An alfalfa yield estimation method based on multispectral and physically constrained sample enhancement

Abstract

The application focuses on the field of agricultural informatization and remote sensing application technology, and particularly relates to a kind of alfalfa yield estimation method based on multispectral and physical constraint sample enhancement. The method first acquires unmanned aerial vehicle multispectral image data corresponding to the alfalfa growth period, and lays out field quadrats with equal area in the image coverage area, collects measured data of alfalfa hay yield, leaf area index, chlorophyll content and leaf equivalent water thickness, and constructs an initial sample set. Then, input each parameter data within a reasonable range, generate predicted multispectral samples that meet the physical consistency constraint using the radiation transfer model, to expand the sample space. The measured samples and predicted samples are fused to construct an alfalfa hay yield estimation training data set. Based on the training data set, a mapping model between unmanned aerial vehicle multispectral features and alfalfa hay yield is established, and the alfalfa hay yield is estimated through machine learning.

Description

Technical Field

[0001] This invention relates to the fields of agricultural informatization and agricultural remote sensing technology, specifically to a method for calculating alfalfa yield based on multispectral and physical constraint sample enhancement. Background Technology

[0002] Alfalfa (Medicago sativa) is an important perennial forage crop, and its yield level directly affects the supply of forage and the efficiency of livestock production. Alfalfa is characterized by its long growth cycle, strong regeneration ability, and multiple harvests. Its canopy structure and biomass vary significantly between different growth stages and different harvesting cycles, which brings considerable uncertainty to yield calculation.

[0003] Existing methods for estimating alfalfa yield mainly rely on manual ground sampling or empirical statistical models, which suffer from problems such as high workload and insufficient spatial representativeness. In recent years, UAV multispectral remote sensing technology has been gradually applied to crop growth monitoring and yield estimation due to its advantages such as high spatiotemporal resolution and rich spectral information. However, existing alfalfa yield estimation methods based on UAV multispectral data mostly rely on empirical regression or purely data-driven models, which are highly sensitive to the number and distribution of samples. When the sample size is insufficient or the growth period varies greatly, the model stability and generalization ability are poor.

[0004] Furthermore, some studies have attempted to introduce radiative transfer models to simulate crop canopy spectra, but these mostly focus on biomass or leaf area index inversion, and often use physical models as the core inversion step, resulting in complex model structures and high engineering application difficulty. Therefore, there is an urgent need for a yield measurement method that is practical for alfalfa production and takes into account both physical constraints and engineering operability. Summary of the Invention

[0005] To overcome the numerous shortcomings of current alfalfa yield measurement technologies, this invention provides an alfalfa yield calculation method based on multispectral and physically constrained sample enhancement. The method first acquires UAV multispectral image data corresponding to the alfalfa growth stage, and then establishes field plots of equal area within the image coverage area to collect measured data on alfalfa hay yield, leaf area index, chlorophyll content, and equivalent leaf water thickness, constructing an initial sample set. Subsequently, within a reasonable range, the data for each parameter are input, and a radiative transfer model is used to generate predicted multispectral samples that satisfy physical consistency constraints, thereby expanding the sample space. The measured samples and predicted samples are then fused to construct a training dataset for alfalfa hay yield calculation. Based on this training dataset, a mapping model between UAV multispectral features and alfalfa hay yield is established to realize the calculation of alfalfa hay yield. This invention, by introducing a physically constrained sample enhancement mechanism, improves the stability and applicability of alfalfa hay yield calculation under limited sample quantity conditions, and has good prospects for widespread application.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention provides a method for calculating alfalfa yield based on multispectral and physical constraint sample enhancement: During the alfalfa growing season, multispectral remote sensing images of the alfalfa canopy at different harvesting times were obtained using drones equipped with multispectral cameras. Radiometric calibration and geometric correction were performed on the multispectral remote sensing images to extract reflectance information for each spectral band, and the Normalized Difference Vegetation Index (NDVI), Normalized Red Edge Index (NDRE), and Normalized Green Difference Vegetation Index (GNDVI) were calculated. Field plots of equal area were set up within the coverage area of ​​the UAV multispectral imagery to measure alfalfa hay yield, leaf area index, chlorophyll content, and equivalent leaf water thickness data, and to construct a measured sample set. Based on the structural characteristics of alfalfa canopy, a radiative transfer model is used to generate predicted multispectral samples that meet physical consistency constraints, which are then used to expand the sample space. By integrating measured and predicted samples, a training dataset for calculating alfalfa hay yield was constructed. Based on the training dataset, a mapping model between the multispectral features of UAVs and alfalfa hay yield is established, and the alfalfa hay yield calculation results are output through machine learning.

