Wheat precision foliar fertilizer topdressing method based on multispectral unmanned aerial vehicle yield prediction

Abstract

The present application belongs to the technical field of unmanned aerial vehicle variable fertilization, and specifically discloses a wheat precision foliar fertilizer topdressing method based on multi-spectral unmanned aerial vehicle yield prediction, which comprises the following steps: in the booting stage and the grain filling stage, first, using the unmanned aerial vehicle carrying spectrum and RTK to obtain the canopy spectrum, combining the ground measured flag leaf SPAD value and leaf area index, and establishing a yield prediction model across the growth period; in the fertilization operation, the unmanned aerial vehicle obtains the spectrum in real time, inputs the above-mentioned agricultural parameters into the model after inversion, and obtains the estimated yield distribution map of the whole field; according to the yield, variable spraying is implemented: the foliar fertilizer is increased in the area with low estimated yield, and the foliar fertilizer is reduced or not applied in the area with high estimated yield. The present application can directly associate the fertilization decision with the yield formation process, realize the precision fertilization according to the need, significantly reduce the fertilizer consumption, reduce the risks such as leaf overgrowth, group shading and element imbalance caused by excessive foliar fertilizer, and has outstanding economic and environmental benefits.

Description

Technical Field

[0001] This invention relates to the field of variable fertilizer application technology using drones, and in particular to a method for precision foliar fertilizer application to wheat based on multispectral drone yield prediction. Background Technology

[0002] Wheat yield and quality are closely related to topdressing management, especially in dryland areas where irrigation is lacking, requiring increased fertilization to ensure wheat growth and development. Foliar fertilization is an important means of supplementing nutrients in the mid-to-late stages; however, traditional foliar fertilization often involves spraying a uniform dosage across the entire field, which has the following shortcomings: First, excessive application of foliar fertilizer in some areas with already high potential yields can lead to excessive growth of the upper and middle leaves of wheat, resulting in large, drooping leaves, increased canopy cover, insufficient light exposure of the lower and middle leaves, and a decrease in overall photosynthetic efficiency. This may, in turn, lead to delayed maturity and a reduction in thousand-grain weight.

[0003] Secondly, excessive amounts of nitrogen, potassium, and other elements entering the body through the leaves can disrupt the original elemental balance. For example, excessive application of urea or potassium dihydrogen phosphate to the leaves can induce nitrogen-potassium antagonism and calcium-potassium antagonism, inhibiting the absorption and transport of some micronutrients and reducing the actual utilization rate of fertilizers.

[0004] Third, ineffective fertilization not only increases fertilizer costs, but also causes excess nutrients to leach and volatilize into the environment, posing a risk of non-point source pollution.

[0005] While existing drone fertilization operations can achieve variable spraying by zone, their decision-making is mostly based on single indicators such as the Normalized Difference Vegetation Index (NDVI) and relies on experience. Without specific reference values, it is difficult to promote directly. Furthermore, there is a lack of quantitative models that link real-time spectral indices, key growth period agronomic parameters, and final yield. This results in a disconnect between fertilization decisions and the actual yield of wheat, making it impossible to achieve precise control of "supplementing only what is needed."

[0006] In agriculture, drone remote sensing, with its high spatiotemporal resolution, flexibility, and ease of use, can efficiently collect crop canopy data and assess crop growth status, such as water, nutrients, pests and diseases, and photosynthesis. Multispectral imaging technology is increasingly widely used. Compared to multispectral cameras, multispectral cameras can provide more detailed spectral information using visible and near-infrared spectra. Furthermore, multispectral cameras offer advantages such as convenience, cost-effectiveness, and ease of data processing. Vegetation indices, as important indicators reflecting crop growth status, are widely used in crop yield prediction. Nevertheless, crop yield prediction based on drone multispectral imagery still faces some challenges, particularly in improving prediction accuracy.

