A deep soil water detection method based on transfer learning
Through transfer learning-based methods and Richard equation constraints, the problem of difficulty in obtaining deep soil water data is solved, the monitoring accuracy is improved, and the efficient detection of deep soil water data is achieved.
Patent Information
- Application Number
- CN202410475451.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-19
- Publication Date
- 2024-12-27
- Estimated Expiration
- 2044-04-19
AI Technical Summary
The prior art is difficult to effectively monitor and detect deep soil water data, and because the data database is small, data accuracy and robustness are difficult to guarantee.
Using a transfer learning method, a neural network model is trained using shallow soil water data sets, and the output of Richard's equation constraint model is carried out to carry out transfer learning and correction of deep soil water data to build a deep soil water migration model.
The accuracy of deep soil water monitoring is improved, and the problem of less deep soil water data is compensated. Combined with the constraints of Richard's equation, the model has higher accuracy.
Smart Images
Figure CN118350447B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a deep soil water detection method, and in particular to a deep soil water detection method based on transfer learning. Background Art
[0002] As one of the factors affecting regional climate and vegetation growth dynamics, soil moisture data detection is of great significance to regional agricultural production and water resources management. The development of satellite remote sensing technology and the increase in the number of detection stations in recent years have made the soil water database more complete than before, but it is mainly shallow soil water data with higher sensitivity and wider application range. For deep soil water data, it is more inconvenient to obtain than shallow soil water, and the database size is much smaller than that of shallow soil water.
[0003] Many studies have used deep learning frameworks to train and detect deep soil water. However, due to the small size of the database and the uneven distribution of data in different regions, it is difficult to carry out large-scale soil water detection. At the same time, it is difficult to ensure data accuracy and improve robustness. Summary of the invention
[0004] Purpose of the invention: To address the above problems, this method proposes a deep soil water detection method based on transfer learning, which can monitor the deep soil water content in the absence of deep soil water data and improve the monitoring accuracy of deep soil water.
[0005] Technical solution: The technical solution adopted by the present invention is a deep soil water detection method based on transfer learning, including: training a neural network model according to a shallow soil water data set; performing transfer learning on the trained neural network model to obtain a first neural network model; constructing a second neural network model, using the Richard equation to constrain the output value of the second neural network model, and using the output value of the second neural network model under the Richard equation constraint to correct the deep soil moisture content output by the first neural network to obtain a deep soil water migration model.
[0006] The shallow soil water data set includes: remote sensing data, geographic data, soil data, meteorological data and shallow soil water data; time.
[0007] The remote sensing data includes synthetic aperture radar observation angle, radar VV polarization, radar VH polarization, and vegetation NDVI value; the geographic data includes longitude and latitude, elevation, slope, aspect, and depth; the soil data includes sand content and soil density; and the meteorological data includes temperature and rainfall.
[0008] The transfer learning refers to: for the neural network model trained using the shallow soil water data set, the shallow soil water data is replaced with the deep soil water data, and the parameters of the hidden layer in the neural network are fine-tuned and then trained again.
[0009] The construction of the second neural network model includes: taking soil depth z and time t as input values, taking matrix potential, permeability coefficient and deep soil moisture content as output values, using shallow soil water data set to obtain the second neural network model through transfer learning, and calculating the cost function as follows:
[0010] ,
[0011] In the formula, is the cost function, , and They are respectively the matrix potential, permeability coefficient, and deep soil moisture content output by the second neural network model, t is time, and z is soil depth.
[0012] The method of using the Richard equation constraint on the trained neural network model also includes: calculating the cost function value and the measured soil moisture content. The second neural network model outputs deep soil moisture content Bring in the loss function for correction, and the calculation formula is:
[0013] ,
[0014] In the formula, is the loss function, i represents the i-th moment of the data taken, represents the deep soil moisture content output by the first neural network model, represents the deep soil moisture content output by the second neural network model, To measure soil moisture content, Calculate the value for the cost function.
[0015] The deep soil water migration model is used to input remote sensing data, geographic data, soil data, and meteorological data measured on the day to output deep soil water data.
[0016] The present invention proposes a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the deep soil water detection method based on transfer learning when executing the computer program.
[0017] The present invention proposes a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the deep soil water detection method based on transfer learning is implemented.
[0018] The present invention proposes a computer program product, including a computer program and / or instructions, which implement the deep soil water detection method based on transfer learning when executed by a processor.
[0019] Beneficial effects: Compared with the prior art, the present invention uses shallow soil water with large data volume to transfer deep soil water to make up for the problem of less deep soil water data and poor training set, and introduces the Richard equation to constrain the model to improve accuracy. This method provides a solution for the lack of data in deep soil water training, and combined with the Richard equation constraint, the model has higher accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 is a flow chart of the deep soil water detection method based on transfer learning described in the present invention;
[0021] Figure 2 It is a schematic diagram of the architecture of the deep soil water migration model described in the present invention. DETAILED DESCRIPTION
[0022] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0023] The deep soil water detection method based on transfer learning described in the present invention has a flow chart as shown in Figure 1 As shown, it includes: training a neural network model based on a shallow soil water dataset; using the Richard equation constraint on the trained neural network model, and using the trained neural network model under the Richard equation constraint to perform transfer learning on the deep soil water dataset to obtain a deep soil water migration model.
[0024] The following is a detailed introduction to each part.
[0025] 1: Extract and process independent and dependent variable data.