[0007] As a further limitation of the first aspect of the present invention, the following details are added when processing and analyzing the measured sample set data and images captured by the UAV multispectral camera: The alfalfa growth period includes the greening stage, the budding stage, and the initial flowering stage; The UAV multispectral remote sensing image data includes blue light, green light, red light, red edge, and near-infrared bands; The different harvesting cycles include two of the first, second, and subsequent regeneration cycles, with an interval of 25 to 35 days between each cycle; The radiometric calibration method is based on a combination of reflectivity calibration plates and airborne illumination information. It uses sensor calibration parameters to convert raw digital values ​​into radiance for the corresponding band, eliminating the influence of different sensor responses and imaging parameters. Subsequently, during flight or before and after takeoff and landing, images of the calibration plate with known reflectance are acquired. Combined with the solar altitude angle and irradiance information recorded by the airborne illumination sensor, the radiance is further converted into surface reflectance. This reduces the impact of differences in illumination intensity and imaging time on multispectral data, unifying multispectral image data acquired at different times and in different sorties onto a consistent reflectance scale. This gives each band of data a clear physical meaning, allowing it to be stably used for vegetation index calculations and maintaining physical consistency with the simulated spectral data generated by the radiative transfer model. This provides a reliable data foundation for subsequent alfalfa hay yield calculations and physical constraint sample enhancement. The geometric correction is based on UAV POS data and uses the direct georegistration method for coarse correction. The image pixels in the UAV body coordinate system are initially mapped to the geodetic coordinate system through a coordinate transformation model. Specifically, an affine transformation model is adopted, using position and attitude parameters from the POS data to calculate the image exterior orientation elements (such as principal point coordinates, focal length, flight altitude, and attitude angle), establishing a preliminary correspondence between pixel coordinates and geodetic coordinates. This corrects the overall offset caused by UAV flight attitude fluctuations (pitch and roll deviations) and flight altitude changes, ensuring that the error between pixel and ground position is controlled within 1-2 pixels. Simultaneously, using the deployed ground control points as a reference, a polynomial correction model is used for fine correction to eliminate residual geometric distortion (including lens distortion, terrain undulation distortion, and distortion caused by minor flight attitude fluctuations). The model uses a 2nd-3rd order polynomial (adaptively selected according to the degree of image distortion; a 2nd order polynomial is used in flat terrain areas of alfalfa fields, and a 3rd order polynomial is used in areas with slight terrain undulations). The pixel coordinates and geodetic coordinates of the control points are fitted using the least squares method, the polynomial coefficients are solved, and then the coefficients are used to perform coordinate transformation on all pixels in the image, achieving a precise correspondence between each pixel and the real ground position. The vegetation index is a type of remote sensing parameter that enhances vegetation information and reduces the influence of soil background and light changes by mathematically combining the reflectance of different bands in multispectral images. The reflectance of each pixel in blue light, green light, red light, red edge and near infrared is extracted from the radiometrically calibrated multispectral image. The reflectance of the relevant band is selected according to the calculation formula of the corresponding vegetation index and substituted into the calculation to obtain the calculated value of the vegetation index. The NDVI reflects vegetation growth and canopy biomass, with values ​​typically ranging from -1 to 1. Higher values ​​indicate greater vegetation cover and growth vigor. The calculation formula is as follows: , Wherein, NIR represents the reflectance in the near-infrared band, and Red represents the reflectance in the red band; The NDRE is mainly used to improve the sensitivity to medium- and high biomass vegetation and chlorophyll content. The value range is usually between -1 and 1. The larger the value, the stronger the vegetation growth and the higher the chlorophyll content. The calculation formula is as follows: , Wherein, NIR represents the reflectance in the near-infrared band, and RedEdge represents the reflectance in the red-edge band; The GNDVI is mainly used to improve the sensitivity to chlorophyll content and growth status. The value range is usually between -1 and 1. The higher the value, the stronger the vegetation growth and the higher the chlorophyll content. The calculation formula is as follows:

[0008] Wherein, NIR represents the reflectance in the near-infrared band, and Green represents the reflectance in the green band; All field quadrats were of equal area. Multiple quadrats were selected in the field, and the alfalfa in each quadrat was uniformly cut, leaving a stubble height of 10cm. The weight of the fresh grass was then measured. The measured sample set includes alfalfa hay yield, leaf area index, chlorophyll, and leaf equivalent water thickness. The alfalfa hay yield refers to the total amount of dried hay harvested from an alfalfa field within a certain growth period. In this invention, the harvested fresh hay samples are weighed in the field and converted according to a preset standard moisture content to obtain the alfalfa hay yield. In this invention, the alfalfa hay yield is used as the main data input into the physical constraint model. The leaf area index (LAI) refers to the ratio of the total area of ​​plant leaves to the total land area per unit land area. In alfalfa fields, the LAI reflects leaf density and coverage, and is directly related to photosynthetic efficiency, chlorophyll content, and dry matter accumulation. In this invention, LAI is used as the main data input into the physical constraint model, and its calculation formula is as follows: ; Chlorophyll is a photosynthetic pigment in plant leaves, mainly found in chloroplasts, and is the core substance of photosynthesis. Chlorophyll content directly affects photosynthetic efficiency, plant growth vigor, and dry matter accumulation. The chlorophyll content data is used as auxiliary data in this invention, which is obtained by inverting the band reflectance data from UAV multispectral images. The equivalent water thickness (EWT) of leaves refers to the amount of water contained per unit area. It is an important indicator for measuring the water status of plant leaves. The level of EWT can reflect whether the plant is in a state of sufficient water or water stress, thereby indirectly affecting dry matter accumulation and hay yield. In this invention, the equivalent water thickness data of leaves is used as auxiliary data, and its calculation formula is as follows:

