Soybean yield remote sensing estimation method and device based on dual drive of data and knowledge
Through the remote sensing estimation method of soybean yield driven by data and knowledge, remote sensing data and recurrent neural network model are used to combine crop growth models to build an agronomic knowledge base, solving the problem of poor crop yield estimation effect in the existing technology, and achieving high precision and efficient yield estimation.
Patent Information
- Application Number
- CN202410904052.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-08
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-07-08
AI Technical Summary
In the prior art, the estimation effect of crop yield is poor, the data quality and model representativeness of a large number of training samples are dependent on, and it consumes a lot of manpower and material resources during large-scale applications, making it difficult to expand.
Using the remote sensing estimation method of soybean yield based on dual driving data and knowledge, the leaf area index is determined by obtaining the remote sensing data of crops, and the yield yield is estimated using the recurrent neural network model. Combined with the crop growth model, various growth scenarios are simulated to construct yield formation. Agronomy knowledge base for processes.
It improves the accuracy and efficiency of crop yield estimation, simplifies the estimation process, reduces the dependence on ground observation samples, and is suitable for large-area applications.
Smart Images

Figure CN118446559B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of agricultural production technology, and in particular to a soybean yield remote sensing estimation method and device based on dual drive of data and knowledge. Background Art
[0002] Early crop yield estimation models were mainly driven by data. Some machine learning models under data-driven backgrounds, such as distributed gradient boosting models, random forest models, support vector machine models, and artificial neural networks, have been widely used in crop yield remote sensing estimation applications. An empirical statistical model is constructed by combining crop growth state parameters inverted by remote sensing (such as Leaf Area Index (LAI), the proportion of photosynthetically active radiation absorbed by the canopy, and various vegetation indices (VIS)) with crop yield.
[0003] However, the accuracy of yield prediction by empirical statistical models depends on the data quality of a large number of training samples and the representativeness of the model. Although remote sensing data provides certain data support for the construction of yield estimation models, the accuracy of crop yield estimation is always limited and the estimation effect is poor. Furthermore, with the continuous in-depth understanding of the crop growth process, crop growth models driven by agronomic mechanism knowledge predict the evolution of crops from sowing to harvesting by simulating photosynthesis, gas exchange between the canopy and the atmosphere, phenology, soil moisture and temperature changes, biomass accumulation and grain formation. The prediction accuracy is compared with empirical statistical models, but the actual promotion and application of crop growth models depends on a large number of ground observation samples, consumes a lot of manpower and material resources, and is difficult to apply on a large scale. Summary of the invention
[0004] In view of the problems existing in the prior art, the present invention provides a soybean yield remote sensing estimation method and device based on dual drive of data and knowledge, so as to solve the problem of poor crop yield estimation effect in the prior art, so as to achieve high-precision estimation of crop yield.
[0005] In a first aspect, the present invention provides a soybean yield remote sensing estimation method based on dual drive of data and knowledge, comprising: acquiring remote sensing data of crops, and determining a first average leaf area index and a second average leaf area index of the crops through the remote sensing data, wherein the first average leaf area index is the average leaf area index from seedling to flowering stage, and the second average leaf area index is the average leaf area index from flowering to maturity stage; inputting the first average leaf area index and the second average leaf area index of the crops into a crop yield estimation model to obtain the yield of the crops output by the crop yield estimation model; the crop yield estimation model is obtained by training a recurrent neural network based on the first average leaf area index, the second average leaf area index and the yield of multiple crop samples.
[0006] Optionally, the first average leaf area index, the second average leaf area index and the yield per unit area of the multiple crop samples are obtained by simulating a plurality of different growth scenarios with a crop growth model, each growth scenario corresponding to a set of scenario parameter combinations, and the scenario parameter combination includes a combination of meteorological parameters, crop parameters, soil parameters and management measure parameters.
[0007] Optionally, the input of the crop growth model is meteorological parameters, crop parameters, soil parameters and management measure parameters, and the output is leaf area index, date, development stage, total aboveground dry weight, storage organ dry weight, leaf dry weight, stem weight, root weight, transpiration rate, actual root depth, actual root zone soil moisture content and total water content in the soil profile.
[0008] Optionally, the acquiring of remote sensing data of crops includes: determining the maturity type of the crops based on the accumulated temperature zone; determining a first time range from the seedling stage to the flowering stage and a second time range from the flowering stage to the maturity stage of the crops based on the maturity type of the crops; and acquiring corresponding remote sensing data based on the first time range and the second time range.
[0009] Optionally, the recurrent neural network is a gated recurrent unit model.
[0010] Optionally, determining the first average leaf area index and the second average leaf area index of the crop through the remote sensing data includes: obtaining the first average leaf area index and the second average leaf area index of the crop based on inverting surface reflectivity data in the remote sensing data.
[0011] In a second aspect, the present invention provides a soybean yield remote sensing estimation device based on dual drive of data and knowledge, the device comprising: a first processing module, used to obtain remote sensing data of crops, and determine a first average leaf area index and a second average leaf area index of the crops through the remote sensing data, the first average leaf area index being the average leaf area index from seedling to flowering stage, and the second average leaf area index being the average leaf area index from flowering to maturity stage; a second processing module, used to input the first average leaf area index and the second average leaf area index of the crop into a crop yield estimation model, to obtain the yield of the crop output by the crop yield estimation model; wherein the crop yield estimation model is obtained by training a recurrent neural network based on the first average leaf area index, the second average leaf area index and the yield of multiple crop samples.
