A soil moisture estimation system and method based on feedforward neural network model

Abstract

The present invention relates to a soil moisture estimation system based on a feedforward neural network model, which belongs to the technical field of soil moisture measurement. The system comprises a data acquisition module, a remote sensing data processing module, a multi-layer feedforward neural network model, a model based on environmental information and a fusion calculation module. The data acquisition module comprises a remote sensing information acquisition module and an environmental information acquisition module. The system inputs the data collected by the remote sensing information acquisition module and standardized by the remote sensing data processing module into the multi-layer feedforward neural network model to obtain the soil moisture. SM FNN , input the data collected by the environmental information collection module into the above-mentioned model based on environmental information to obtain soil moisture SM NM , the soil moisture SM FNN and soil moisture SM NM The soil moisture SM estimated by the system is obtained by inputting the above fusion calculation module. Meanwhile, the present invention also includes a soil moisture estimation method based on the above system.

Description

Technical Field

[0001] The present invention relates to the technical field of soil moisture measurement, and in particular to a soil moisture estimation system and method, a computer program product and a readable storage medium. Background Art

[0002] Traditionally, soil moisture content is measured by the drying method, in which soil samples are dried in an oven and the difference between the initial weight and the dry weight of the soil sample is the moisture content. This method is highly accurate but time-consuming.

[0003] Sensors based on soil water dielectric technology, such as time domain reflectometry (TDR) and frequency domain reflectometry (FDR), solve the time-consuming problem of the drying method, but they still require manual collection and still consume a lot of manpower when there are many sampling points.

[0004] Based on the traditional soil moisture measurement methods such as drying method, time domain reflection and frequency domain reflection, some new soil moisture measurement methods based on satellite remote sensing technology, drones and artificial intelligence models are constantly emerging. For example, CN116482128A uses radar mounted on drones to detect soil, extracts soil moisture features from the collected radar signals, inputs the features into a trained deep neural network model, and outputs soil moisture estimation results; for another example, JP2024031096A proposes a low-cost technology for estimating soil moisture content, which uses an estimation model that uses RGB image data obtained by photographing soil and measurement data obtained by measuring soil moisture to learn the correlation between RGB image data and measurement data. The accuracy of these soil moisture measurement methods is highly dependent on the accuracy of the model, and the training of high-precision models requires a large number of tests and large sample data sets to achieve. Summary of the invention

[0005] To solve the above problems in the prior art, the present invention proposes a fusion soil moisture estimation system and method based on a feedforward neural network model, as well as a computer program product for running the method, and a readable storage medium storing the computer program.

[0006] The soil moisture estimation system based on the feedforward neural network model of the present invention comprises a data acquisition module, a remote sensing data processing module, a multi-layer feedforward neural network model, a model based on environmental information and a fusion calculation module;

[0007] The data acquisition module includes a remote sensing information acquisition module and an environmental information acquisition module. The remote sensing information acquisition module acquires four characteristic values, namely, the backscattering value of VV and VH polarization, the image incidence angle and the normalized vegetation index NDVI; the environmental information acquisition module acquires net radiation NR, air humidity AH, and land surface temperature LST;

[0008] The remote sensing data processing module standardizes the backscatter values ​​of VV and VH polarizations, the image incidence angle and NDVI collected by the remote sensing information collection module by the Z-score standardization method. The standardization process is based on the following formula:

[0009]

[0010] Where X represents the initial eigenvalue; X Mean represents the average value; X Std represents the standard deviation of the eigenvalues;

[0011] The multi-layer feedforward neural network model includes an input layer, two hidden layers and an output layer. The input layer selects the backscattering values ​​of VV and VH polarization, the image incident angle and the normalized vegetation index NDVI after the standardization processing of the remote sensing data processing module as input; the output layer is the estimated soil moisture SM FNN , each hidden layer has 4 neurons;

[0012] The model based on environmental information is expressed by the following formula:

[0013]

[0014] Where NR is the net radiation, AH is the air humidity, LST is the surface temperature, SM NM is the soil moisture estimated by the model based on environmental information, a, b, c, d, e are coefficients;

[0015] The fusion calculation module includes a fusion calculation model, and the fusion calculation model is expressed by the following formula:

[0016]

[0017] Where SM is the soil moisture estimated by the system, w1 and w2 are SM FNN and SM NM The corresponding weights;

[0018] The system inputs the data collected by the remote sensing information collection module and standardized by the remote sensing data processing module into the multi-layer feedforward neural network model to obtain soil moisture SM FNN , input the data collected by the environmental information collection module into the above-mentioned model based on environmental information to obtain soil moisture SM NM , the soil moisture SM FNNand soil moisture SM NM The soil moisture SM estimated by the system is obtained by inputting the above fusion calculation model.