[0007] The development of machine learning technology has significantly improved the accuracy of crop yield prediction. Machine learning can analyze the nonlinear relationship between predictor and response variables, and generally outperforms traditional linear regression methods in terms of fitting accuracy. Models such as Long Short-Term Memory networks in deep learning can better handle time series data and effectively predict wheat yield. Researchers are now using wheat growth indices at different stages to predict wheat yield, making the data more realistic and reliable. Summary of the Invention

[0008] The purpose of this invention is to provide a method for precision foliar fertilizer application for wheat based on multispectral drone yield prediction, in order to solve the problems of weak correlation between fertilization and final yield target and inability to dynamically and precisely regulate foliar fertilizer supply in the prior art.

[0009] To achieve the above objectives, this invention provides a method for precision foliar fertilizer application of wheat based on multispectral unmanned aerial vehicle (UAV) yield prediction, comprising the following steps: S1. Construct a wheat yield prediction model spanning multiple growth stages; S2. Real-time acquisition of the canopy spectrum of the field to be operated, inversion to obtain the SPAD value and leaf area index (LAI) of each operation unit, input into the yield prediction model to generate a spatial distribution map of the estimated yield; S3. Compare the estimated yield with the preset target yield threshold to generate a variable fertilization prescription map; S4. The drone applies the appropriate type of foliar fertilizer to each growth stage according to the variable fertilization prescription map.

[0010] Preferably, S1 includes the following steps: S11. During the heading and grain-filling stages of wheat, canopy multispectral images of the experimental plots were collected using drones equipped with multispectral cameras and RTK positioning modules. At the same time, N calibration points were evenly distributed in the field, and the SPAD value and leaf area index (LAI) of each point at each stage were measured manually. After the wheat matured, the yield of each calibration point was measured separately to obtain grain yield data. S12. For canopy spectral images at different times, extract vegetation indices and use multiple regression or machine learning algorithms to establish a spectral inversion model that takes vegetation indices as input and outputs SPAD value and leaf area index (LAI) respectively. S13. Using the SPAD values ​​and leaf area index (LAI) of the calibrated sampling points during the booting and grain-filling stages as input variables and the yield at maturity as the output variable, a yield prediction model spanning the growth stages is constructed.

[0011] Preferably, in S12, the vegetation index includes one or more of NDVI, RVI, DVI, and red-edged chlorophyll index.

[0012] Preferably, in S13, a wheat yield prediction model spanning the growth period is constructed using partial least squares regression, random forest, or artificial neural network, and the optimal SPAD value and leaf area index (LAI) are determined through cross-validation.

[0013] Preferably, S2 includes the following steps: S21. During the heading and grain-filling stages of wheat, drones equipped with spraying systems, multispectral imagers, and RTK were used to conduct low-altitude remote sensing flights of 12m-130m to acquire real-time canopy spectral images of the fields to be treated. The spectral inversion model constructed in S12 was then used to calculate the SPAD value and leaf area index (LAI) of each treatment unit from the spectral images. S22. Input the calculated SPAD value and leaf area index (LAI) into the wheat yield prediction model constructed in S13 to generate a spatial distribution map of the estimated yield for the current period.

[0014] Preferably, in step S3, the estimated yield is compared with a preset target yield threshold to generate a variable fertilization prescription map: When the estimated yield is lower than the lower limit of the target yield, it is identified as a nutrient-deficient area, and the amount of foliar fertilizer spraying should be controlled and increased. When the estimated yield is higher than the upper limit of the target yield, it is identified as an area of ​​excessive growth, and foliar fertilizer spraying should be reduced or stopped. When the estimated yield is between the upper and lower limits of the target yield, maintain the baseline spraying rate.