[0012] 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, it implements any of the above-mentioned methods for remote sensing estimation of soybean yield based on dual drive of data and knowledge.
[0013] 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 above-described methods for remote sensing estimation of soybean yield based on dual drive of data and knowledge.
[0014] In a fifth aspect, the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned methods for remote sensing estimation of soybean yield based on dual drive of data and knowledge.
[0015] The soybean yield remote sensing estimation method based on dual drive of data and knowledge provided by the present invention obtains the remote sensing data of the crop, determines the first average leaf area index and the second average leaf area index of the crop, and inputs the first average leaf area index and the second average leaf area index into the crop yield estimation model obtained by the recurrent neural network model, so as to realize the estimation of the crop yield. The problem of poor crop yield estimation effect in the prior art is solved. By using the leaf area index as the input indicator, the crop yield estimation process is simplified with the help of the recurrent neural network model, and the estimation accuracy is improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the technical solutions in the present invention or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. 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.
[0017] Figure 1 It is a flow chart of the soybean yield remote sensing estimation method based on dual drive of data and knowledge provided by the present invention.
[0018] Figure 2 It is a schematic diagram of the GRU model structure provided by the present invention.
[0019] Figure 3 It is a structural schematic diagram of the WOFOST model provided by the present invention.
[0020] Figure 4 It is a schematic diagram of a process for estimating crop yields based on a hybrid modeling method driven by both data and knowledge provided by the present invention.
[0021] Figure 5 This is a schematic diagram of soybean LAI simulation provided by the present invention.
[0022] Figure 6 This is a simulated diagram of the dry weight of soybean storage organs provided by the present invention.
[0023] Figure 7 This is a single-point scale soybean yield estimation accuracy verification chart for 2022-2023 provided by the present invention.
[0024] Figure 8 This is a regional-scale soybean yield estimation accuracy verification chart for 2019-2022 provided by the present invention.
[0025] Fig. 9 It is a structural schematic diagram of the soybean yield remote sensing estimation device based on dual drive of data and knowledge provided by the present invention.
[0026] Fig.10 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION
[0027] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments in 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. The above method provided by the present invention is illustrated below by examples of specific application scenarios.
[0028] Figure 1 Schematic diagram of the process of soybean yield remote sensing estimation method based on dual drive of data and knowledge provided by the present invention, such as Figure 1 As shown, the method includes the following steps S102-S104:
[0029] Step S102: Acquire remote sensing data of crops, and determine a first average leaf area index and a second average leaf area index of the crops through the remote sensing data, wherein the first average leaf area index is the average leaf area index from seedling emergence to flowering stage, and the second average leaf area index is the average leaf area index from flowering to maturity stage.
[0030] It should be noted that the crop described in this application is soybean.
[0031] It should be noted that the Leaf Area Index (LAI) is an important indicator for measuring the degree of vegetation coverage, which refers to the vegetation leaf area per unit projected area on unit ground.
[0032] In an exemplary embodiment, the acquisition of crop remote sensing data may be achieved by following steps S11-S13:
[0033] Step S11: determining the maturity type of the crop according to the accumulated temperature zone.
[0034] It should be noted that due to climate differences in different regions, the research area is divided according to the accumulated temperature zone in practice. The maturity types of soybeans include: early maturity, medium early maturity, medium maturity, medium late maturity, and late maturity. According to the division of accumulated temperature zones, combined with literature data, it is determined which type of soybeans should be planted in different accumulated temperature zones.
[0035] Step S12: Determine a first time range from the emergence to the flowering stage and a second time range from the flowering to the maturity stage of the crop according to the maturity type of the crop.
[0036] It should be noted that the stage division of crops is mainly defined by effective accumulated temperature. Effective accumulated temperature is the accumulation of daily average temperature above the base temperature of the crop. When the effective accumulated temperature reaches the accumulated temperature threshold required for a growth stage, the growth of the crop will enter the next growth stage. The time of the crop growth stage is usually expressed in calendar days. The time range is determined by converting calendar time into thermal time. The length and duration of the crop growth stage will be adjusted according to the temperature conditions in different years. The calculation formula for effective accumulated temperature is as follows:
[0037]
[0038] Where T is the average daily temperature. is the lower limit of the critical temperature for development. It is the upper limit temperature for soybean development.
[0039] Step S13: Acquire corresponding remote sensing data based on the first time range and the second time range.
[0040] As an optional embodiment, the time range used by crops to reach different growth stages is mainly based on the ERA5-Land Daily Aggregated - ECMWF ClimateReanalysis data provided by the GEE (Google Earth Engine) platform, which provides daily average temperature data at an altitude of 2 meters above the surface from 1950 to the present.
[0041] In an exemplary embodiment, determining the first average leaf area index and the second average leaf area index of the crop by using the remote sensing data can be achieved by the following steps S21:
[0042] Step S21: obtaining a first average leaf area index and a second average leaf area index of the crop based on inversion of the surface reflectance data in the remote sensing data.
[0043] As an optional embodiment, the remote sensing data is usually Sentinel-2 Level-2A data obtained by searching, processing and downloading through the GEE (Google Earth Engine) platform, and the LA2-level data is the surface reflectance data that has completed radiation calibration and atmospheric correction.