[0019] Furthermore, the data acquisition module also includes a time domain reflectometer. Based on the data acquisition module and the remote sensing data processing module, N groups of data DATE_a are obtained to train the multi-layer feedforward neural network model, each group of data DATE_a includes: soil moisture collected by the time domain reflectometer at a certain time, backscatter values ​​of VV and VH polarizations obtained at the same time and standardized by the remote sensing data processing module, image incident angle and normalized vegetation index NDVI; based on the N groups of data DATE_a, the weight of each neuron in the multi-layer feedforward neural network model and the bias of each layer are trained to obtain the multi-layer feedforward neural network model.

[0020] Furthermore, based on the time domain reflectometer and the environmental information acquisition module, M groups of data DATE_b are obtained to calculate the coefficients a, b, c, d, e of the model based on the environmental information, each group of data DATE_b includes: soil moisture collected by the time domain reflectometer at a certain time, and net radiation NR, air humidity AH, and surface temperature LST obtained at the same time; based on the M groups of data DATE_b, the coefficients a, b, c, d, e of the model based on the environmental information are calculated, and then the model based on the environmental information is obtained.

[0021] Furthermore, the data acquisition module also includes a drying measurement module, which obtains Z groups of data DATE_c based on the data acquisition module and the remote sensing data processing module to calculate SM FNN and SM NM Corresponding weights w1 and w2, each set of data DATE_c includes: soil moisture measured by drying of soil samples collected at a certain time SM OD , the standardized VV, VH polarized backscatter values, image incidence angle, normalized vegetation index NDVI collected at the same time, and the net radiation NR, air humidity AH and surface temperature LST collected at the same time; based on the multi-layer feedforward neural network model and the model based on environmental information, the Z group of soil moisture is obtained from the above Z group of data DATE_c SM OD , SM FNN and SM NM , the values ​​of group Z weights w1 and w2 can be obtained by bringing them into the fusion calculation model, and the average value of group Z weights w1 and w2 can be obtained by calculating the average value SM FNN and SMNM The corresponding values ​​of weights w1 and w2.

[0022] Furthermore, the remote sensing information acquisition module includes a synthetic aperture radar and a camera. The synthetic aperture radar is carried on a remote sensing satellite, and the camera is carried on a drone. The remote sensing information acquisition module obtains the backscattering values ​​and image incident angles of VV and VH polarizations based on the synthetic aperture radar, and obtains the normalized vegetation index NDVI based on the low-altitude remote sensing image of the camera.

[0023] Furthermore, the environmental information acquisition module includes a net radiation meter, a hygrometer and a surface temperature inversion module, which are used to collect net radiation NR, air humidity AH and surface temperature LST respectively.

[0024] Furthermore, the drying measurement module includes a drying box, which weighs the soil sample to be tested to obtain the sample wet weight. W WET Then the soil sample to be tested is completely dried in a drying oven to obtain the sample dry weight W DRY , drying measured soil moisture .

[0025] The present invention also includes a soil moisture estimation method according to the above soil moisture estimation system, comprising the following steps:

[0026] S1. Configure a multi-layer feedforward neural network model, which includes an input layer, two hidden layers and an output layer. The input layer selects the backscatter values ​​of VV and VH polarization, the image incident angle and NDVI after standardization by the remote sensing data processing module as input; the output layer is the estimated soil moisture SM FNN ; Each hidden layer has 4 neurons;

[0027] S2. Establishing a model based on environmental information, wherein the model based on environmental information is represented by the following formula:

[0028]

[0029] Where NR is the net radiation, AH is the air humidity, LST is the surface temperature, SM NM is the soil moisture estimated by the model based on environmental information, a, b, c, d, e are coefficients;

[0030] S3. Establish a fusion calculation model based on soil moisture SM FNN and soil moisture SM NM Obtain soil moisture SM, and the fusion calculation model is expressed by the following formula:

[0031]

[0032] Among them, w1 and w2 are SM FNN and SM NM The corresponding weights;

[0033] S4. Data collection, including soil moisture collected by time domain reflectometer, VV, VH polarization backscattering values ​​and image incidence angles obtained by synthetic aperture radar, normalized vegetation index NDVI, net radiation NR, air humidity AH, surface temperature LST obtained by camera low-altitude remote sensing images, and soil moisture obtained by drying measurement module SM OD ; Step S4 specifically includes:

[0034] S41. Collect N sets of data DATE_a to train a multi-layer feedforward neural network model, each set of data DATE_a includes: soil moisture collected by a time domain reflectometer at a certain time, backscatter values ​​of VV and VH polarizations obtained at the same time and standardized by a remote sensing data processing module, image incidence angle, and normalized vegetation index NDVI;

[0035] The standardization refers to standardizing the backscatter values ​​of all four features VV, VH polarization, image incident angle and NDVI by Z-score standardization method, and the standardization process is based on the following formula:

[0036]

[0037] Where X represents the initial eigenvalue; X Mean represents the average value; X Std represents the standard deviation of the eigenvalues;

[0038] S42. Collect M groups of data DATE_b to calculate coefficients a, b, c, d, e of the model based on environmental information, each group of data DATE_b includes: soil moisture collected by a time domain reflectometer at a certain time, net radiation NR, air humidity AH, and surface temperature LST obtained at the same time;

[0039] S43. Collect Z group data DATE_c to calculate SM FNN and SM NM Corresponding weights w1 and w2, each set of data DATE_c includes: soil moisture measured by drying of soil samples collected at a certain time SM OD, the standardized VV, VH polarized backscatter values, image incidence angle, normalized vegetation index NDVI collected at the same time, as well as the net radiation NR, air humidity AH and land surface temperature LST collected at the same time;

[0040] S5. Training the multilayer feedforward neural network model based on the data collected in step S4; specifically, training the weight of each neuron and the bias of each layer in the multilayer feedforward neural network model based on the N sets of data DATE_a, and then training to obtain a multilayer feedforward neural network model;

[0041] S6. Calculate the coefficients a, b, c, d, e of the model based on the environmental information based on the data collected in step S4; specifically, calculate the coefficients a, b, c, d, e of the model based on the environmental information based on the M set of data DATE_b, and then obtain the model based on the environmental information;

[0042] S7. Calculate the weights w1, w2 of the fusion calculation model based on the data collected in step S4; specifically, calculate based on the Z group data DATE_c SM FNN and SM NM Corresponding weights, based on the multi-layer feedforward neural network model obtained in step S5 and the model based on environmental information obtained in step S6, the soil moisture of group Z can be obtained from the group Z data DATE_c SM OD , SM FNN and SM NM , the fusion calculation model established in step S3 above can be used to obtain the values ​​of the Z group weights w1 and w2, and the average value of the Z group weights w1 and w2 is calculated. SM FNN and SM NM The corresponding weights w1, w2 are then used to obtain the fusion calculation model;

[0043] S8. Collecting data for soil moisture fusion calculation; Step S8 specifically includes:

[0044] S81. Collect the backscatter values ​​and image incident angle of VV and VH polarizations obtained by synthetic aperture radar, obtain the normalized vegetation index NDVI based on the low-altitude remote sensing image of the camera, and standardize them; input the standardized backscatter values ​​and image incident angle of VV and VH polarizations, and the normalized vegetation index NDVI into the multi-layer feedforward neural network model obtained in the above step S5 to obtain the estimated soil moisture SM FNN ;

[0045] S82. Collect net radiation NR, air humidity AH, and surface temperature LST, and input the net radiation NR, air humidity AH, and surface temperature LST into the model based on environmental information obtained in step S6 above to obtain estimated soil moisture SM NM ;

[0046] S83. The soil moisture obtained in step S81 above SM FNN and the soil moisture obtained in step S82 SM NM The fusion calculation model obtained in the above step S7 is input to obtain the soil moisture SM.

[0047] The present invention also includes a computer program product, which includes a computer program, and when the computer program is executed by a processor, the soil moisture estimation method can be implemented.

[0048] The present invention also includes a readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the soil moisture estimation method can be implemented.