[0015] Preferably, in S3, a fertilization adjustment coefficient is introduced. Quantify the deviation of the estimated yield from the target yield, based on the fertilizer adjustment coefficient. Numerical range matching-based differentiated fertilization control strategy; Fertilizer adjustment coefficient The calculation formula is as follows: ; in, To estimate production, Target output; when If a nutrient deficiency is identified, the fertilizer application rate should be increased from the baseline amount. And not exceeding 50% of the benchmark amount; when At that time, it was determined that the plant was growing excessively, and the amount of fertilizer was reduced. ; when At that time, maintain the baseline amount.

[0016] Preferably, in S3, the upper and lower limits of the target yield are preset based on the historical average yield of the production area where the field is located, the yield potential of the variety, and the soil fertility level, and are adjusted and manually modified before each operation.

[0017] Preferably, in S3, the foliar fertilizer sprayed during the booting stage is a phosphorus-potassium fertilizer and a boron fertilizer; the foliar fertilizer sprayed during the grain-filling stage is a potassium fertilizer and an anti-premature aging functional foliar fertilizer.

[0018] Therefore, the present invention employs the above-mentioned method for precision foliar fertilizer application of wheat based on multispectral UAV yield prediction, and the beneficial effects are as follows: (1) This invention quantitatively correlates the spectral data of UAVs, agronomic parameters across growth stages and final yield, and the pesticide application decision directly serves the yield target, realizing "fertilization based on yield" and making the basis for fertilization more scientific.

[0019] (2) By predicting yield, this invention identifies areas in the field that have the potential for high yield but do not require much fertilizer, and stops or reduces the supply of foliar fertilizer. Under the premise of ensuring yield potential, it can save 15% to 30% of the total amount of foliar fertilizer, directly reducing planting costs.

[0020] (3) This invention provides on-demand supplementation for areas with low yields, avoiding excessive topdressing across the entire field. It can effectively prevent problems such as excessive growth of functional leaves, severe shading of the plant population, increased plant respiration consumption, and imbalance of antagonistic elements caused by excessive foliar fertilizer, thus ensuring the efficiency of photosynthetic production and material transport in the later stages of wheat growth, thereby improving the grain setting rate and grain weight of wheat.

[0021] (4) The present invention utilizes RTK high-precision positioning throughout the process, which enables precise matching of multi-period spectral data with ground measurement points and subsequent fertilization points, ensuring the spatial accuracy of model calculation and prescription map generation, and providing support for centimeter-level variable spraying of UAVs.

[0022] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0023] Figure 1 This is an overall flowchart of an embodiment of the wheat precision foliar fertilizer application method based on multispectral drone yield prediction of the present invention; Figure 2 This is the fitting result of the optimal spectral index NDVI for SPAD during the booting stage of the present invention, based on the method of wheat precision foliar fertilizer application for yield prediction using multispectral drones. Figure 3 This is the fitting result of the optimal spectral index NDRE for the LAI (Lead Inlet Aperture) during the booting stage of the present invention, based on the wheat precision foliar fertilizer topdressing method for yield prediction using multispectral drones. Figure 4This is the result of fitting the optimal spectral index NDRE to the LAI (Labor Intake) during the grain-filling stage of the present invention, which is based on the wheat precision foliar fertilizer topdressing method for yield prediction using multispectral drones. Figure 5 This is the fitting result of the optimal spectral index NDVI for SPAD during the grain-filling stage of the present invention, based on the wheat precision foliar fertilizer topdressing method for yield prediction using multispectral drones. Detailed Implementation

[0024] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0025] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0026] like Figure 1 As shown, the method for precision foliar fertilizer application of wheat based on multispectral UAV yield prediction includes the following steps: S1. Construct a wheat yield prediction model spanning multiple growth stages, including the following steps: S11. Data Acquisition: During the wheat heading and grain-filling stages, canopy multispectral images of the experimental plots were collected using drones equipped with multispectral cameras and RTK positioning modules. Simultaneously, N calibration sampling points were evenly distributed in the field. The number of calibration sampling points could be determined based on the area size, approximately 10-20 per acre. The SPAD value and leaf area index (LAI) of each sampling point at each stage were measured manually. After the wheat matured, each calibration sampling point was harvested and yield measured separately to obtain grain yield data.