[0044] As an optional embodiment, when performing inversion based on surface reflectivity data, the The vegetation index is implemented, where is defined as:
[0045]
[0046] In the above formula, NIR is the reflectivity of the near-infrared band (B8A) of the Sentinel-2 image, and RE is the reflectivity of the red edge band (B5, 704nm) of the Sentinel-2 image.
[0047] Further, based on The LAI of soybean can be further expressed as:
[0048]
[0049] Step S104: inputting the first average leaf area index and the second average leaf area index of the crop into a crop yield estimation model to obtain the yield of the crop output by the crop yield estimation model.
[0050] The crop yield estimation model is obtained by training a recurrent neural network based on the first average leaf area index, the second average leaf area index and the yield of multiple crop samples.
[0051] It should be noted that the recurrent neural network is a gated recurrent unit model. The gated recurrent unit (GRU) is mainly used to solve problems such as long-term memory and gradients in back propagation. GRU merges the gate structure of LSTM and improves the complex unit structure of the previous recurrent neural network. It can increase the training speed of the network and ensure that the training accuracy of the model will not decrease. Compared with the "three-gate" structure of LSTM, GRU only has two gate structures, the update gate and the reset gate, to control the flow of long-term information. The network structure is simplified and the training speed of the model can be increased. The calculation process of the GRU model is:
[0052]
[0053] Among them, sigmoid() is the S-type activation function; tanh() represents the hyperbolic tangent activation function; represents the output of the reset gate at time t; represents the output of the update gate at time t; represents the weight matrix of the reset gate; represents the weight matrix of the update gate; The weight matrix representing the candidate hidden state; Represents the input of the GRU model at time t; Represents the hidden layer state output of the GRU model at time t; represents the candidate hidden state of the current input, Represents the hidden layer state output of the GRU model at time t-1.
[0054] It should be noted that the present invention uses the average LAI of crops from the emergence stage to the flowering stage (0 < DVS ≤ 1) and from the flowering stage to the maturity stage (1 < DVS ≤ 2) as the temporal input features of the model, and the yield per unit area of crops as the output feature of the model.
[0055] As an alternative embodiment, during the process of training the model, the model optimizer is Adam, and the loss function is mean squared error. The network is trained multiple times to find the optimal values of parameters such as the number of GRU layers, the number of neurons units, epochs, batch size, dropout, etc. As Figure 2 shown, it is a schematic diagram of the GRU model structure provided by the present invention.
[0056] It should be noted that the first average leaf area index, the second average leaf area index, and the yield per unit area of the multiple crop samples are obtained by simulating various growth scenarios through a crop growth model. Each growth scenario corresponds to a set of scenario parameter combinations, and the scenario parameter combinations include combinations of meteorological parameters, crop parameters, soil parameters, and management measure parameters.
[0057] As an alternative embodiment, the crop growth model can select the World Food Studies Model (abbreviated as WOFOST). Figure 3 It is a schematic diagram of the WOFOST model provided by the present invention. As shown in the figure, the model mainly includes four modules: meteorology, crops, soil, and field management. With the corresponding meteorological conditions, crop varieties, soil parameters, and field management measures as the constraint mechanisms, the growth and development processes of various crops are simulated. The inputs of the crop growth model are meteorological parameters, crop parameters, soil parameters, and management measure parameters, and the outputs are leaf area index, date, development stage, total aboveground dry weight, storage organ dry weight, leaf dry weight, stem weight, root weight, transpiration rate, actual root depth, actual soil water content in the root zone, and total soil water content in the soil profile.
[0058] It should be noted that the meteorological parameters include: date (DAY, d), incident shortwave radiation (IRRAD, KJ / m2 / day), average water vapor pressure (VAP, kPa), maximum temperature (TMAX, °C), minimum temperature (TMIN, °C), average wind speed (WIND, m / s), precipitation (RAIN, mm), and snow depth (SNOWDEPTH, cm).
[0059] It should be noted that the crop parameters include: low temperature threshold for emergence, high temperature threshold for emergence, accumulated temperature from sowing to emergence, accumulated temperature from emergence to flowering, accumulated temperature from flowering to maturity, initial total dry weight of crops, maximum growth rate of leaf area index, specific leaf area (DVS=0.0), specific leaf area (DVS=0.45), specific leaf area (DVS=0.90), specific leaf area (DVS=2.00), leaf life at 35 degrees Celsius, low temperature threshold of leaf age, extinction coefficient of visible light diffusion as a function of DVS, light energy utilization rate of a single leaf as a function of daily average temperature, maximum leaf Assimilation rate (DVS=0.0), maximum leaf Assimilation rate (DVS=1.7), maximum leaf Assimilation rate (DVS=2.0), assimilate conversion efficiency of leaves, assimilate conversion efficiency of storage organs, assimilate conversion efficiency of roots, assimilate conversion efficiency of stems, relative variable of respiration rate when temperature increases by 10℃, maintenance respiration rate of leaves, maintenance respiration rate of storage organs, maintenance respiration rate of roots, maintenance respiration rate of stems, ratio of total dry matter to roots as a function of DVS, ratio of aboveground dry matter to leaves as a function of DVS, ratio of aboveground dry matter to stems as a function of DVS, ratio of aboveground dry matter to organs as a function of DVS, maximum relative mortality of leaves caused by water stress, initial root depth, maximum daily increase in root depth, maximum root depth.
