A crop yield inversion method and system based on fusion network model

By constructing an LSTM-CNN coupled neural network model and combining meteorological data and remote sensing image data, the data adaptation and accuracy problems of existing crop growth models in regional scale applications were solved, and high-resolution crop yield inversion was achieved.

CN118735140BActive Publication Date: 2024-11-22WUHAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411223873.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-03
Publication Date
2024-11-22
Estimated Expiration
2044-09-03

AI Technical Summary

Technical Problem

Existing crop growth models suffer from limited data source adaptability, poor model generalization, and low accuracy when applied at the regional scale. Furthermore, remote sensing technology is costly and difficult to promote on a large scale, while meteorological data has low resolution and cannot effectively reflect plot differences.

Method used

An LSTM-CNN coupled neural network model was constructed using LSTM and CNN. Combined with meteorological data and remote sensing image data, the model was pre-trained using the WOFOST crop growth model, and the LSTM layer parameters were frozen to obtain a crop yield inversion model.

Benefits of technology

It achieves higher spatial resolution crop yield inversion, integrates plot information, improves the model's generalization and accuracy, and reduces the cost of remote sensing data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118735140B_ABST
    Figure CN118735140B_ABST
Patent Text Reader

Abstract

The present invention provides a crop yield inversion method and system based on a fusion network model, which belongs to the technical field of crop yield measurement, including: obtaining meteorological data and remote sensing image data in a study area; using LSTM and CNN to construct an LSTM-CNN coupled neural network model; using simulated data generated by a WOFOST crop growth model to pre-train the LSTM, freezing LSTM layer parameters, and using meteorological data and remote sensing image data to train the LSTM-CNN coupled neural network model to obtain a crop yield inversion model. The present invention combines the advantages of a crop growth model and a neural network model, extracts information from meteorological data and remote sensing image data respectively, and fuses feature information into a feature vector, so that a priori yield inversion information can be obtained from meteorological data, and plot information can also be introduced to achieve crop yield inversion with higher spatial resolution.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of crop yield estimation, and in particular to a crop yield inversion method and system based on a fusion network model. Background Art

[0002] In the field of crop management, estimating crop yields can adjust resource inputs and predict crop price trends. It is not only a way to set prices in advance and effectively increase profits, but also an effective way to reduce resource waste and improve the quality of agricultural products, arable land productivity and agricultural production efficiency. At the same time, high-precision yield estimates can optimize the spatial planting pattern of crops, strengthen crop production management, and further tap the land production potential.

[0003] At present, crop growth models are often used to simulate crop growth processes using environmental factors to reveal the causes and nature of crop growth and development. However, when crop growth models develop from single-point research to regional scale applications, the surface and near-surface environmental heterogeneity problems caused by the increase in spatial scale lead to many difficulties in obtaining some macro data in the model and regionalization of parameters, and the model simulation results will also have great uncertainty. Studying the adaptability of the crop growth model combined with the neural network and finding a method for simulating the crop growth model with the neural network can not only reflect the crop growth and development process in terms of mechanism, but also use the powerful transfer learning ability of deep learning to improve the generalization of the model, which is conducive to achieving complementary advantages and enhancing application potential. In addition, there are still some problems in the existing remote sensing technology for crop yield estimation, such as insufficient high-resolution remote sensing data sources, high cost of use, and purely empirical yield estimation models are only applicable to small-scale study areas and difficult to be extended to large-scale yield estimation. The data used as the input source of the model generally also involves meteorological data, which often has a low spatial resolution. In fact, the meteorological environment in a larger spatial range will not change much, but the yield per unit area of ​​each farmland plot is different. Therefore, to achieve crop yield inversion with higher spatial resolution, it is necessary to introduce plot information. Remote sensing images are ideal data that can provide plot information. However, the input data of the existing WOFOST (World FOodSTudies) crop growth model is fixed and cannot input remote sensing image data. The model is a crop growth model developed by the World Food Research Center (CWFS) and Wageningen University in the Netherlands. It simulates a dynamic explanatory model of annual crop growth under specific soil and climatic conditions. The model emphasizes quantitative land evaluation, regional yield forecasting, risk analysis, and quantitative applications of inter-annual yield changes and the impact of climate change.

