remote sensing image generation model; the total loss function includes: ; in, is the restoration loss of the predicted latent vector; The predicted latent vector is decoded by the decoder of the VQ-VAE model to obtain the similarity measurement loss in the image dimension between the predicted block image and the training block image.

5. The remote sensing time series image generation method according to claim 4, characterized in that: The historical block images of all historical moments of any spatial angle of the current crop are input into the remote sensing image generation model to obtain the predicted block images of the corresponding spatial angle at the time to be predicted, including: Determine any spatial angle of the current crop as the current angle; The historical block images of all historical moments of the current angle are input into the encoder of the VQ-VAE model in the remote sensing image generation model to obtain a latent vector of size 16×16×4 corresponding to the historical block image; The latent vector flattening module in the remote sensing image generation model is used to flatten each latent vector, and the meteorological factor information display embedding module is used to couple the meteorological factor to the latent vector to obtain a Token vector of size 1024 corresponding to the historical block image. Based on all the token vectors of the current angle and the encoder of the Transformer network, determine the crop-specific block image symbol of the current angle; The crop-specific block image symbol of the current angle is input into the decoder of the VQ-VAE model in the remote sensing image generation model to obtain the predicted block image of the current angle at the time to be predicted.

6. The remote sensing time series image generation method according to claim 1, characterized in that: Determine the remote sensing time series images of various types of crops in the area to be predicted, including: Acquire historical time-series satellite remote sensing data of the area to be predicted; the historical time-series satellite remote sensing data includes: satellite remote sensing data of multiple consecutive historical moments; The remote sensing time series images of various types of crops in the area to be predicted are determined according to the historical time series satellite remote sensing data.

7. The remote sensing time series image generation method according to claim 6, characterized in that: Determining remote sensing time series images of various types of crops in the area to be predicted based on the historical time series satellite remote sensing data includes: Obtain the correlation between the spectral reflectance and leaf area index images of various types of crops; Based on the satellite remote sensing data at each historical moment and the association relationship, determining an initial leaf area index data set corresponding to the historical moment; Based on the leaf area index images in the initial leaf area index data set at each historical moment, the crop distribution map of each type of crop at the corresponding historical moment is determined, and the remote sensing time series images of each type of crop in the area to be predicted are further obtained.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the remote sensing time series image generation method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the remote sensing time series image generation method described in any one of claims 1 to 7 is implemented.

Images