[0060] It should be noted that soil parameters mainly include the wilting coefficient (SMW), saturated water content (SM0) and field holding capacity (SMFCF), and the values of these parameters mainly depend on the soil texture and structure.
[0061] It should be noted that management measure parameters include planting date, etc.
[0062] As an optional embodiment, the present invention can widely set the four types of input data based on the "lookup table" concept to generate numerous growth scenarios through their permutations and combinations, and use the model to simulate the crop growth and development and yield formation process under each growth scenario one by one, thereby constructing a huge agronomic knowledge base for the yield formation process. The huge agronomic knowledge base for the yield formation process constructed through crop samples solves the problem that the existing estimation method relies on a large number of ground observation samples, and the acquisition of observation samples requires a lot of manpower and material resources. The constructed agronomic knowledge base for the yield formation process can be more suitable for large-scale crop yield estimation scenarios.
[0063] It should be noted that the storage organ dry weight on the last day of the growth period is used to characterize the crop yield per unit area in this application.
[0064] As an optional embodiment, after obtaining the crop yield per unit area output by the yield estimation model, the estimation accuracy can be further verified. The estimation accuracy verification includes two aspects: actual sample point verification and statistical data verification. Among them, the actual sample point verification data comes from sample data collected in the field, and the accuracy verification indicators include the determination coefficient (Coefficient of Determination), root mean square error RMSE (Root Mean Square Error) and mean relative error MRE (Mean Relative Error). The higher it is, the smaller the RMSE and MRE are, which means the performance of the yield estimation model is better. The calculation formula is as follows:
[0065]
[0066] in, and Represent the actual yield and estimated yield, respectively. is the mean of the actual yield.
[0067] Through the above steps S102-S104, by acquiring the remote sensing data of the crop, the first average leaf area index and the second average leaf area index of the crop are determined, and the first average leaf area index and the second average leaf area index are input into the crop yield estimation model obtained by the recurrent neural network model, so as to realize the estimation of the crop yield. The problem of poor crop yield estimation effect in the prior art is solved, and by using the leaf area index as an input indicator, the crop yield estimation process is simplified with the help of the recurrent neural network model, and the estimation accuracy is improved.
[0068] The executor of each step in the method can be a soybean yield remote sensing estimation device based on dual drive of data and knowledge. The device can be implemented through software and / or hardware. The device can be integrated in an electronic device. The electronic device can be a terminal device (such as a smart phone, a personal computer, etc.), or a server (such as a local server or a cloud server, or a server cluster, etc.), or a processor, or a chip, etc.
[0069] Obviously, the above-described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. In order to better understand the above method, the above process is described below in conjunction with the embodiments, but the technical solutions of the embodiments of the present invention are not limited. Specifically:
[0070] Based on agronomic mechanisms, the present invention couples machine learning methods with crop growth models to establish a hybrid modeling method driven by data and knowledge to estimate crop yields. Figure 4This is a flow chart of the hybrid modeling method based on data and knowledge driven to estimate crop yields. For ease of understanding, the present invention takes soybean as an example. The present invention solves the problems of sample scarcity and poor crop yield estimation in yield estimation research, and expands the spatiotemporal generalization ability of the yield estimation model.
[0071] 1. Construction of agronomic knowledge base on soybean yield formation process
[0072] WOFOST can quantitatively simulate the crop growth process under the influence of meteorological and other environmental factors with a day-long step, and the model's description of the process is universal, and different crops in different geographical locations can be simulated by changing parameters. The WOFOST model is based on the simulation of crop physiological and ecological processes such as assimilation, respiration, transpiration, and dry matter distribution, mainly including the simulation of crop growth under potential growth conditions, water-limited conditions, and nutrient-limited conditions. Taking the Northeast Black Soil Region as an example, the model parameters are set as follows.
[0073] (1) Meteorological parameters
[0074] The meteorological parameters required by the WOFOST model include date (DAY, d), incident shortwave radiation (IRRAD, KJ / m2 / day), average vapor pressure (VAP, kpa), maximum temperature (TMAX, ℃), minimum temperature (TMIN, ℃), average wind speed (WIND, m / s), precipitation (RAIN, mm) and snow depth (SNOWDEPTH, cm). The meteorological observation dataset used in the study comes from the meteorological website, and 319 meteorological stations in the Northeast Black Soil Region are selected, with a time range of 1980-2023. In order to further screen meteorological stations, based on the 2023 soybean planting distribution map, a total of 51 meteorological stations within 1 km of the soybean planting plot were selected to provide meteorological input data for the model. The meteorological observation dataset contains daily observation data, including: maximum temperature (℃), average temperature (℃), minimum temperature (℃), average wind speed (m / sec), precipitation (mm), average vapor pressure (kPa) and sunshine hours (h).
[0075] (2) Crop parameters
[0076] The crop variety parameters required by the WOFOST model are shown in Table 1. Since the growth process of soybeans is regulated by accumulated temperature, the present invention has graded the accumulated temperature from emergence to flowering (TSUM1) and the accumulated temperature from flowering to maturity (TSUM2) of soybeans according to existing research, and divided the soybeans planted in the Northeast Black Soil Region into five maturity types: early, medium-early, medium, medium-late and late. Among them, early-maturing soybeans require less accumulated temperature to reach different growth stages, and late-maturing soybeans have a greater demand for accumulated temperature. For other non-sensitive crop parameters, such as the low temperature threshold (TBASEM) of soybean emergence, the high temperature threshold (TEFFMX), the initial crop total dry weight (TDWI), and the specific leaf area (SLATB), they are mainly set with reference to the soybean default parameter values provided by the WOFOST model and the calibration values of related studies in the same study area.