[0004] Therefore, it is necessary to propose a new model for crop yield estimation that can effectively solve the above problems. Summary of the invention

[0005] The present invention provides a crop yield inversion method and system based on a fusion network model, which are used to solve the defects of the crop growth model used for crop yield estimation in the prior art, such as limited data source adaptation range, weak model generalization and low result accuracy.

[0006] In a first aspect, the present invention provides a crop yield inversion method based on a fusion network model, comprising:

[0007] Obtain meteorological data and remote sensing image data in the study area;

[0008] Use LSTM and CNN to build a LSTM-CNN coupled neural network model;

[0009] Pre-training the LSTM using simulated data generated by the WOFOST crop growth model, freezing LSTM layer parameters, and training the LSTM-CNN coupled neural network model using the meteorological data and the remote sensing image data to obtain a crop yield inversion model;

[0010] The meteorological data and remote sensing image data of the target area are input into the crop yield inversion model to obtain the crop yield inversion result of the target area.

[0011] According to a crop yield inversion method based on a fusion network model provided by the present invention, meteorological data and remote sensing image data in a study area are obtained, including:

[0012] determining a plurality of target boxes within the study area;

[0013] Using a visual large model SAM to segment the plots within the multiple target frames, and determining the coordinates of the center points corresponding to the multiple target frames;

[0014] The meteorological data and the remote sensing image data released by the Climate Change Service Center are obtained based on the coordinates of the center point. According to a crop yield inversion method based on a fusion network model provided by the present invention, the meteorological data includes solar irradiance, maximum temperature, minimum temperature, wind speed, air pressure and precipitation, and the solar irradiance, the maximum temperature, the minimum temperature, the wind speed, the air pressure and the precipitation constitute a multidimensional meteorological data matrix;

[0015] The remote sensing image data includes R channel image data, G channel image data and B channel image data, and the R channel image data, the G channel image data and the B channel image data constitute a multi-dimensional remote sensing data matrix.

[0016] According to a crop yield inversion method based on a fusion network model provided by the present invention, LSTM and CNN are used to construct an LSTM-CNN coupled neural network model, including:

[0017] Determine that the LSTM includes an LSTM input layer, a one-dimensional convolution layer, a batch normalization layer, and an LSTM output layer connected in sequence;

[0018] Determine that the CNN includes a CNN input layer, a CNN layer, and a fully connected layer connected in sequence, and the CNN layer includes two stacked layers connected in sequence, each stacked layer is a two-dimensional convolutional layer and a maximum pooling layer;

[0019] Input the meteorological data into the LSTM to obtain an LSTM output vector, and input the remote sensing image data into the CNN to obtain a CNN output vector;

[0020] Merging the LSTM output vector and the CNN output vector to obtain a fused feature vector;

[0021] The fused feature vector is input into the overall fully connected layer to obtain the output result of the coupled neural network model.

[0022] According to a crop yield inversion method based on a fusion network model provided by the present invention, it is determined that the LSTM includes an LSTM input layer, a one-dimensional convolution layer, a batch normalization layer and an LSTM output layer connected in sequence, including:

[0023] The LSTM input layer expands the meteorological data according to the multi-dimensional meteorological matrix and inputs it into the one-dimensional convolutional layer;

[0024] The one-dimensional convolution layer extracts information from the meteorological data from a time dimension to obtain meteorological extraction time information, and the convolution layer includes the same number of input channels and output channels;

[0025] The batch normalization layer performs normalization processing on the meteorological extraction time information, and outputs the normalization processing result to the LSTM layer;

[0026] The LSTM layer flattens the normalization processing result into the LSTM output vector, and the number of hidden neurons in the LSTM layer corresponds to the dimension of the number of output channels of the convolution layer.