[0009] in, For the fresh weight of the leaves, Where A is the dry weight of the blade and A is the blade area; As a further limitation of the first aspect of the present invention, the following details are added when constructing the alfalfa hay yield calculation model: The alfalfa canopy structure features refer to the overall morphological characteristics of alfalfa plants at the field canopy scale, formed by the stems, leaves and their spatial distribution. These features mainly include canopy height, canopy coverage, leaf area index and leaf spatial distribution. The alfalfa canopy structure features directly affect light interception, water use and dry matter accumulation processes, and are reflected in UAV multispectral images by changing the canopy spectral reflectance characteristics, thus providing important structural information support for alfalfa hay yield measurement. The radiative transfer model is a crop canopy spectral simulation model formed by coupling the leaf optical model and the canopy radiative transfer model. The leaf optical model is used to describe the absorption, reflection and transmission process of incident radiation by plant leaves. By introducing leaf structural parameters (fixed according to the growth period of alfalfa) and physiological parameters (chlorophyll content, equivalent water thickness of leaves), a physical relationship between leaf spectral characteristics and physiological state is established. The canopy radiative transfer model is used to describe the propagation, absorption, reflection and scattering of solar radiation in the alfalfa canopy. By introducing leaf optical properties, canopy structure parameters (leaf area index, canopy height, leaf tilt angle distribution) and observation geometry (solar altitude angle, sensor observation angle, azimuth angle), it simulates the spectral reflectance characteristics of the alfalfa canopy in different bands. Based on the alfalfa canopy structure and growth period characteristics, a reasonable range of input parameters for the radiative transfer model is set, and then the radiative transfer model outputs a continuous spectrum. The continuous spectrum output by the radiation model is convolved with the center wavelength and response function of the UAV multispectral camera to obtain multispectral reflectance data consistent with the actual UAV image, and to generate predicted multispectral samples. The predicted multispectral samples are used to cover the range of spectral variations of alfalfa at different growth stages, different canopy densities, and different leaf water content states. The alfalfa hay yield calculation training dataset is constructed from measured samples and predicted samples. The alfalfa hay yield calculation training dataset contains the hay yield of alfalfa quadrats cut in the field, as well as the corresponding spectral information. The alfalfa hay yield calculation training dataset is compared with the multispectral remote sensing image data taken for subsequent alfalfa yield measurement to predict the alfalfa hay yield. The machine learning algorithm includes random forest algorithm, gradient boosting algorithm, support vector machine algorithm or combination thereof. It takes as input data the relationship between the multispectral features of the UAV and alfalfa hay yield, and completes the prediction of alfalfa hay yield through machine learning. The random forest algorithm is a supervised learning algorithm based on multi-decision tree ensemble. It constructs multiple independent decision trees by randomly sampling samples and features, and integrates the prediction results of each decision tree to improve the model's prediction accuracy and generalization ability. The gradient boosting algorithm is an ensemble learning method based on an additive model. It builds multiple weak learners step by step and uses the prediction residual of the previous stage model as the optimization target. It iteratively updates the model along the negative gradient direction of the loss function to achieve high-precision prediction. The Support Vector Machine (SVM) algorithm is a supervised learning method based on statistical learning theory. It models the relationship between input features and output variables by constructing an optimal margin hyperplane in the feature space, and is suitable for nonlinear regression and classification problems. The alfalfa hay production The calculation formula is as follows:

[0010] in, denoted as the weight of hay within the quadrat, and A as the area of ​​the quadrat. The spatial distribution results of alfalfa hay yield can be used for alfalfa harvesting operation planning or refined field management.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. Under the condition of limited measured sample quantity, this invention combines the radiative transfer model to construct multispectral prediction samples that meet physical consistency constraints, effectively expanding the training sample space and alleviating the dependence of traditional data-driven methods on large-scale measured samples. By fusing measured samples with samples generated by physical constraints for modeling, the stability of the model training process is improved, and the risk of overfitting due to insufficient samples is reduced, thereby enhancing the applicability and reliability of the alfalfa hay yield calculation model under different plots and different growth stages. 2. This invention addresses the characteristics of alfalfa, such as a long growth period, multiple harvests within a year, and significant changes in growth status. During model construction, multispectral feature information from multiple growth stages and harvest cycles is incorporated. This allows the model to comprehensively reflect the changes in alfalfa canopy structure, leaf area index, and physiological parameters as the growth process progresses. This enhances the adaptability of the yield measurement model to different growth stages and harvest cycles, avoiding the limited applicability of traditional single-phase or single-crop modeling methods. 3. This invention uses the harvestable hay yield calculated based on a preset standard moisture content as the measurement target. Compared to fresh hay yield or biomass indicators, this is more in line with the practical application needs in alfalfa production management, harvest scheduling, and forage trading. By directly outputting hay yield information, it can provide growers with more valuable data support for decision-making, which is beneficial for guiding the selection of alfalfa harvesting timing, yield assessment, and production benefit analysis. 4. The method of this invention includes steps such as UAV multispectral image acquisition, image preprocessing, sample construction, model training, and yield calculation. Each step has clear logic and a well-defined implementation path, making it easy to integrate into existing UAV remote sensing operations and agricultural information systems. This method can achieve rapid, non-contact yield calculation at the alfalfa field scale, supports spatial expression of yield results, and is suitable for refined field management and regional scale production monitoring of alfalfa. It has good prospects for engineering promotion and application.