[0077] Table 1 WOFOST model crop parameter settings
[0078]
[0079] Table 1. WOFOST model crop parameter settings
[0080]
[0081] Table 1. WOFOST model crop parameter settings
[0082]
[0083] Table 1. WOFOST model crop parameter settings
[0084]
[0085] (3) Soil parameters
[0086] The main soil parameters required by the WOFOST model include the wilting coefficient (SMW), saturated water content (SM0) and field water holding capacity (SMFCF), etc. The values of these parameters mainly depend on soil texture and structure. It mainly includes soil spatial distribution, soil physical properties, soil chemical properties and soil nutrient data. The soil types in the Northeast Black Soil Region are divided into four types: sandy loam, light loam, medium loam and heavy loam. The parameter values of different soil types are shown in Tables 2 and 3. The parameters involved mainly include the wilting coefficient (SMW), field water holding capacity (SMFCF), saturated water content (SM0), saturated hydraulic conductivity (K0), critical air content of soil ventilation (CRAIRC), deep seedling bed first layer topsoil seepage parameters (SPADS), deep seedling bed second layer topsoil seepage parameters (SPODS), shallow seedling bed first layer topsoil seepage parameters (SPASS), shallow seedling bed second layer topsoil seepage parameters (SPOSS), etc. The values of other parameters refer to the existing literature research and the default parameter values provided by the WOFOST model.
[0087] Table 2 Main soil parameter settings of WOFOST model
[0088]
[0089] Table 3 Main soil parameter settings of WOFOST model (continued)
[0090]
[0091] (4) Management measures parameters
[0092] The WOFOST model is based on the simulation of crop physiological and ecological processes such as assimilation, respiration, transpiration and dry matter distribution, mainly including the simulation of crop growth under potential growth conditions, water limitation conditions and nutrient limitation conditions. Since soybeans planted in the black soil region of Northeast China are basically rain-fed, the study used the water stress model to simulate the soybean growth process. According to the ground survey results, four soybean sowing dates were set for crop growth simulation, namely April 20, April 30, May 10 and May 20.
[0093] In order to simulate the growth process of soybeans under various agricultural scenarios and solve the problem of sample dependence of existing models, a wide range of settings for four types of input data were made based on the "lookup table" idea, and their arrangement and combination generated many growth scenarios. The model was used to simulate the growth and yield formation process of soybeans under each growth scenario one by one, with a total of more than 150,000 (38 51 4 5 4) soybean planting scenarios and finally constructed a large multi-scenario simulation data set (as shown in Table 4).
[0094] Table 4 WOFOST model simulation scenarios
[0095]
[0096] Further, as shown in Table 5, the input simulation parameters of the model mainly include date (day), development stage (DVS), leaf area index (LAI), total aboveground dry weight (TAGP), storage organ dry weight (TWSO), leaf dry weight (TWLV), stem weight (TWST), root weight, transpiration rate (TRA), actual root depth (RD), actual root zone soil moisture content (SM) and soil profile total water content (WWLOW). After the parameters are given, the WOFOST model can realize continuous simulation of the crop growth process with a step length of days starting from the emergence of crops, wherein the storage organ dry weight on the last day of the growth period is the soybean yield of concern in the present invention.
[0097] Table 5 Simulation parameters output by WOFOST model
[0098]
[0099] 2. Intelligent Modeling of Soybean Yield Estimation
[0100] The present invention uses a recurrent neural network GRU model to perform intelligent modeling for soybean yield estimation. The calculation process of the GRU model is:
[0101]
[0102] Among them, sigmoid() is the S-type activation function; tanh() represents the hyperbolic tangent activation function; represents the output of the reset gate at time t; represents the output of the update gate at time t; represents the weight matrix of the reset gate; represents the weight matrix of the update gate; The weight matrix representing the candidate hidden state; Represents the input of the GRU model at time t; Represents the hidden layer state output of the GRU model at time t; represents the candidate hidden state of the current input, Represents the hidden layer state output of the GRU model at time t-1.
[0103] During the model construction process, considering that the WOFOST model divides the growth stages of crops into three stages: emergence, flowering, and maturity, the average LAI (LAImean) of soybeans from the emergence to the flowering stage (0 < DVS ≤ 1) and from the flowering to the maturity stage (1 < DVS ≤ 2) is used as the temporal input feature of the model, and the yield per unit area of soybeans is used as the output feature of the model. The simulated dataset is divided into training samples and test samples according to a ratio of 9:1. The model optimizer is Adam (Adaptive Moment Estimation), and the loss function is the mean squared error. The network is trained multiple times to find the optimal values of parameters such as the number of GRU layers, the number of neurons units, epochs, batch size, and dropout.