[0027] According to a crop yield inversion method based on a fusion network model provided by the present invention, it is determined that the CNN includes a CNN input layer, a CNN layer and a fully connected layer connected in sequence, and the CNN layer includes two stacked layers connected in sequence, each stacked layer is a two-dimensional convolutional layer and a maximum pooling layer, including:

[0028] The CNN input layer resamples the remote sensing image data to a preset size and normalizes it into a tensor;

[0029] Input the tensor to the CNN layer, wherein the number of output channels of the first two-dimensional convolutional layer in the CNN layer is equal to the number of input channels of the second two-dimensional convolutional layer and has the same convolution kernel size, and the maximum pooling layer in each stacked layer has the same number of kernels;

[0030] The output result of the CNN layer is input into the fully connected layer to obtain the CNN output vector.

[0031] According to a crop yield inversion method based on a fusion network model provided by the present invention, the LSTM is pre-trained using simulated data generated by the WOFOST crop growth model, the LSTM layer parameters are frozen, and the LSTM-CNN coupled neural network model is trained using the meteorological data and the remote sensing image data to obtain a crop yield inversion model, including:

[0032] Acquire the simulation data, and load the simulation data into the LSTM of the LSTM-CNN coupled neural network model;

[0033] Freezing the LSTM layer parameters in the LSTM so that the LSTM layer parameters stop updating during the training process;

[0034] Based on a preset training cycle, batch size, preset loss function and gradient optimization method, the LSTM-CNN coupled neural network model is trained using sample data consisting of the meteorological data and the remote sensing image data until the model meets the preset convergence conditions, the training is stopped, and the current training model is saved as the crop yield inversion model.

[0035] In a second aspect, the present invention further provides a crop yield inversion system based on a fusion network model, comprising:

[0036] Acquisition module, used to obtain meteorological data and remote sensing image data in the study area;

[0037] A building block for building an LSTM-CNN coupled neural network model using LSTM and CNN;

[0038] A training module, used to pre-train the LSTM using the simulated data generated by the WOFOST crop growth model, freeze the LSTM layer parameters, and train the LSTM-CNN coupled neural network model using the meteorological data and the remote sensing image data to obtain a crop yield inversion model;

[0039] The inversion module is used to input the meteorological data and remote sensing image data of the target area into the crop yield inversion model to obtain the crop yield inversion result of the target area.

[0040] In a third aspect, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a crop yield inversion method based on a fusion network model as described above is implemented.

[0041] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the crop yield inversion methods based on the fusion network model as described above.

[0042] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0043] The crop yield inversion method and system based on the fusion network model provided by the present invention combine the advantages of the crop growth model and the neural network model, extract information from meteorological data and remote sensing image data respectively, and fuse the feature information into a single feature vector. It can not only obtain a priori yield inversion information from meteorological data, but also introduce plot information to achieve crop yield inversion with higher spatial resolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 It is one of the flow charts of the crop yield inversion method based on the fusion network model provided by the present invention;

[0046] Figure 2 This is the second flow chart of the crop yield inversion method based on the fusion network model provided by the present invention;

[0047] Figure 3 It is a schematic diagram of sampling and selecting a target frame when acquiring data provided by the present invention;

[0048] Figure 4 It is a schematic diagram of using the SAM model to segment a plot provided by the present invention;

[0049] Figure 5 It is a LSTM-CNN neural network framework diagram provided by the present invention;

[0050] Figure 6 It is a structural schematic diagram of a crop yield inversion system based on a fusion network model provided by the present invention;

[0051] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] In view of the shortcomings of the existing technology, the present invention proposes a crop yield inversion method based on network model combination, which combines the crop growth model with the neural network model, integrates meteorological data and remote sensing image data, and utilizes the mechanism of the WOFOST model to realize the inversion of crop yield.