[0012] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0013] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0014] Figure 1 A schematic diagram of the overall technical route for an alfalfa yield calculation method based on multispectral and physical constraint sample enhancement provided by the present invention; Figure 2 This is a schematic diagram of the physical constraint sample enhancement mechanism in this invention; Figure 3 This is a schematic diagram illustrating the construction and application process of the alfalfa hay yield calculation model in this invention; Detailed Implementation

[0015] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0016] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0017] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0018] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0019] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings.

[0020] A method for calculating alfalfa yield based on multispectral and physical constraint sample enhancement includes the following steps: S10. During the alfalfa growing season, use a drone equipped with a multispectral camera to acquire multispectral remote sensing images of the alfalfa canopy at different harvesting times. S20. Perform radiometric calibration and geometric correction on the multispectral remote sensing image, extract reflectance information for each spectral band, and calculate NDVI, NDRE, and GNDVI indices. S30. Field plots of equal area are set up within the coverage area of ​​the UAV multispectral image to measure alfalfa hay yield, leaf area index, chlorophyll content and equivalent leaf water thickness data to construct a measured sample set. S40. Based on the structural characteristics of alfalfa canopy, predictive multispectral samples that meet physical consistency constraints are generated using a radiative transfer model to expand the sample space. S50. By integrating measured and predicted samples, a training dataset for calculating alfalfa hay yield is constructed. S60. Based on the training dataset, establish a mapping model between the multispectral features of the UAV and the alfalfa hay yield, and output the alfalfa hay yield calculation results through machine learning.

[0021] Preferably, S10 also includes: S11, The alfalfa growth period includes the greening stage, the budding stage, and the initial flowering stage; S12, The UAV multispectral remote sensing image data includes blue light, green light, red light, red edge and near-infrared bands; S13. The different harvesting cycles include two of the first cycle, the second cycle, and subsequent regeneration cycles, with an interval of 25 to 35 days between each cycle. Preferably, S20 also includes: S21. The radiometric calibration method is based on a combination of reflectance calibration plates and airborne illumination information. It uses sensor calibration parameters to convert raw digital values ​​into radiance for the corresponding band, eliminating the influence of different sensor responses and imaging parameters. Subsequently, during flight or before and after takeoff and landing, images of the calibration plate with known reflectance are acquired. Combined with the solar altitude angle and irradiance information recorded by the airborne illumination sensor, the radiance is further converted into surface reflectance. This weakens the influence of differences in illumination intensity and imaging time on multispectral data, unifying multispectral image data acquired at different times and in different sorties to a consistent reflectance scale. This gives each band of data a clear physical meaning, allowing for stable use in vegetation index calculations and maintaining physical consistency with the simulated spectral data generated by the radiative transfer model. This provides a reliable data foundation for subsequent alfalfa hay yield calculations and physical constraint sample enhancement. S22. The geometric correction is based on UAV POS data and uses the direct georegistration method for coarse correction. The image pixels in the UAV body coordinate system are initially mapped to the geodetic coordinate system through the coordinate transformation model. Specifically, an affine transformation model is adopted, using position and attitude parameters from the POS data to calculate the image exterior orientation elements (such as principal point coordinates, focal length, flight altitude, and attitude angle), establishing a preliminary correspondence between pixel coordinates and geodetic coordinates. This corrects the overall offset caused by UAV flight attitude fluctuations (pitch and roll deviations) and flight altitude changes, ensuring that the error between pixel and ground position is controlled within 1-2 pixels. Simultaneously, using the deployed ground control points as a reference, a polynomial correction model is used for fine correction to eliminate residual geometric distortion (including lens distortion, terrain undulation distortion, and distortion caused by minor flight attitude fluctuations). The model uses a 2nd-3rd order polynomial (adaptively selected according to the degree of image distortion; a 2nd order polynomial is used in flat terrain areas of alfalfa fields, and a 3rd order polynomial is used in areas with slight terrain undulations). The pixel coordinates and geodetic coordinates of the control points are fitted using the least squares method, the polynomial coefficients are solved, and then the coefficients are used to perform coordinate transformation on all pixels in the image, achieving a precise correspondence between each pixel and the real ground position. S23. The vegetation index is a type of remote sensing parameter that enhances vegetation information and weakens the influence of soil background and light changes by mathematically combining the reflectance of different bands in multispectral images. The reflectance of each pixel in blue light, green light, red light, red edge and near infrared is extracted from the radiometrically calibrated multispectral image. The reflectance of the relevant band is selected according to the calculation formula of the corresponding vegetation index and substituted into the calculation to obtain the calculated value of the vegetation index. S24. The NDVI can reflect vegetation growth status and canopy biomass. The numerical range is usually between -1 and 1. The larger the value, the higher the vegetation coverage and growth vigor. The calculation formula is as follows: , Wherein, NIR represents the reflectance in the near-infrared band, and Red represents the reflectance in the red band; S25. The NDRE is mainly used to improve the sensitivity to medium- and high biomass vegetation and chlorophyll content. The value range is usually between -1 and 1. The larger the value, the stronger the vegetation growth and the higher the chlorophyll content. The calculation formula is as follows: , Wherein, NIR represents the reflectance in the near-infrared band, and RedEdge represents the reflectance in the red-edge band; S26. The GNDVI mentioned above is mainly used to improve the sensitivity to chlorophyll content and growth status. The value range is usually between -1 and 1. The larger the value, the stronger the vegetation growth and the higher the chlorophyll content. Its calculation formula is as follows:

[0022] Wherein, NIR represents the reflectance in the near-infrared band, and Green represents the reflectance in the green band; Preferably, S30 also includes: S31. All field quadrats are of equal area. Multiple quadrats are selected in the field, and the alfalfa in the quadrats is uniformly cut, leaving a stubble height of 10cm. The weight of the fresh grass is measured. S32. The measured sample set includes alfalfa hay yield, leaf area index, chlorophyll, and leaf equivalent water thickness. S33. The alfalfa hay yield refers to the total amount of dried hay harvested from the alfalfa field within a certain growth period. In this invention, the fresh hay samples are weighed in the field and converted according to the preset standard moisture content to obtain the alfalfa hay yield. In this invention, the alfalfa hay yield is used as the main data input into the physical constraint model. S34. The Leaf Area Index (LAI) refers to the multiple of the total leaf area of ​​plants per unit land area to the total land area. In alfalfa fields, the LAI reflects the density and coverage of leaves and is directly related to photosynthetic efficiency, chlorophyll content, and dry matter accumulation. In this invention, LAI is used as the main data input into the physical constraint model, and its calculation formula is as follows: ; S35. Chlorophyll is a photosynthetic pigment in plant leaves, mainly found in chloroplasts, and is the core substance of photosynthesis. Chlorophyll content directly affects photosynthetic efficiency, plant growth vigor, and dry matter accumulation. The chlorophyll content data is used as auxiliary data in this invention, which is obtained by inverting the band reflectance data from UAV multispectral images. S36. The equivalent water thickness (EWT) of leaves refers to the amount of water contained per unit area. It is an important indicator for measuring the water status of plant leaves. The level of EWT can reflect whether the plant is in a state of sufficient water or water stress, thereby indirectly affecting dry matter accumulation and hay yield. In this invention, the equivalent water thickness data of leaves is used as auxiliary data, and its calculation formula is as follows:

[0023] in, For the fresh weight of the leaves, Where A is the dry weight of the blade and A is the blade area; Preferably, S40 also includes: S41. The alfalfa canopy structure features refer to the overall morphological features of alfalfa plants at the field population scale, formed by the stems, leaves and their spatial distribution. These features mainly include canopy height, canopy coverage, leaf area index and leaf spatial distribution. The alfalfa canopy structure features directly affect light energy interception, water use and dry matter accumulation processes, and are reflected in UAV multispectral images by changing the canopy spectral reflectance characteristics, thus providing important structural information support for alfalfa hay yield calculation. S42. The radiative transfer model is a crop canopy spectral simulation model formed by coupling the leaf optical model and the canopy radiative transfer model. S43. The leaf optical model is used to describe the absorption, reflection and transmission process of incident radiation by plant leaves. By introducing leaf structural parameters (fixed according to the growth period of alfalfa) and physiological parameters (chlorophyll content, equivalent water thickness of leaves), a physical relationship between leaf spectral characteristics and physiological state is established. S44. The canopy radiative transfer model is used to describe the propagation, absorption, reflection and scattering process of solar radiation in the alfalfa canopy. By introducing leaf optical properties, canopy structure parameters (leaf area index, canopy height, leaf tilt angle distribution) and observation geometry (solar altitude angle, sensor observation angle, azimuth angle), the spectral reflectance characteristics of the alfalfa canopy in different bands are simulated. S45. Based on the alfalfa canopy structure and growth period characteristics, a reasonable range of input parameters for the radiative transfer model is set, and then the radiative transfer model outputs a continuous spectrum. S46. The continuous spectrum output by the radiation model is convolved with the center wavelength and response function of the UAV multispectral camera to obtain multispectral reflectance data consistent with the actual UAV image, and a predicted multispectral sample is generated. S47. The predicted multispectral samples are used to cover the range of spectral variations of alfalfa at different growth stages, different canopy densities, and different leaf water content states. Preferably, S50 also includes: S51. The alfalfa hay yield calculation training dataset is constructed from measured samples and predicted samples. The alfalfa hay yield calculation training dataset contains the hay yield of alfalfa quadrats cut in the field, as well as the corresponding spectral information. The alfalfa hay yield calculation training dataset is compared with the multispectral remote sensing image data taken for subsequent alfalfa yield measurement to predict the alfalfa hay yield. Preferably, S60 also includes: S61. The machine learning algorithm includes random forest algorithm, gradient boosting algorithm, support vector machine algorithm or combination thereof. Input the data corresponding to the multispectral features of the UAV and the alfalfa hay yield, and complete the prediction of alfalfa hay yield through machine learning. S62. The random forest algorithm is a supervised learning algorithm based on multi-decision tree ensemble. It constructs multiple independent decision trees by randomly sampling samples and features, and integrates the prediction results of each decision tree to improve the model's prediction accuracy and generalization ability. S63. The gradient boosting algorithm is an ensemble learning method based on an additive model. It gradually builds multiple weak learners and uses the prediction residual of the previous stage model as the optimization target. It iteratively updates the model along the negative gradient direction of the loss function to achieve high-precision prediction. S64. The support vector machine algorithm is a supervised learning method based on statistical learning theory. It models the relationship between input features and output variables by constructing an optimal margin hyperplane in the feature space. It is suitable for nonlinear regression and classification problems. S65, the alfalfa hay yield The calculation formula is as follows:

[0024] in, denoted as the weight of hay within the quadrat, and A as the area of ​​the quadrat. S66. The spatial distribution results of alfalfa hay yield can be used for alfalfa harvesting operation planning or refined field management.