[0104] III. Remote Sensing Estimation of Soybean Yield Based on Multi-Scenario Simulated Dataset
[0105] Due to climate differences in different regions, it is necessary to divide regions according to accumulated temperature and then download remote sensing images accordingly. Taking the Northeast Black Soil Region as an example, referring to the accumulated temperature zone division standard in Heilongjiang Province in the 1990s, using the active accumulated temperature with the daily average temperature ≥ 10°C as the heat resource index, based on the daily average temperature data of meteorological stations in the Northeast Black Soil Region from 1981 to 2019, adopting the five-day moving average method, with a range of 200 °C·d, and using the empirical frequency method to take values with an 80% guarantee rate, the Northeast Black Soil Region is divided into 10 accumulated temperature zones.
[0106] Referring to the division of soybean planting maturity types on different accumulated temperature belts in Heilongjiang Province, the present invention sets five different soybean planting maturity types: early maturity, medium-early maturity, medium maturity, medium-late maturity, and late maturity for different accumulated temperature zones in the Northeast Black Soil Region, and thus calculates the phenological periods of soybeans in different soybean maturity type planting areas. Among them, provinces such as Jilin and Liaoning mainly plant medium maturity, medium-late maturity, and late maturity type soybeans, and the northern region is more inclined to plant early maturity and medium-early maturity type soybeans.
[0107] The division of different growth stages of soybeans is mainly judged by the effective accumulated temperature ( ), and the calculation formula of the effective accumulated temperature is as follows:
[0108]
[0109] where T is the daily average temperature. is the lower limit of the development critical temperature, is the upper limit temperature of soybean development. Referring to the setting of soybean variety parameters in the WOFOST model when constructing the simulation data set, the accumulated temperature thresholds corresponding to the growth stages for different soybean planting maturity types are shown in Table 6, where TSUM is the effective accumulated temperature required for soybean from sowing to germination, TSUM1 is the effective accumulated temperature required for soybean from germination to flowering, and TSUM2 is the effective accumulated temperature required for soybean from flowering to maturity. According to the results of field surveys, the soybean sowing time in Heilongjiang Province and Inner Mongolia Autonomous Region was uniformly set to May 5 in the study, while the soybean sowing time in Jilin Province and Liaoning Province was earlier and uniformly set to May 1. The soybean emergence time in the Northeast Black Soil Region is set to no later than June 1, and the maturity time is set to no later than October 1.
[0110] Table 6 Accumulated temperature thresholds corresponding to different soybean maturity types
[0111]
[0112] The calculation of soybean phenology is mainly based on the ERA5-LandDaily Aggregated - ECMWF Climate Reanalysis data provided by the GEE (Google Earth Engine) platform. Based on the calculation results of soybean phenological period, the study used the K-means clustering method (K-means) to divide the phenological extraction results into 10 categories, obtained remote sensing images for different regions, inverted the soybean LAI, and completed the data download.
[0113] 4. Results and Analysis
[0114] 1. Analysis of simulated data sets
[0115] Figure 5 This is a schematic diagram of soybean LAI simulation provided by the present invention. Figure 6 This is a simulation diagram of the dry weight of soybean storage organs provided by the present invention. It can be seen from the results that the diversified setting of parameters provides rich simulation results for the simulation. The growth process of soybeans under different production scenarios is also different. The simulation data provides a wide range of variability, ensuring the completeness of the training sample data set and the representativeness of the soybean growth conditions in the target area, providing data support for the subsequent intelligent modeling of yield estimation.
[0116] 2. Yield accuracy evaluation
[0117] The single-point scale accuracy of the model prediction results was verified using the ground-measured soybean yield data in 2022 and 2023. Figure 7This is a single-point scale soybean yield estimation accuracy verification diagram for 2022-2023 provided by the present invention. It can be seen from the results that the prediction model achieves a relatively accurate estimation of soybean yield at a single-point scale. The accuracy verification results show that the determination coefficient is 0.73, which passes the significance test (P<0.01, where P is used to characterize probability), reaching an extremely significant level, with a root mean square error of 288.72 kg / ha, an average relative error of 10.06%, and an overall prediction accuracy of more than 85%. The model prediction results were verified for regional scale accuracy using municipal soybean yield statistical data from 2019 to 2022, such as Figure 8 As shown, it is a regional scale soybean yield estimation accuracy verification diagram provided by the present invention from 2019 to 2022. The overall estimation accuracy determination coefficient of the model from 2019 to 2022 is 0.62, the root mean square error is 272.36 kg / ha, and the average relative error is 12.08%.
[0118] In general, the pixel-scale yield estimation not only obtains yield information at a finer scale, but also performs well at the municipal scale. In addition, this method can realize dynamic and flexible yield estimation based on the available data, showing great potential in large-scale soybean yield estimation.