[0054] Figure 1 is one of the flow charts of the crop yield inversion method based on the fusion network model provided by the embodiment of the present invention, such as Figure 1 As shown, including:

[0055] Step 100: Acquire meteorological data and remote sensing image data in the study area;

[0056] Step 200: Using LSTM and CNN to construct an LSTM-CNN coupled neural network model;

[0057] Step 300: pre-training the LSTM using the simulated data generated by the WOFOST crop growth model, freezing the LSTM layer parameters, and training the LSTM-CNN coupled neural network model using the meteorological data and the remote sensing image data to obtain a crop yield inversion model;

[0058] Step 400: inputting the meteorological data and remote sensing image data of the target area into the crop yield inversion model to obtain the crop yield inversion result of the target area.

[0059] Specifically, Figure 2 As shown, an embodiment of the present invention proposes a long short-term memory network combined with a convolutional neural network (LSTM-CNN) that combines meteorological data and remote sensing image data.

[0060] First, the meteorological data and remote sensing image data in the study area are obtained, and then the LSTM-CNN coupled neural network model is constructed, and some parameters of the LSTM network are frozen. This part is achieved by pre-training the LSTM network with a simulated meteorological database, and freezing these parameters. The LSTM-CNN coupled neural network model with frozen LSTM network parameters is further trained using the real meteorological data and remote sensing images obtained. After the training is completed, the crop yield inversion model is obtained. Finally, in practical applications, the meteorological data and remote sensing image data of the target area are input into the trained crop yield inversion model to obtain the crop yield inversion results of the target area.

[0061] The present invention combines the advantages of crop growth models and neural network models, extracts information from meteorological data and remote sensing image data respectively, and fuses the feature information into a single feature vector. It can not only obtain a priori yield inversion information from meteorological data, but also introduce plot information to achieve crop yield inversion with higher spatial resolution.

[0062] In some embodiments, obtaining meteorological data and remote sensing image data in the study area includes:

[0063] determining a plurality of target boxes within the study area;

[0064] Using a visual large model SAM to segment the plots within the multiple target frames, and determining the coordinates of the center points corresponding to the multiple target frames;

[0065] The meteorological data and the remote sensing image data released by the Climate Change Service Center are obtained based on the center point coordinates.

[0066] Specifically, the embodiment of the present invention selects an experimental area, selects a target frame in the experimental area, and a generation result of the target frame is as follows: Figure 3 As shown in the figure, for example, 7 counties in a province are selected as experimental areas, 1000 sampling points are randomly scattered in the experimental area, 1000 latitude and longitude coordinates are obtained, and then 1000 points with a radius of about 1000m are selected with these coordinates as the center point. 1000m target box.

[0067] In the target box, the Segment Anything Model (SAM) is used to segment the plots in the box. The segmentation results are as follows: Figure 4 As shown, each plot will have a center point coordinate. With this coordinate, the meteorological data published by the Copernicus Climate Change Service (C3S) and the remote sensing image data provided by Google can be downloaded on Google Earth Engine, and the data can be preprocessed to obtain the meteorological data set required by this method, including solar irradiance I, maximum temperature , minimum temperature , wind speed , air pressure , precipitation Meteorological data and remote sensing image data corresponding to the plot. Among them, solar irradiance I, maximum temperature , minimum temperature , wind speed , air pressure , precipitation Corresponding to Figure 2 Input1-6 of Weather Data Inputs in the downloads daily frequency weather data, that is, each type of weather data corresponding to each plot is a vector, including solar irradiance I, maximum temperature , minimum temperature , wind speed , air pressure , precipitation There are 6 types of meteorological data, that is, Input1-6 of each plot is a matrix. The downloaded remote sensing image data has three channels of R, G, and B, which can also form a matrix.