[0049] The soil moisture estimation system and method of the present invention integrates a multi-layer feedforward neural network model, a model based on environmental information and a drying measurement method, thereby reducing the dependence on the accuracy of the neural network model while ensuring the accuracy of soil moisture estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a schematic diagram of the soil moisture estimation system of the present invention;

[0051] Figure 2 It is a schematic diagram of a multi-layer feedforward neural network model of the present invention;

[0052] Figure 3 This is a flow chart of the soil moisture estimation method of the present invention. DETAILED DESCRIPTION

[0053] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0054] like Figure 1 As shown, the soil moisture estimation system based on the feedforward neural network model of the present invention includes: a data acquisition module, a remote sensing data processing module, a multi-layer feedforward neural network model, a model based on environmental information and a fusion calculation module. The data acquisition module includes a time domain reflectometer (TDR), a remote sensing information acquisition module, an environmental information acquisition module and a drying measurement module. Among them, the soil moisture output by the multi-layer feedforward neural network model SMFNN and soil moisture from the model output based on environmental information SM NM Input to the fusion calculation module to obtain soil moisture SM; time domain reflectometer (TDR), remote sensing information acquisition module, remote sensing data processing module provide training data and input data for the multi-layer feedforward neural network model; time domain reflectometer (TDR) and environmental information acquisition module provide coefficient calculation data and input data for the model based on environmental information; drying measurement module is used to provide calculation data for each weight of the fusion calculation model in the fusion calculation module. The details are as follows.

[0055] 1. Time Domain Reflectometry (TDR)

[0056] The speed at which high-frequency electromagnetic waves propagate in the soil depends on the dielectric constant of the soil, and soil moisture greatly affects the dielectric constant of the soil. Therefore, the dielectric constant of the soil can be calculated based on the time it takes for electromagnetic waves to propagate in the soil. Then, the soil moisture content can be calculated using the empirical relationship between the soil dielectric constant and the soil volume moisture content. This is the principle of estimating soil moisture using time domain reflectometry.

[0057] 2. Remote sensing information acquisition module

[0058] The remote sensing information acquisition module includes synthetic aperture radar and camera. The synthetic aperture radar is carried on remote sensing satellites, such as Sentinel-1, Fengyun-3 and other remote sensing satellites; the camera is carried on drones. The remote sensing information acquisition module obtains the backscatter value and image incident angle of VV and VH polarization based on the synthetic aperture radar, and obtains the normalized vegetation index NDVI based on the low-altitude remote sensing image of the camera.

[0059] 3. Remote sensing data processing module

[0060] All four features VV, VH polarized backscatter values, image incidence angle, and NDVI are standardized by the Z-score standardization method. The standardization process is based on the following formula:

[0061]

[0062] Where X represents the initial eigenvalue; X Mean represents the average value; X Std Represents the standard deviation of the eigenvalues.

[0063] 4. Multilayer feedforward neural network model

[0064] The present invention estimates soil moisture based on a multi-layer feedforward neural network model. Figure 2As shown in FIG. 1 , the model includes an input layer, two hidden layers and an output layer. The input layer selects the backscatter values ​​of VV and VH polarization, the image incident angle and NDVI after the standardization processing by the remote sensing data processing module as input; the output layer is the estimated soil moisture SM FNN ; Each hidden layer has 4 neurons.

[0065] The present invention collects N groups of data DATE_a to train a multi-layer feedforward neural network model. Each group of data DATE_a includes: soil moisture collected by a time domain reflectometer at a certain time, backscatter values ​​of VV and VH polarizations obtained at the same time and standardized by a remote sensing data processing module, image incidence angle, and normalized vegetation index NDVI. Each group of data DATE_a is collected at different times.

[0066] The weight of each neuron and the bias of each layer in the multi-layer feedforward neural network model are trained based on N groups of data DATE_a, thereby obtaining a multi-layer feedforward neural network model.

[0067] 5. Environmental information collection module

[0068] Soil moisture is affected by environmental factors. Studies have shown that soil moisture has a certain correlation with environmental information such as net radiation, atmospheric humidity, and surface temperature. To this end, the present invention estimates soil moisture based on net radiation NR, air humidity AH, and surface temperature LST. The environmental information acquisition module includes a net radiation meter, a hygrometer, and a surface temperature inversion module, which are used to collect net radiation, atmospheric humidity, and surface temperature, respectively. The surface temperature inversion module is preferably based on the inversion calculation of unmanned aerial vehicle thermal infrared remote sensing data. The unmanned aerial vehicle thermal infrared remote sensing technology has the advantage of controllable time resolution. Combined with the net radiation and atmospheric humidity that can be collected in a timely manner, the environmental information collection of the present invention is highly real-time, which is convenient for experimenters to collect environmental information that matches the remote sensing information collection time.

[0069] 6. Models based on environmental information

[0070] The present invention also estimates soil moisture through a model based on environmental information. SM NM The model based on environmental information is expressed as follows:

[0071]

[0072] Among them, a, b, c, d, e are coefficients.