[0026] Among them, the independent variables and dependent variables in the present invention use shallow soil water as the dependent variable for initial training, and use deep soil water as the dependent variable for migration training; the independent variables include: remote sensing data, including synthetic aperture radar observation angle, radar VV polarization, radar VH polarization, vegetation NDVI value; geographic data, including longitude and latitude, elevation, slope, aspect, depth; soil data, including sand content, soil density; meteorological data, including temperature, rainfall; time. The independent variable is resampled to the daily average value of the station observation within the unit resolution grid dependent variable grid. The independent variable data is normalized so that the model does not use incorrect weights due to different data units. The calculation formula is:
[0027] ,
[0028] in, is a normalized independent variable, x is a variable, , The maximum and minimum values of the independent variable are set manually. For example, the minimum value of sand content is set to 0 and the maximum value is set to 100.
[0029] In addition, for the shallow soil water data and deep soil water data, 70% of the shallow soil water data were randomly used for training and 30% for verification; 10% of the deep soil water data were randomly used for training and 90% for verification.
[0030] 2: Build a neural network.
[0031] Constructing a neural network means taking the independent variable as the input layer, and after training with two hidden layers, outputting the dependent variable shallow soil water data, which is then compared with the true value and corrected.
[0032] 3: Use deep soil water data for transfer learning.
[0033] Transfer learning means: for a model trained with a large amount of shallow soil water data and various data, the shallow soil water data of the dependent variable is replaced with deep soil water data, the parameters of the hidden layer in the neural network are fine-tuned, and then the model is trained again to output the deep soil water data. .
[0034] After replacing with deep soil water data, the shallow soil depth z in the input independent variable should be changed to the soil depth where the deep soil water data is located.
[0035] 4: Constrain the model using the Richard equation.
[0036] Construct another neural network, taking soil depth z and time t as input values, and transforming matrix potential , Permeability coefficient Deep soil moisture content As the output value, transfer learning obtains the second neural network model, whose output value should satisfy the Richard equation. The cost function is used to describe the degree of deviation between the neural network and the Richard equation as follows:
[0037] ,
[0038] For deep soil moisture , the calculated value of the cost function and the measured soil moisture content should be Deep soil moisture content output by the second neural network model Bring in the loss function for correction, and its calculation formula is:
[0039] ,
[0040] in, is the measured value.
[0041] The final deep soil water transport model architecture is as follows: Figure 2 As shown. By using the obtained deep soil water migration model and inputting the remote sensing data, geographic data and other data measured on the day, the output of deep soil water data is realized.
[0042] In one 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 implements the above-mentioned deep soil water detection method based on transfer learning when executing the computer program.
[0043] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned deep soil water detection method based on transfer learning is implemented.
[0044] In one embodiment, a computer program product is provided, comprising a computer program / instruction, which, when executed by a processor, implements the deep soil water detection method based on transfer learning.
[0045] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt 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 codes.
[0046] The present application 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 application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes 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.
[0047] 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.
[0048] 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 The steps for the functions specified in one or more boxes.
Claims
1. A deep soil water detection method based on transfer learning, characterized in that: include: A neural network model is trained based on a shallow soil water data set; transfer learning is performed on the trained neural network model to obtain a first neural network model; a second neural network model is constructed, the output value of the second neural network model is constrained by the Richard equation, and the deep soil water content output by the first neural network is corrected using the output value of the second neural network model constrained by the Richard equation to obtain a deep soil water migration model; The construction of the second neural network model includes: taking soil depth z and time t as input values, taking matrix potential, permeability coefficient and deep soil moisture content as output values, using shallow soil water data set to obtain the second neural network model through transfer learning, and calculating the cost function as follows: In the formula, is the cost function, and are the matrix potential, permeability coefficient, and deep soil moisture content output by the second neural network model, t is time, and z is soil depth; The output value of the second neural network model under the constraint of the Richard equation is used to correct the deep soil moisture content output by the first neural network, including: the calculated value of the cost function, the measured soil moisture content θ (i) The second neural network model outputs deep soil moisture content Bring in the loss function for correction, and the calculation formula is: In the formula, L(W,b) is the loss function, i represents the i-th moment of the data taken, represents the deep soil moisture content output by the first neural network model, represents the deep soil moisture content output by the second neural network model, θ is the measured soil moisture content, Calculate the value for the cost function.
2. The deep soil water detection method based on transfer learning according to claim 1 is characterized in that: The shallow soil water data set includes: remote sensing data, geographic data, soil data, meteorological data, shallow soil water data, and time.
3. The deep soil water detection method based on transfer learning according to claim 2 is characterized in that: The remote sensing data includes synthetic aperture radar observation angle, radar VV polarization, radar VH polarization, and vegetation NDVI value; the geographic data includes longitude and latitude, elevation, slope, aspect, and depth; the soil data includes sand content and soil density; and the meteorological data includes temperature and rainfall.
4. The deep soil water detection method based on transfer learning according to claim 1 is characterized in that: The transfer learning refers to: for the neural network model trained using the shallow soil water data set, the shallow soil water data is replaced with the deep soil water data, and the parameters of the hidden layer in the neural network are fine-tuned and then trained again.
5. The deep soil water detection method based on transfer learning according to claim 1 is characterized in that: The deep soil water migration model is used to input remote sensing data, geographic data, soil data, and meteorological data measured on the day to output deep soil water data.
6. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the deep soil water detection method based on transfer learning described in any one of claims 1 to 5 is implemented.
7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the deep soil water detection method based on transfer learning described in any one of claims 1 to 5 is implemented.
8. A computer program product comprising a computer program and / or instructions, characterized in that: When the computer program / instructions are executed by a processor, the deep soil water detection method based on transfer learning described in any one of claims 1 to 5 is implemented.