[0068] In some embodiments, LSTM and CNN are used to construct an LSTM-CNN coupled neural network model, including:

[0069] Determine that the LSTM includes an LSTM input layer, a one-dimensional convolution layer, a batch normalization layer, and an LSTM output layer connected in sequence;

[0070] Determine that the CNN includes a CNN input layer, a CNN layer, and a fully connected layer connected in sequence, and the CNN layer includes two stacked layers connected in sequence, each stacked layer is a two-dimensional convolutional layer and a maximum pooling layer;

[0071] Input the meteorological data into the LSTM to obtain an LSTM output vector, and input the remote sensing image data into the CNN to obtain a CNN output vector;

[0072] Merging the LSTM output vector and the CNN output vector to obtain a fused feature vector;

[0073] The fused feature vector is input into the overall fully connected layer to obtain the output result of the coupled neural network model.

[0074] Specifically, Figure 5 As shown, it is a structural diagram of the LSTM-CNN coupled neural network model proposed in an embodiment of the present invention.

[0075] The first part is to build LSTM, where LSTM has an input layer, inputs the meteorological data set, namely Input 1-6, and then passes through a Conv1d convolution layer, namely a one-dimensional convolution layer, to extract information from the meteorological data in the time dimension. The number of input channels of the Conv1d convolution layer is 6, the number of output channels is also 6, the convolution kernel size is 10, and the step size is 5. After passing through this convolution layer, the size becomes 72 6; A batch normalization layer is followed by the convolution layer to obtain vectors X1 to X6, and then the output results are input into the long short-term memory network layer. The number of hidden neurons in the long short-term memory network layer is set to correspond to the dimension of the convolution layer output, which is 72 in this embodiment. Then the output result is flattened into a long vector, that is, Figure 5 The LSTM output vector in .

[0076] The second part is to build a CNN. The CNN part has an input layer to input remote sensing image data. Before input, the remote sensing image data is preprocessed and the image size is uniformly resampled to 32. 32, and normalized to a tensor. Then input to a Conv2d convolutional layer, that is, a two-dimensional convolutional neural network layer, the number of input channels of the two-dimensional convolutional neural network convolutional layer is 3, the number of output channels is 6, and the convolution kernel size is 3. 3, stride 2, padding 2, so that the shape of the tensor is (6,16,16), and then through a maximum pooling layer, the kernel size of the maximum pooling layer is set to (2,2), the stride is 2, and the padding is 0. Then it is input to a Conv2d convolution layer, the number of input channels of the Conv2d convolution layer is 6, the number of output channels is 1, and the convolution kernel size is 3 3, with a step size of 1 and a padding of 2. Then it passes through a maximum pooling layer, with the kernel size of the maximum pooling layer set to (2,2), a step size of 2, and a padding of 2. Finally, it is input into a fully connected layer to obtain the output result of the CNN part, that is, Figure 5 The convolutional neural network output vector in .

[0077] The third part combines the obtained LSTM output vector and the convolutional neural network output vector into one vector using fusion, that is, Figure 5 The fused feature vector in .

[0078] In the fourth part, the obtained fused feature vector is input into a fully connected layer to obtain the final output result of the network model.

[0079] In some embodiments, the LSTM is pre-trained using simulated data generated by the WOFOST crop growth model, the LSTM layer parameters are frozen, and the LSTM-CNN coupled neural network model is trained using the meteorological data and the remote sensing image data to obtain a crop yield inversion model, including:

[0080] Acquire the simulation data, and load the simulation data into the LSTM of the LSTM-CNN coupled neural network model;

[0081] Freezing the LSTM layer parameters in the LSTM so that the LSTM layer parameters stop updating during the training process;

[0082] Based on a preset training cycle, batch size, preset loss function and gradient optimization method, the LSTM-CNN coupled neural network model is trained using sample data consisting of the meteorological data and the remote sensing image data until the model meets the preset convergence conditions, the training is stopped, and the current training model is saved as the crop yield inversion model.

[0083] Specifically, the training process of the LSTM-CNN coupled neural network model in the embodiment of the present invention includes:

[0084] Before starting training, load the pre-trained parameters of the LSTM network model. The pre-trained parameters are obtained by pre-training with simulated data. After loading the parameters, freeze the parameters of the LSTM layer (i.e. Figure 5 The parameters of the long short-term memory network layer in ), in the subsequent training process, these parameters are no longer updated.