[0073] The present invention collects M groups of data DATE_b to calculate the above coefficients when establishing the above model. Each group of data DATE_b includes: soil moisture collected by a time domain reflectometer at a certain time, net radiation NR, air humidity AH, and surface temperature LST obtained at the same time. Each group of data DATE_b is collected at different times.

[0074] The collection of DATE_a and DATE_b can be carried out simultaneously to avoid repeated collection of soil moisture.

[0075] 7. Drying measurement module

[0076] The present invention also includes a drying measurement module, which includes a drying box, and the soil sample to be tested is weighed to obtain the sample wet weight. W WET Then the soil sample to be tested is completely dried in a drying oven to obtain the sample dry weight W DRY , drying measured soil moisture .

[0077] 8. Fusion computing module

[0078] The fusion calculation module obtains the final soil moisture SM fusion calculation model based on the multi-layer feedforward neural network model and the model based on environmental information:

[0079]

[0080] Among them, w1 and w2 are SM FNN and SM NM The corresponding weight.

[0081] The drying measurement method has high accuracy, and the fusion calculation module is based on the soil moisture obtained from the drying measurement module. SM OD Calculate w1 and w2 to obtain a fusion calculation model.

[0082] Specifically, Z groups of data DATE_c are collected to calculate when establishing the above fusion calculation model. SM FNN and SM NM Each set of data DATE_c includes: soil moisture obtained from the drying measurement module for soil samples collected at a certain time SM OD, the backscatter values ​​of VV and VH polarizations, image incidence angle, normalized vegetation index NDVI, net radiation NR, air humidity AH and surface temperature LST collected at the same time. Based on the above multi-layer feedforward neural network model and the model based on environmental information, the soil moisture of group Z can be obtained from the above group Z data DATE_c SM OD , SM FNN and SM NM , the values ​​of the Z group weights w1 and w2 can be obtained by bringing them into the above soil moisture SM fusion calculation model (i.e., each group weight includes a w1 value and a w2 value). The present invention takes the mean value of the Z group weights w1 and w2 as SM FNN and SM NM The corresponding weights w1 and w2 are obtained.

[0083] After the fusion calculation model is built, the backscatter values ​​of VV and VH polarization, image incidence angle, normalized vegetation index NDVI, net radiation NR, air humidity AH and surface temperature LST collected by the remote sensing information acquisition module and the environmental information acquisition module are input into the fusion calculation model to obtain the estimated soil moisture SM.

[0084] The present invention also implements a soil moisture estimation method, such as Figure 3 As shown, it includes the following steps:

[0085] S1. Configure a multi-layer feedforward neural network model, which includes an input layer, two hidden layers and an output layer. The input layer selects the backscatter values ​​of VV and VH polarization, image incident angle and NDVI after standardization by the remote sensing data processing module as input; the output layer is the estimated soil moisture SM FNN ; Each hidden layer has 4 neurons.

[0086] S2. Establishing a model based on environmental information, wherein the model based on environmental information is represented by the following formula:

[0087]

[0088] Where NR is the net radiation, AH is the air humidity, LST is the surface temperature, SM NM is the soil moisture estimated by the model based on environmental information, and a, b, c, d, and e are coefficients.

[0089] S3. Establish a fusion calculation model based on soil moisture SMFNN and soil moisture SM NM To obtain the final soil moisture SM, the fusion calculation model is expressed by the following formula:

[0090]

[0091] Among them, w1 and w2 are SM FNN and SM NM The corresponding weight.

[0092] S4. Data collection, including soil moisture collected by time domain reflectometer, VV, VH polarization backscattering values ​​and image incidence angles obtained by synthetic aperture radar, normalized vegetation index NDVI, net radiation NR, air humidity AH, surface temperature LST obtained by camera low-altitude remote sensing images, and soil moisture obtained by drying measurement module SM OD Step S4 specifically includes:

[0093] S41. Collect N sets of data DATE_a to train a multi-layer feedforward neural network model, each set of data DATE_a includes: soil moisture collected by a time domain reflectometer at a certain time, backscatter values ​​of VV and VH polarizations obtained at the same time and standardized by a remote sensing data processing module, image incidence angle, and normalized vegetation index NDVI. Each set of data DATE_a is collected at different times.

[0094] The standardization refers to standardizing the backscatter values ​​of all four features VV, VH polarization, image incident angle and NDVI by Z-score standardization method, and the standardization process is based on the following formula:

[0095]

[0096] Where X represents the initial eigenvalue; X Mean represents the average value; X Std Represents the standard deviation of the eigenvalues.