[0027] When collecting spectra and spraying, the drone maintains centimeter-level positioning through the RTK positioning module to ensure spatial consistency of spectra, agronomic parameters and spraying areas across multiple periods.

[0028] S12. Construct a spectral inversion model: Extract vegetation indices from canopy spectral images at different times, and use multiple regression or machine learning algorithms to establish a spectral inversion model that takes vegetation indices as input and outputs SPAD values ​​and leaf area index (LAI) respectively.

[0029] The vegetation index of this invention includes one or more of NDVI, RVI, DVI, and red-edged chlorophyll index.

[0030] S13. Using the SPAD values ​​and leaf area index (LAI) of the calibrated sampling points during the booting and grain-filling stages as input variables and the yield at maturity as the output variable, a yield prediction model spanning the growth stages is constructed.

[0031] This invention can use partial least squares regression, random forest or artificial neural network to construct a wheat yield prediction model across the growth period, and determine the optimal SPAD value and leaf area index (LAI) through cross-validation.

[0032] The process of precise application of foliar fertilizer: S2. Real-time acquisition of the canopy spectrum of the field to be operated on, inversion to obtain the SPAD value and leaf area index (LAI) of each operation unit, inputting them into the yield prediction model to generate a spatial distribution map of the estimated yield, including the following steps: S21. Real-time Remote Sensing and Parameter Inversion: During the wheat heading and grain-filling stages, drones equipped with spraying systems, multispectral imagers, and RTK were used to conduct low-altitude remote sensing flights (12m-130m) to acquire real-time canopy spectral images of the fields to be treated. The flight altitude was determined according to the size of the plot; the larger the plot, the higher the required flight altitude. The spectral inversion model constructed in S12 was then used to calculate the SPAD value and leaf area index (LAI) of each treatment unit from the spectral images. For example, the calculation formula for the grain-filling stage is as follows: ; ; NDVI, or Spectral Vegetation Index, is obtained by performing band calculations on multispectral data captured by drones. The results can be directly obtained from... Figure 5 Read from [the source]. Similarly, the red-edged vegetation index (NDRE) can be read from [the source]. Figure 4 Read from above; coefficients The data was fitted based on the sampled dataset obtained from the actual sampling points, such as... Figure 5 As shown ,like Figure 4 As shown .

[0033] S22. Input the calculated SPAD value and leaf area index (LAI) into the wheat yield prediction model constructed in S13 to generate a spatial distribution map of the estimated yield for the current period.

[0034] S3. Compare the estimated yield with the preset target yield threshold to generate a variable fertilization prescription map: When the estimated yield is lower than the lower limit of the target yield, it is identified as a nutrient-deficient area, and the amount of foliar fertilizer spraying is controlled and increased.

[0035] When the estimated yield exceeds the upper limit of the target yield, it is identified as an area of ​​excessive growth, and foliar fertilizer spraying should be reduced or stopped.

[0036] When the estimated yield is between the upper and lower limits of the target yield, the baseline spraying amount is maintained, thereby achieving precision fertilization based on yield potential.

[0037] In this invention, the upper and lower limits of the target yield are preset based on the historical average yield of the production area where the field is located, the yield potential of the variety, and the soil fertility level, and can be adjusted and manually modified before each operation.

[0038] Introducing fertilization adjustment coefficient Quantify the deviation of the estimated yield from the target yield, based on the fertilizer adjustment coefficient. The numerical range matching-based differentiated fertilization control strategy performs a secondary correction on the fertilization amount. The specific control principle is as follows: A simulated predicted yield is obtained from spectral data using a wheat yield prediction model. Then, a standard yield, such as the average yield of a normal year, is set as the benchmark target yield. When the predicted yield is lower than this benchmark target yield, foliar fertilizer is sprayed to supplement nutrients and increase future yield. The amount of foliar fertilizer sprayed depends on the difference; the lower the yield is from the standard, the higher the amount of foliar fertilizer to be sprayed.