[0119] The present invention is driven by meteorological data, crop varieties, soil parameters and management measures data, and uses the WOFOST model to simulate various scenarios of soybean planting, so as to realize the construction of an agronomic knowledge base for the soybean yield formation process. The simulation scenarios are rich, the simulation process is continuous, and the simulation results are scientific and reasonable, which provides agronomic knowledge support and sufficient sample data for the intelligent modeling of yield estimation. Based on the soybean multi-scenario simulation data set, the deep learning method is used to explore the complex relationship between the key indicator factors of soybean yield formation and the yield, and the intelligent modeling of yield estimation is realized. The modeling does not rely on ground measured samples, which provides theoretical guidance for the realization of early high-precision yield estimation, agricultural production and agricultural decision-making. At the same time, the study area is divided into accumulated temperature zones, and different soybean planting maturity types are set. Combined with medium and high-resolution remote sensing data, meteorological data, statistical yields, and measured yield data, the hybrid modeling method proposed in the present invention is used to carry out pixel-scale soybean yield remote sensing estimation, and the demonstration application and accuracy evaluation of the method are realized. The results show that the prediction model achieves high yield estimation accuracy at both single point and regional scales. The accuracy of the model was evaluated using the measured yield data of the Northeast Black Soil Region in 2022 and 2023. The coefficient of determination was 0.73, which passed the significance test (P<0.01) and reached an extremely significant level. The root mean square error was 288.72 kg / ha, the average relative error was 10.06%, and the overall prediction accuracy was higher than 85%. Compared with the municipal statistical data, the overall performance of the Northeast Black Soil Region's yield estimation is , RMSE =272.36 kg / ha, MRE = 12.08%, passing the significance test (P<0.01) and reaching an extremely significant level. The yield estimation model showed a high spatiotemporal continuity in regional scale demonstration applications, and was able to achieve large-scale, precise, and flexible yield estimation.
[0120] In short, the present invention proposes a hybrid modeling method that combines remote sensing data, crop growth models and deep learning to estimate soybean yield at the pixel scale. The method is supported by agronomic mechanisms and gives full play to the advantages of process model mechanisms and data mining advantages of deep learning. The research method meets the requirements of agronomic knowledge support and intelligent level for yield estimation modeling. Combined with high-resolution remote sensing images, it can achieve a relatively ideal soybean yield estimation performance at the pixel scale, providing a new idea and method for crop yield estimation. It solves the problem of poor crop yield estimation in the existing technology under the premise of insufficient ground samples.
[0121] The following is a description of the soybean yield remote sensing estimation device based on dual drive of data and knowledge provided by the present invention. The soybean yield remote sensing estimation device based on dual drive of data and knowledge described below and the soybean yield remote sensing estimation method based on dual drive of data and knowledge described above can be referenced to each other.
[0122] Fig. 9 Schematic diagram of the structure of the soybean yield remote sensing estimation device based on dual drive of data and knowledge provided by the present invention. Fig. 9 As shown, the device comprises:
[0123] The first processing module 902 is used to obtain remote sensing data of crops, and determine a first average leaf area index and a second average leaf area index of the crops through the remote sensing data, wherein the first average leaf area index is an average leaf area index from seedling to flowering stage, and the second average leaf area index is an average leaf area index from flowering to maturity stage;
[0124] The second processing module 904 is used to input the first average leaf area index and the second average leaf area index of the crop into the crop yield estimation model to obtain the yield of the crop output by the crop yield estimation model; wherein the crop yield estimation model is obtained by training a recurrent neural network based on the first average leaf area index, the second average leaf area index and the yield of multiple crop samples.
[0125] By using the above-mentioned soybean yield remote sensing estimation device based on dual drive of data and knowledge, the first average leaf area index and the second average leaf area index of the crop are determined by acquiring the remote sensing data of the crop, and the first average leaf area index and the second average leaf area index are input into the crop yield estimation model obtained by the recurrent neural network model, so as to realize the estimation of the crop yield. The problem of poor crop yield estimation effect in the prior art is solved. By using the leaf area index as the input indicator, the crop yield estimation process is simplified with the help of the recurrent neural network model, and the estimation accuracy is improved.
[0126] Optionally, the first average leaf area index, the second average leaf area index and the yield per unit area of the multiple crop samples are obtained by simulating a plurality of different growth scenarios with a crop growth model, each growth scenario corresponding to a set of scenario parameter combinations, and the scenario parameter combination includes a combination of meteorological parameters, crop parameters, soil parameters and management measure parameters.
[0127] Optionally, the input of the crop growth model is meteorological parameters, crop parameters, soil parameters and management measure parameters, and the output is leaf area index, date, development stage, total aboveground dry weight, storage organ dry weight, leaf dry weight, stem weight, root weight, transpiration rate, actual root depth, actual root zone soil moisture content and total water content in the soil profile.
[0128] Optionally, the first processing module 902 is also used to determine the maturity type of the crop based on the accumulated temperature zone; determine a first time range from the seedling stage to the flowering stage and a second time range from the flowering stage to the maturity stage of the crop based on the maturity type of the crop; and obtain corresponding remote sensing data based on the first time range and the second time range.
[0129] Optionally, the recurrent neural network is a gated recurrent unit model.
[0130] Optionally, the first processing module 902 is further used to invert the first average leaf area index and the second average leaf area index of the crop based on the surface reflectance data in the remote sensing data.
[0131] It should be noted here that the above-mentioned device provided by the present invention can implement all the method steps implemented by the above-mentioned method embodiment, and can achieve the same technical effect. The parts and beneficial effects that are the same as the method embodiment in this embodiment will not be described in detail here.
[0132] Fig.10 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Fig.10As shown, the electronic device may include: a processor 1010, a communications interface 1020, a memory 1030 and a communication bus 1040, wherein the processor 1010, the communications interface 1020 and the memory 1030 communicate with each other through the communication bus 1040. The processor 1010 may call the logic instructions in the memory 1030 to execute the soybean yield remote sensing estimation method based on dual drive of data and knowledge.
[0133] In addition, the logic instructions in the above-mentioned memory 1030 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 such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for 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.