[0097] S42. Collect M sets of data DATE_b to calculate coefficients a, b, c, d, e of the model based on environmental information. Each set of data DATE_b includes: soil moisture collected by a time domain reflectometer at a certain time, net radiation NR, air humidity AH, and surface temperature LST obtained at the same time. Each set of data DATE_b is collected at different times.

[0098] S43. Collect Z group data DATE_c to calculate SMFNN and SM NM The corresponding weights w1 and w2. Each set of data DATE_c includes: the soil moisture measured by drying the soil samples collected at a certain time SM OD , the standardized VV, VH polarized backscatter values, image incidence angle, normalized vegetation index NDVI collected at the same time, as well as the net radiation NR, air humidity AH and land surface temperature LST collected at the same time.

[0099] S5. Training the multi-layer feedforward neural network model based on the data collected in step S4. Specifically, training the weight of each neuron and the bias of each layer in the multi-layer feedforward neural network model based on N sets of data DATE_a, and then obtaining the multi-layer feedforward neural network model.

[0100] S6. Calculate the coefficients of the model based on the environment information based on the data collected in step S4. Specifically, calculate the coefficients a, b, c, d, e of the model based on the environment information based on the M sets of data DATE_b, and then obtain the model based on the environment information.

[0101] S7. Calculate the weight of the fusion calculation model based on the data collected in step S4. Specifically, calculate based on the Z group data DATE_c SM FNN and SM NM Based on the multi-layer feedforward neural network model obtained in step S5 and the model based on environmental information obtained in step S6, the soil moisture of group Z can be obtained from the group Z data DATE_c. SM OD , SM FNN and SM NM , by bringing into the fusion calculation model established in step S3 above, we can obtain the values ​​of the Z group weights w1 and w2 (that is, each group weight includes a w1 value and a w2 value), and calculate the mean of the values ​​of the Z group weights w1 and w2 to obtain SM FNN and SM NM The corresponding values ​​of weights w1 and w2 are then used to obtain the fusion calculation model.

[0102] S8. Collect data for fusion calculation of soil moisture. Specifically include:

[0103] S81. Collect the backscatter values ​​and image incident angle of VV and VH polarizations obtained by synthetic aperture radar, obtain the normalized vegetation index NDVI based on the low-altitude remote sensing image of the camera, and standardize them. Input the standardized backscatter values ​​and image incident angle of VV and VH polarizations, and the normalized vegetation index NDVI into the multi-layer feedforward neural network model obtained in the above step S5 to obtain the estimated soil moisture. SM FNN .

[0104] S82. Collect net radiation NR, air humidity AH, and ground surface temperature LST. Input the net radiation NR, air humidity AH, and ground surface temperature LST into the model based on environmental information obtained in step S6 above to obtain the estimated soil moisture. SM NM .

[0105] S83. The soil moisture obtained in step S81 above SM FNN and the soil moisture obtained in step S82 SM NM The fusion calculation model obtained in the above step S7 is input to obtain the soil moisture SM.

[0106] The soil moisture estimation system and method of the present invention integrates a multi-layer feedforward neural network model, a model based on environmental information and a drying measurement method, thereby reducing the dependence on the accuracy of the neural network model while ensuring the accuracy of soil moisture estimation.

[0107] The present invention also includes a computer program product, which includes a computer program, and when the computer program is executed by a processor, the soil moisture estimation method can be implemented.

[0108] The present invention also includes a readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the soil moisture estimation method can be implemented.

[0109] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A soil moisture estimation system based on a feedforward neural network model, It is characterized in that The system includes a data acquisition module, a remote sensing data processing module, a multi-layer feedforward neural network model, a model based on environmental information and a fusion calculation module; the data acquisition module includes a remote sensing information acquisition module and an environmental information acquisition module, and the remote sensing information acquisition module collects four characteristic values, namely, backscatter values ​​of VV and VH polarizations, image incidence angle and normalized vegetation index NDVI; The environmental information collection module collects net radiation NR, air humidity AH, and ground surface temperature LST; The remote sensing data processing module standardizes the backscatter values ​​of VV and VH polarizations, the image incidence angle and NDVI collected by the remote sensing information collection module by the Z-score standardization method. The standardization process is based on the following formula: Where X represents the initial eigenvalue; X Mean represents the average value; X Std represents the standard deviation of the eigenvalues; The multi-layer feedforward neural network model includes an input layer, two hidden layers and an output layer, and the input layer selects the backscattering values ​​of VV and VH polarizations, the image incident angle and the normalized vegetation index NDVI after the standardization processing by the remote sensing data processing module as input; The output layer is the estimated soil moisture SM FNN , each hidden layer has 4 neurons; The model based on environmental information is expressed by the following formula: Where NR is the net radiation, AH is the air humidity, LST is the surface temperature, SM NM is the soil moisture estimated by the model based on environmental information, a, b, c, d, e are coefficients; The fusion calculation module includes a fusion calculation model, and the fusion calculation model is expressed by the following formula: Where SM is the soil moisture estimated by the system, and w 1， w 2 They are SM FNN and SM NM The corresponding weights; The system inputs the data collected by the remote sensing information collection module and standardized by the remote sensing data processing module into the multi-layer feedforward neural network model to obtain soil moisture SM FNN , the data collected by the environmental information collection module is input into the above-mentioned model based on environmental information to obtain soil moisture SM NM , the soil moisture SM FNN and soil moisture SM NM The soil moisture SM estimated by the system is obtained by inputting the above fusion calculation model.