[0039] In this invention, the foliar fertilizer sprayed during the booting stage is a phosphorus-potassium fertilizer and a boron fertilizer; the foliar fertilizer sprayed during the grain-filling stage is a potassium fertilizer and an anti-premature aging functional foliar fertilizer.

[0040] Fertilizer adjustment coefficient The calculation formula is as follows: ; in, To estimate production, The target output.

[0041] when If a nutrient deficiency is identified, the fertilizer application rate should be increased from the baseline amount. And not exceeding 50% of the benchmark amount.

[0042] when At that time, it was determined that the plant was growing excessively, and the amount of fertilizer was reduced. (The lowest possible value is 0).

[0043] when At that time, maintain the baseline amount.

[0044] S4. The drone applies the appropriate type of foliar fertilizer to each growth stage according to the variable fertilizer prescription map, and performs targeted spraying to achieve precise targeted spraying.

[0045] Example 1: In Li Village, a city in XX province, an experimental demonstration field was selected.

[0046] I. Constructing a production forecasting model: During the wheat heading stage (from the appearance of the flag leaf to before heading) and the grain-filling stage (10-15 days after flowering), on sunny, cloudless days with winds less than level 3, between 10:00 and 14:00, a DJI Mavic 3M RTK drone equipped with a multispectral camera (green light 560nm±16nm, red light 650nm±16nm, red edge 730nm±16nm, and near-infrared 860nm±26nm) and an RTK positioning module was used to acquire hyperspectral images of the experimental field canopy with a flight altitude of 20m, a forward overlap of 80%, and a lateral overlap of 70%. Calibration was performed using a whiteboard set up on the ground.

[0047] On the day of the first spectral acquisition, 50 fixed sampling points were evenly marked in the field (precise coordinates were recorded using RTK, with each quadrat having a side length of 1m). The SPAD value of the flag leaf (the first leaf measured from the greening to jointing stage) representing the plants in each quadrat was measured using a SPAD-502 chlorophyll meter, and the leaf area index was measured using a LAI-2200 canopy analyzer. After the flight during the grain-filling stage, the above parameters were measured again simultaneously in the same quadrat. At the full maturity stage of wheat, the wheat was harvested, threshed, dried, and weighed in separate quadrats to obtain the actual yield.

[0048] After background segmentation of the multispectral images, candidate vegetation indices, such as NDVI, RVI, MCARI, NDRE, and LCI, are extracted for corresponding pixels of each time period. Pearson correlation analysis is then used to screen the spectral indices for fitting.

[0049] Multiple linear regression (MLR) was used to establish prediction models for SPAD values ​​and leaf area index (LAI) for two different periods.

[0050] First, spectral inversion models for SPAD values ​​and leaf area index (LAI) at two different time periods are constructed. Taking experimental data from Limou Village as an example, such as... Figure 2 and Figure 3 The figures show the fitting results of SPAD during the booting stage using the optimal spectral index NDVI and the fitting results of LAI during the booting stage using the optimal spectral index NDRE, respectively. Figure 4 , Figure 5 The figures show the fitting results of LAI during the grouting period using the optimal spectral index NDRE and the fitting results of SPAD during the grouting period using the optimal spectral index NDVI, respectively.

[0051] Then, two multiple linear regression yield prediction models for different periods were constructed: the linear prediction models for leaf area index (LAI) and SPAD values ​​for the two periods were substituted into the spectral data for calculation, and then the LAI and SPAD values ​​calculated from the spectra were fitted with the measured yield values ​​using multiple linear regression to obtain yield prediction models for different periods, which were used to construct the basis for variable fertilization data. The final regression model is as follows: Spike incubation period: ; in, Indicates the final yield during the booting stage; The simulated leaf area index represents the heading stage; This represents the simulated chlorophyll index during the booting stage.