[0085] It should be noted that the above simulation data is obtained by the WOFOST crop growth model. The specific steps include:

[0086] (1) Crop development stage

[0087] In the WOFOST model, phenology is described by the dimensionless state variable DevelopmentStage (DVS). For most annual crops, DVS is set to 0 at seedling emergence, 1 at flowering, and 2 at maturity. Phenological development in WOFOST is calculated by adding the development stage with the development rate at each time step.

[0088] Development rate The calculation method is:

[0089]

[0090] in, represents the daily effective temperature, It indicates the total accumulated temperature required to enter the next development stage, which varies according to the growth stage. Take different values, namely TSUM1 and TSUM2 in the WOFOST input parameters.

[0091] yes Accumulation of effective sunshine time:

[0092]

[0093] in, Expressed as sunshine duration, Indicates the most suitable sunshine duration, Indicates the minimum sunshine duration.

[0094] (2) Respiration and dry matter accumulation

[0095] The total daily assimilation rate of a crop is calculated based on the absorbed radiation and the photosynthetic response curve of the leaves. This response depends on temperature. The absorbed radiation is calculated based on the total incident radiation and the leaf area. Since the response of photosynthesis to light intensity is nonlinear, changes in radiation levels need to be taken into account. The effective amount of radiation absorbed by the leaf surface can be calculated using the following formula:

[0096]

[0097] in, It is the leaf layer The net radiation is the reflection coefficient, is the radiation level of the plant canopy on a sunny day, k is the extinction coefficient, and L is the leaf area index.

[0098] The light absorbed by the leaf layer can be calculated using the following formula:

[0099]

[0100] in, Represents the leaf layer Absorbed radiation.

[0101] The maximum total assimilation rate can be calculated using the following formula.

[0102]

[0103] in, is the total assimilation rate, is the maximum assimilation rate, and ε is the original light energy utilization rate of a single leaf.

[0104] (3) Dry matter distribution

[0105] After the crop respires, the dry matter produced by various organs will be redistributed. Through the distribution of dry matter, we can understand in detail the substances accumulated by the crop and the nutrients it contains. The formula for dry matter W distribution is:

[0106]

[0107] in, is the distribution coefficient, Expressed as the daily assimilation rate, is the respiration rate of crops at real temperature or simulated temperature. In the distribution of crop dry matter, various organs of crops are distributed in different proportions. The dry matter distribution of the model is divided into aboveground dry matter and underground dry matter ;

[0108]

[0109]

[0110] in, It represents the dry matter distribution coefficient to roots, which is the FRTB parameter in the WOFOST input parameters.

[0111] Above ground dry matter Will be distributed to different organs of crops:

[0112]

[0113] in, For the dry matter distribution of crop leaves, For the dry matter distribution of crop stems, Dry matter distribution for crop storage organs, , , is the distribution coefficient that sums to 1.

[0114] When the growth stage reaches full maturity, that is, DVS=2, This is the desired simulated output value.

[0115] When the WOFOST model is actually used for yield simulation, various parameters need to be set. In further model training, the upper limit of the training cycle epochs is set to 300, the batch size is set to 32, the loss function is the mean square error (MSE), the gradient optimization method is the Adam optimization method, the Adam initial learning rate parameter is set to 0.001, and the momentum parameter is set to 0.09. Input the training samples into the set LSTM-CNN coupled neural network model for training. When the loss function value loss does not improve for 10 consecutive training cycles, stop training, save the model, and obtain the trained crop yield inversion model.

[0116] It is understandable that the input data of the existing WOFOST crop growth model is fixed and image data cannot be input. However, the present invention first uses a neural network to learn the WOFOST model into network parameters, and then combines it with another image input neural network, which is equivalent to introducing image information into the WOFOST model. The present invention effectively combines the advantages of the crop growth model and the neural network model, extracts information from meteorological data and remote sensing image data respectively, and fuses feature information into a single feature vector. It can not only obtain a priori yield inversion information from meteorological data, but also introduce plot information to achieve crop yield inversion with higher spatial resolution.