2. The soil moisture estimation system according to claim 1, It is characterized in that The data acquisition module also includes a time domain reflectometer. Based on the data acquisition module and the remote sensing data processing module, N groups of data DATE_a are obtained to train the multi-layer feedforward neural network model. Each group of data DATE_a includes: soil moisture collected by the time domain reflectometer at a certain time, backscatter values ​​of VV and VH polarizations obtained at the same time and standardized by the remote sensing data processing module, image incident angle, and normalized vegetation index NDVI; based on the N groups of data DATE_a, the weight of each neuron in the multi-layer feedforward neural network model and the bias of each layer are trained to obtain the multi-layer feedforward neural network model.

3. The soil moisture estimation system according to claim 2, It is characterized in that Based on the time domain reflectometer and the environmental information acquisition module, M groups of data DATE_b are obtained to calculate the coefficients a, b, c, d, e of the model based on environmental information, each group of data DATE_b includes: soil moisture collected by the time domain reflectometer at a certain time, net radiation NR, air humidity AH, and surface temperature LST obtained at the same time; based on the M groups of data DATE_b, the coefficients a, b, c, d, e of the model based on environmental information are calculated, and then the model based on environmental information is obtained.

4. The soil moisture estimation system according to claim 3, It is characterized in that The data acquisition module also includes a drying measurement module, which obtains Z groups of data DATE_c based on the data acquisition module and the remote sensing data processing module to calculate SM FNN and SM NM The corresponding weight w 1 and w 2 Each set of data DATE_c includes: soil moisture measured by drying the soil sample collected at a certain time SM OD , the standardized VV, VH polarized backscatter values, image incidence angle, normalized vegetation index NDVI collected at the same time, and the net radiation NR, air humidity AH and surface temperature LST collected at the same time; based on the multi-layer feedforward neural network model and the model based on environmental information, the Z group of soil moisture is obtained from the above Z group of data DATE_c. SM OD , SM FNN and SM NM , into the fusion calculation model to obtain the Z group weight w 1 , w 2 The value of the weight w for the Z group 1 , w 2 The average value of SM FNN and SM NM The corresponding weight w 1 , w 2 The value of .

5. The soil moisture estimation system according to any one of claims 1 to 4, It is characterized in that The remote sensing information acquisition module includes a synthetic aperture radar and a camera. The synthetic aperture radar is carried on a remote sensing satellite, and the camera is carried on a drone. The remote sensing information acquisition module obtains the backscattering value and image incident angle of VV and VH polarizations based on the synthetic aperture radar, and obtains the normalized vegetation index NDVI based on the low-altitude remote sensing image of the camera.

6. The soil moisture estimation system according to any one of claims 1 to 4, It is characterized in that The environmental information acquisition module includes a net radiation meter, a hygrometer and a land surface temperature inversion module, which are used to collect net radiation NR, air humidity AH and land surface temperature LST respectively.

7. The soil moisture estimation system according to claim 4, It is characterized in that The drying measurement module includes a drying box, which weighs the soil sample to be tested to obtain the wet weight of the sample. W WET Then the soil sample to be tested is completely dried in a drying oven to obtain the sample dry weight W DRY , drying measured soil moisture .