[0052] Grouting period: ; in, The grouting period is the final output period; The simulated leaf area index during the grouting period; This represents the simulated chlorophyll index during the grouting period.

[0053] This model characterizes the contribution of comprehensive spectral parameters of plant growth at different stages to the final yield. Although the multiple linear regression model is relatively simple, it is easy to visualize, apply, and promote.

[0054] II. Implementation of Precision Fertilization In the same wheat field, agricultural drones (such as DJI T40 / T50 modified with variable sprayers and equipped with multispectral cameras) are deployed at the same two growth stages to carry out operations.

[0055] After takeoff, the multispectral camera acquires images in real time. The airborne edge computing module or ground station calls the pre-loaded PLSR inversion model to invert the SPAD value and leaf area index (LAI) of each grid (1m×1m) in real time. Then, it is input into the yield prediction model to obtain the yield prediction map.

[0056] Set floating thresholds based on the target yield for the field (set at 400 kg / mu): the lower limit is 300 kg / mu, and the upper limit is 500 kg / mu. Patches with an estimated yield below 300 kg / mu are marked orange (over-application area), and the spraying amount is increased by 20%; patches with an estimated yield above 500 kg / mu are marked blue (no or minimal application area), and the amount is reduced by 100% (i.e., shut off) or maintained at 5% for only wetting; for yields between 400-520 kg / mu, spray at the normal baseline amount of 150 ml / mu.

[0057] The drone's spraying system adjusts the solenoid valve duty cycle and flow rate in real time according to the prescription map, with a flight speed of 5m / s and a spray width of 6m, precisely targeting each work area. A mixture of potassium dihydrogen phosphate and boron fertilizer is sprayed during the heading stage; potassium dihydrogen phosphate is sprayed during the grain-filling stage with the addition of brassinolide to prevent premature aging; all stages are based on the same yield prediction logic and threshold determination.

[0058] Two consecutive years of trial demonstrations have shown that, compared with the traditional method of uniformly spraying foliar fertilizer, this method reduces the total amount of foliar fertilizer used by 23.7%, maintains stable grain yield without significant yield reduction, reduces the lodging area in the field by more than 40%, decreases the incidence of leaf diseases (such as powdery mildew and leaf rust) in the middle and late stages, and increases the thousand-grain weight by 1.2-2.0g.

[0059] Therefore, this invention adopts the above-mentioned wheat precision foliar fertilizer topdressing method based on multispectral UAV yield prediction. It collects field data through a multispectral UAV with an RTK positioning module, and constructs a yield prediction model by combining measured SPAD and leaf area index. According to the estimated yield and the target yield threshold, the amount of foliar fertilizer sprayed is dynamically adjusted to achieve yield-based fertilizer determination and variable precision topdressing, which can save fertilizer, reduce losses, improve quality and efficiency.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for precision foliar fertilizer application of wheat based on multispectral unmanned aerial vehicle (UAV) yield prediction, characterized in that, Includes the following steps: S1. Construct a wheat yield prediction model spanning multiple growth stages; S2. Real-time acquisition of the canopy spectrum of the field to be operated, inversion to obtain the SPAD value and leaf area index (LAI) of each operation unit, input into the yield prediction model to generate a spatial distribution map of the estimated yield; S3. Compare the estimated yield with the preset target yield threshold to generate a variable fertilization prescription map; S4. The drone applies the appropriate type of foliar fertilizer to each growth stage according to the variable fertilization prescription map.