[0117] The crop yield inversion system based on the network model provided by the present invention is described below. The crop yield inversion system based on the network model described below and the crop yield inversion method based on the network model described above can be referenced to each other.

[0118] Figure 6 is a schematic diagram of the structure of a crop yield inversion system based on a fusion network model provided by an embodiment of the present invention, such as Figure 6 As shown, it includes: an acquisition module 61, a construction module 62, a training module 63 and an inversion module 64, wherein:

[0119] The acquisition module 61 is used to acquire the meteorological data and remote sensing image data in the study area; the construction module 62 is used to construct an LSTM-CNN coupled neural network model using LSTM and CNN; the training module 63 is used to pre-train the LSTM using the simulated data generated by the WOFOST crop growth model, freeze the LSTM layer parameters, and use the meteorological data and the remote sensing image data to train the LSTM-CNN coupled neural network model to obtain a crop yield inversion model; the inversion module 64 is used to input the meteorological data and remote sensing image data of the target area into the crop yield inversion model to obtain the crop yield inversion result of the target area.

[0120] Figure 7 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 7 As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730 and a communication bus 740, wherein the processor 710, the communication interface 720 and the memory 730 communicate with each other through the communication bus 740. The processor 710 may call the logic instructions in the memory 730 to execute the crop yield inversion method based on the fusion network model, the method comprising: obtaining meteorological data and remote sensing image data in the study area; using LSTM and CNN to construct an LSTM-CNN coupled neural network model; using the simulated data generated by the WOFOST crop growth model to pre-train the LSTM, freezing the LSTM layer parameters, using the meteorological data and the remote sensing image data to train the LSTM-CNN coupled neural network model, and obtaining a crop yield inversion model; inputting the meteorological data and remote sensing image data of the target area into the crop yield inversion model to obtain the crop yield inversion result of the target area.

[0121] In addition, the logic instructions in the above-mentioned memory 730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0122] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the crop yield inversion method based on the fusion network model provided by the above-mentioned methods, the method comprising: obtaining meteorological data and remote sensing image data in the study area; using LSTM and CNN to construct an LSTM-CNN coupled neural network model; pre-training the LSTM using simulated data generated by the WOFOST crop growth model, freezing the LSTM layer parameters, and training the LSTM-CNN coupled neural network model using the meteorological data and the remote sensing image data to obtain a crop yield inversion model; inputting the meteorological data and remote sensing image data of the target area into the crop yield inversion model to obtain the crop yield inversion result of the target area.