[0134] It should be noted here that the electronic device provided by the present invention can implement all the method steps implemented by the above-mentioned method embodiment, and can achieve the same technical effects. The parts and beneficial effects of this embodiment that are the same as the method embodiment will not be described in detail here.
[0135] 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 soybean yield remote sensing estimation method based on dual drive of data and knowledge provided by the above-mentioned methods.
[0136] It should be noted here that the non-transitory computer-readable storage medium provided by the present invention can implement all the method steps implemented by the above-mentioned method embodiment, and can achieve the same technical effect. The parts and beneficial effects that are the same as the method embodiment in this embodiment will not be described in detail here.
[0137] The present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the soybean yield remote sensing estimation method based on dual data and knowledge drive provided by the above-mentioned methods.
[0138] It should be noted here that the computer program product provided by the present invention can implement all the method steps implemented by the above-mentioned method embodiment, and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as those of the method embodiment will not be described in detail here.
[0139] The device embodiments described above are merely illustrative, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules 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. Those of ordinary skill in the art may understand and implement it without creative effort.
[0140] 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.
[0141] 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 soybean yield remote sensing estimation method based on dual drive of data and knowledge, characterized in that: include: Acquire remote sensing data of crops, and determine a first average leaf area index and a second average leaf area index of the crops through the remote sensing data, wherein the first average leaf area index is an average leaf area index from seedling emergence to flowering stage, and the second average leaf area index is an average leaf area index from flowering to maturity stage; Inputting the first average leaf area index and the second average leaf area index of the crop into a crop yield estimation model to obtain the yield of the crop output by the crop yield estimation model; The crop yield estimation model is obtained by training a recurrent neural network based on a first average leaf area index, a second average leaf area index and a yield of a plurality of crop samples; The first average leaf area index, the second average leaf area index and the yield per unit area of the multiple crop samples are obtained by simulating a plurality of different growth scenarios with a crop growth model, and each growth scenario corresponds to a set of scenario parameter combinations, and the scenario parameter combination includes a combination of meteorological parameters, crop parameters, soil parameters and management measure parameters.
2. The soybean yield remote sensing estimation method based on dual drive of data and knowledge according to claim 1 is characterized in that: The input of the crop growth model is meteorological parameters, crop parameters, soil parameters and management measure parameters, and the output is leaf area index, date, development stage, total aboveground dry weight, storage organ dry weight, leaf dry weight, stem weight, root weight, transpiration rate, actual root depth, actual root zone soil moisture content and total soil profile moisture.
3. The soybean yield remote sensing estimation method based on dual drive of data and knowledge according to claim 1 is characterized in that: The step of obtaining remote sensing data of crops comprises: Determining the maturity type of the crop according to the accumulated temperature zone; Determining a first time range from the emergence to the flowering stage and a second time range from the flowering to the maturity stage of the crop according to the maturity type of the crop; Corresponding remote sensing data is acquired based on the first time range and the second time range.
4. The soybean yield remote sensing estimation method based on dual drive of data and knowledge according to any one of claims 1 to 3, characterized in that: The recurrent neural network is a gated recurrent unit model.
5. The soybean yield remote sensing estimation method based on dual drive of data and knowledge according to claim 1 or 3, characterized in that: Determining the first average leaf area index and the second average leaf area index of the crop by using the remote sensing data comprises: A first average leaf area index and a second average leaf area index of the crop are obtained by inverting the surface reflectance data in the remote sensing data.
6. A soybean yield remote sensing estimation device based on dual drive of data and knowledge, characterized in that: The device comprises: A first processing module is used to obtain remote sensing data of crops, and determine a first average leaf area index and a second average leaf area index of the crops through the remote sensing data, wherein the first average leaf area index is an average leaf area index from seedling to flowering stage, and the second average leaf area index is an average leaf area index from flowering to maturity stage; A second processing module is used to input the first average leaf area index and the second average leaf area index of the crop into a crop yield estimation model to obtain the yield of the crop output by the crop yield estimation model; The crop yield estimation model is obtained by training a recurrent neural network based on a first average leaf area index, a second average leaf area index and a yield of a plurality of crop samples; The first average leaf area index, the second average leaf area index and the yield per unit area of the multiple crop samples are obtained by simulating a plurality of different growth scenarios with a crop growth model, and each growth scenario corresponds to a set of scenario parameter combinations, and the scenario parameter combination includes a combination of meteorological parameters, crop parameters, soil parameters and management measure parameters.
7. An electronic 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 program, it implements the soybean yield remote sensing estimation method based on dual drive of data and knowledge as described in any one of claims 1 to 5.
8. 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 soybean yield remote sensing estimation method based on dual drive of data and knowledge as described in any one of claims 1 to 5 is implemented.
9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the soybean yield remote sensing estimation method based on dual drive of data and knowledge as described in any one of claims 1 to 5 is implemented.
Citation Information
Patent Citations
Crop yield per unit area remote sensing estimation method, device and system
CN108985588A
Soybean growth vigour predication method based on satellite remote sensing data
CN109187356A
Multi-time-phase remote sensing inversion method of soybean biomass in black soil area by introducing topographic factor
CN109325433A
Cross-scale high-precision dynamic crop growth monitoring and yield assessment method based on high-resolution remote sensing data and a crop model
CN109829234A
Multi-scale crop yield evaluation method and system
CN111898922A