2. The method for precision foliar fertilizer application of wheat based on multispectral UAV yield prediction according to claim 1, characterized in that, S1 includes the following steps: S11. During the heading and grain-filling stages of wheat, canopy multispectral images of the experimental plots were collected using drones equipped with multispectral cameras and RTK positioning modules. At the same time, N calibration points were evenly distributed in the field, and the SPAD value and leaf area index (LAI) of each point at each stage were measured manually. After the wheat matured, the yield of each calibration point was measured separately to obtain grain yield data. S12. For canopy spectral images at different times, extract vegetation indices and use multiple regression or machine learning algorithms to establish a spectral inversion model that takes vegetation indices as input and outputs SPAD value and leaf area index (LAI) respectively. S13. Using the SPAD values ​​and leaf area index (LAI) of the calibrated sampling points during the booting and grain-filling stages as input variables and the yield at maturity as the output variable, a yield prediction model spanning the growth stages is constructed.

3. The method for precision foliar fertilizer application of wheat based on multispectral UAV yield prediction according to claim 2, characterized in that, In S12, vegetation indices include one or more of NDVI, RVI, DVI, and red-edged chlorophyll index.

4. The method for precision foliar fertilizer application of wheat based on multispectral UAV yield prediction according to claim 2, characterized in that, In S13, partial least squares regression, random forest, or artificial neural network are used to construct a wheat yield prediction model across the growth period, and the optimal SPAD value and leaf area index (LAI) are determined through cross-validation.

5. The method for precision foliar fertilizer application of wheat based on multispectral UAV yield prediction according to claim 4, characterized in that, S2 includes the following steps: S21. During the heading and grain-filling stages of wheat, drones equipped with spraying systems, multispectral imagers, and RTK are used to conduct low-altitude remote sensing flights of 12m-130m to acquire real-time canopy spectral images of the fields to be treated. The spectral inversion model constructed in S12 is called to solve the spectral images into SPAD values ​​and leaf area index (LAI) for each treatment unit. S22. The calculated SPAD values ​​and LAI are input into the wheat yield prediction model constructed in S13 to generate a spatial distribution map of the estimated yield for the current period.

6. The method for precision foliar fertilizer application of wheat based on multispectral UAV yield prediction according to claim 5, characterized in that, In S3, the estimated yield is compared with the preset target yield threshold to generate a variable fertilizer prescription map: When the estimated yield is lower than the lower limit of the target yield, it is identified as a nutrient-deficient area, and the amount of foliar fertilizer spraying should be controlled and increased. When the estimated yield is higher than the upper limit of the target yield, it is identified as an area of ​​excessive growth, and foliar fertilizer spraying should be reduced or stopped. When the estimated yield is between the upper and lower limits of the target yield, maintain the baseline spraying rate.

7. The method for precision foliar fertilizer application of wheat based on multispectral UAV yield prediction according to claim 6, characterized in that, In S3, a fertilization adjustment coefficient is introduced. Quantify the deviation of the estimated yield from the target yield, based on the fertilizer adjustment coefficient. Numerical range matching-based differentiated fertilization control strategy; Fertilizer adjustment coefficient The calculation formula is as follows: ； in, To estimate production, Target output; when If a nutrient deficiency is identified, the fertilizer application rate should be increased from the baseline amount. And not exceeding 50% of the benchmark amount; when At that time, it was determined that the plant was growing excessively, and the amount of fertilizer was reduced. ; when At that time, maintain the baseline amount.

8. The method for precision foliar fertilizer application of wheat based on multispectral UAV yield prediction according to claim 6, characterized in that, In S3, the target yield upper and lower limits are preset based on the historical average yield of the production area where the field is located, the yield potential of the variety, and the soil fertility level, and are adjusted and manually modified before each operation.

9. The method for precision foliar fertilizer application of wheat based on multispectral UAV yield prediction according to claim 5, characterized in that, In S3, the foliar fertilizer sprayed during the booting stage is phosphorus and potassium fertilizer and boron fertilizer; the foliar fertilizer sprayed during the grain filling stage is potassium fertilizer and anti-premature aging functional foliar fertilizer.

Images