[0123] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0124] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A crop yield inversion method based on a fusion network model, characterized in that: include: Obtain meteorological data and remote sensing image data in the study area; The LSTM-CNN coupled neural network model is constructed using the long short-term memory network LSTM and the convolutional neural network CNN; Pre-training the LSTM using simulated data generated by the WOFOST crop growth model, freezing LSTM layer parameters, and training the LSTM-CNN coupled neural network model using the meteorological data and the remote sensing image data to obtain a crop yield inversion model; Inputting the meteorological data and remote sensing image data of the target area into the crop yield inversion model to obtain the crop yield inversion result of the target area; LSTM and CNN are used to build an LSTM-CNN coupled neural network model, including: Determine that the LSTM includes an LSTM input layer, a one-dimensional convolution layer, a batch normalization layer, and an LSTM output layer connected in sequence; Determine that the CNN includes a CNN input layer, a CNN layer, and a fully connected layer connected in sequence, and the CNN layer includes two stacked layers connected in sequence, each stacked layer is a two-dimensional convolutional layer and a maximum pooling layer; Input the meteorological data into the LSTM to obtain an LSTM output vector, and input the remote sensing image data into the CNN to obtain a CNN output vector; Merging the LSTM output vector and the CNN output vector to obtain a fused feature vector; Inputting the fused feature vector into the overall fully connected layer to obtain the output result of the coupled neural network model; Determining that the LSTM includes an LSTM input layer, a one-dimensional convolution layer, a batch normalization layer, and an LSTM output layer connected in sequence, including: The LSTM input layer expands the meteorological data according to the multi-dimensional meteorological matrix and inputs it into the one-dimensional convolutional layer; The one-dimensional convolution layer extracts information from the meteorological data from a time dimension to obtain meteorological extraction time information, and the convolution layer includes the same number of input channels and output channels; The batch normalization layer performs normalization processing on the meteorological extraction time information, and outputs the normalization processing result to the LSTM layer; The LSTM layer flattens the normalization processing result into the LSTM output vector, and the number of hidden neurons in the LSTM layer corresponds to the dimension of the number of output channels of the convolution layer; Determine that the CNN includes a CNN input layer, a CNN layer, and a fully connected layer connected in sequence, the CNN layer includes two stacked layers connected in sequence, each stacked layer is a two-dimensional convolutional layer and a maximum pooling layer, including: The CNN input layer resamples the remote sensing image data to a preset size and normalizes it into a tensor; Input the tensor to the CNN layer, wherein the number of output channels of the first two-dimensional convolutional layer in the CNN layer is equal to the number of input channels of the second two-dimensional convolutional layer and has the same convolution kernel size, and the maximum pooling layer in each stacked layer has the same number of kernels; Input the output result of the CNN layer to the fully connected layer to obtain the CNN output vector; The LSTM is pre-trained using the simulated data generated by the WOFOST crop growth model, the LSTM layer parameters are frozen, and the LSTM-CNN coupled neural network model is trained using the meteorological data and the remote sensing image data to obtain a crop yield inversion model, including: Acquire the simulation data, and load the simulation data into the LSTM of the LSTM-CNN coupled neural network model; Freezing the LSTM layer parameters in the LSTM so that the LSTM layer parameters stop updating during the training process; Based on a preset training cycle, batch size, preset loss function and gradient optimization method, the LSTM-CNN coupled neural network model is trained using sample data consisting of the meteorological data and the remote sensing image data until the model meets the preset convergence conditions, the training is stopped, and the current training model is saved as the crop yield inversion model.

2. The crop yield inversion method based on the fusion network model according to claim 1 is characterized in that: Obtain meteorological data and remote sensing image data in the study area, including: determining a plurality of target boxes within the study area; Using a visual large model SAM to segment the plots within the multiple target frames, and determining the coordinates of the center points corresponding to the multiple target frames; The meteorological data and the remote sensing image data released by the Climate Change Service Center are obtained based on the center point coordinates.

3. The crop yield inversion method based on the fusion network model according to claim 2 is characterized in that: The meteorological data includes solar irradiance, maximum temperature, minimum temperature, wind speed, air pressure and precipitation, and the solar irradiance, maximum temperature, minimum temperature, wind speed, air pressure and precipitation constitute a multidimensional meteorological matrix; The remote sensing image data includes R channel image data, G channel image data and B channel image data, and the R channel image data, the G channel image data and the B channel image data constitute a multi-dimensional remote sensing matrix.

4. A crop yield inversion system based on a fusion network model, based on the crop yield inversion method based on a fusion network model as claimed in any one of claims 1 to 3, characterized in that: include: Acquisition module, used to obtain meteorological data and remote sensing image data in the study area; A building block for building an LSTM-CNN coupled neural network model using LSTM and CNN; A training module, used to pre-train the LSTM using the simulated data generated by the WOFOST crop growth model, freeze the LSTM layer parameters, and train the LSTM-CNN coupled neural network model using the meteorological data and the remote sensing image data to obtain a crop yield inversion model; The inversion module is used to input the meteorological data and remote sensing image data of the target area into the crop yield inversion model to obtain the crop yield inversion result of the target area.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the crop yield inversion method based on the fusion network model as described in any one of claims 1 to 3 is implemented.

6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the crop yield inversion based on the fusion network model as described in any one of claims 1 to 3 is realized.