8. The soil moisture estimation method of the soil moisture estimation system according to any one of claims 1 to 7, It is characterized in that The following steps are involved: S1. configuring a multi-layer feedforward neural network model, wherein the multi-layer feedforward neural network model includes an input layer, two hidden layers and an output layer, wherein the input layer selects the backscattering values ​​of VV and VH polarizations, the image incident angle and NDVI after being standardized by the remote sensing data processing module as input; The output layer is the estimated soil moisture SM FNN ; Each hidden layer has 4 neurons; S2. Establishing a model based on environmental information, wherein the model based on environmental information is represented by the following formula: Where NR is the net radiation, AH is the air humidity, LST is the surface temperature, SM NM is the soil moisture estimated by the model based on environmental information, a, b, c, d, e are coefficients; S3. Establish a fusion calculation model based on soil moisture SM FNN and soil moisture SM NM Obtain soil moisture SM, and the fusion calculation model is expressed by the following formula: Among them, w 1 ,w 2 They are SM FNN and SM NM The corresponding weights; S4. Data collection, including soil moisture collected by time domain reflectometer, VV, VH polarization backscattering values ​​and image incidence angles obtained by synthetic aperture radar, normalized vegetation index NDVI, net radiation NR, air humidity AH, surface temperature LST obtained by camera low-altitude remote sensing images, and soil moisture obtained by drying measurement module SM OD ; Step S4 specifically includes: S41. Collect N sets of data DATE_a to train a multi-layer feedforward neural network model, each set of data DATE_a includes: soil moisture collected by a time domain reflectometer at a certain time, backscatter values ​​of VV and VH polarizations obtained at the same time and standardized by a remote sensing data processing module, image incidence angle, and normalized vegetation index NDVI; The standardization refers to standardizing the backscatter values ​​of all four features VV, VH polarization, image incident angle and NDVI by Z-score standardization method, and the standardization process is based on the following formula: Where X represents the initial eigenvalue; X Mean represents the average value; X Std represents the standard deviation of the eigenvalues; S42. Collect M groups of data DATE_b to calculate coefficients a, b, c, d, e of the model based on environmental information, each group of data DATE_b includes: soil moisture collected by a time domain reflectometer at a certain time, net radiation NR, air humidity AH, and surface temperature LST obtained at the same time; S43. Collect Z group data DATE_c to calculate SM FNN and SM NM The corresponding weight w 1 and w 2 Each set of data DATE_c includes: soil moisture measured by drying the soil sample collected at a certain time SM OD , the standardized VV, VH polarized backscatter values, image incidence angle, normalized vegetation index NDVI collected at the same time, as well as the net radiation NR, air humidity AH and land surface temperature LST collected at the same time; S5. Training the multilayer feedforward neural network model based on the data collected in step S4; specifically, training the weight of each neuron and the bias of each layer in the multilayer feedforward neural network model based on the N sets of data DATE_a, and then training to obtain a multilayer feedforward neural network model; S6. Calculate the coefficients a, b, c, d, e of the model based on the environmental information based on the data collected in step S4; specifically, calculate the coefficients a, b, c, d, e of the model based on the environmental information based on the M set of data DATE_b, and then obtain the model based on the environmental information; S7. Calculate the weight w of the fusion calculation model based on the data collected in step S4 1 ,w 2 Specifically, based on the Z group data DATE_c calculation SM FNN and SM NM Corresponding weights, based on the multi-layer feedforward neural network model obtained in step S5 and the model based on environmental information obtained in step S6, the soil moisture of group Z can be obtained from the group Z data DATE_c SM OD , SM FNN and SM NM , by bringing it into the fusion calculation model established in step S3 above, we can get the Z group weight w 1 ,w 2 The value of the weight w for the Z group 1 ,w 2 The average value of SM FNN and SM NM The corresponding weight w 1 ,w 2 The value of , and then get the fusion calculation model; S8. Collecting data for soil moisture fusion calculation; Step S8 specifically includes: S81. Collect the backscatter values ​​and image incident angle of VV and VH polarizations obtained by synthetic aperture radar, obtain the normalized vegetation index NDVI based on the low-altitude remote sensing image of the camera, and standardize them; input the standardized backscatter values ​​and image incident angle of VV and VH polarizations, and the normalized vegetation index NDVI into the multi-layer feedforward neural network model obtained in the above step S5 to obtain the estimated soil moisture SM FNN ; S82. Collect net radiation NR, air humidity AH, and surface temperature LST, and input the net radiation NR, air humidity AH, and surface temperature LST into the model based on environmental information obtained in step S6 above to obtain estimated soil moisture SM NM ; S83. The soil moisture obtained in step S81 above SM FNN and the soil moisture obtained in step S82 SM NM The fusion calculation model obtained in the above step S7 is input to obtain the soil moisture SM.

Images