Claims

1. A method for calculating alfalfa yield based on multispectral and physical constraint sample enhancement, characterized in that, Includes the following steps: S10. During the alfalfa growing season, use a drone equipped with a multispectral camera to acquire multispectral remote sensing images of the alfalfa canopy at different harvesting times. S20. Perform radiometric calibration and geometric correction on the multispectral remote sensing image, extract reflectance information for each spectral band, and calculate the normalized vegetation index NDVI, normalized red edge index NDRE, and normalized green vegetation index GNDVI. Within the coverage area of ​​S30 and UAV multispectral remote sensing images, field plots of equal area were set up to measure alfalfa hay yield, leaf area index, chlorophyll content and equivalent leaf water thickness data to construct a measured sample set. S40. Based on the structural characteristics of alfalfa canopy, predictive multispectral samples that meet physical consistency constraints are generated using a radiative transfer model to expand the sample space. S50. By integrating measured and predicted samples, a training dataset for calculating alfalfa hay yield is constructed. S60. Based on the training dataset, establish a mapping model between the multispectral features of the UAV and the alfalfa hay yield, and output the alfalfa hay yield calculation results through machine learning.

2. The alfalfa yield calculation method according to claim 1, characterized in that: Step S10 also includes: S11, The alfalfa growth period includes the greening stage, the budding stage, and the initial flowering stage; S12, The multispectral remote sensing image data includes blue light, green light, red light, red edge, and near-infrared bands; S13. The different harvesting cycles include two of the first, second, and subsequent regeneration cycles, with an interval of 25 to 35 days between each cycle.

3. The alfalfa yield calculation method according to claim 1, characterized in that: Step S20 also includes: S21. The radiometric calibration method is based on a combination of reflectivity calibration board and airborne illumination information. The original digital values ​​are converted into radiance of the corresponding band using sensor calibration parameters. Subsequently, during flight or before and after takeoff and landing, images of the calibration board with known reflectance are acquired. Combined with the solar altitude angle and irradiance information recorded by the airborne illumination sensor, the radiance is further converted into surface reflectance. This reduces the impact of differences in light intensity and imaging time on multispectral data, providing a data basis for subsequent alfalfa hay yield calculation and physical constraint sample enhancement. S22. The geometric correction is based on UAV POS data and uses direct georegistration for coarse correction. A coordinate transformation model is used to initially map image pixels from the UAV body coordinate system to the geodetic coordinate system. Specifically, an affine transformation model is used to calculate the image exterior orientation elements using position and attitude parameters from the POS data, establishing a preliminary correspondence between pixel coordinates and geodetic coordinates. This corrects the overall offset caused by fluctuations in UAV flight attitude and altitude changes, keeping the error between pixel and ground position within 1-2 pixels. Simultaneously, using the established ground control points as a reference, a polynomial correction model is used for fine correction to eliminate residual geometric distortion. The model uses a 2nd-3rd order polynomial, and the least squares method is used to fit the pixel coordinates and geodetic coordinates of the control points. The polynomial coefficients are then solved, and these coefficients are used to perform coordinate transformation on all pixels in the image, achieving a precise correspondence between each pixel and the actual ground position.

4. The alfalfa yield calculation method according to claim 3, characterized in that: Step S30 further includes: S31. All field quadrats are of equal area. Multiple quadrats are selected in the field, and the alfalfa in the quadrats is uniformly cut, leaving a stubble height of 10cm. The weight of the fresh grass is measured. S32. The alfalfa hay yield refers to the total amount of dried hay harvested from the alfalfa field within a certain growth period. The harvested fresh hay samples are weighed in the field and converted according to the preset standard moisture content to obtain the alfalfa hay yield. The alfalfa hay yield is used as the main data input into the physical constraint model.

5. The alfalfa yield calculation method according to claim 3, characterized in that: Step S40 further includes: S41. The alfalfa canopy structure characteristics refer to the overall morphological characteristics of alfalfa plants at the field population scale, formed by the stems, leaves and their spatial distribution, including canopy height, canopy coverage, leaf area index and leaf spatial distribution. S42. The radiative transfer model is a crop canopy spectral simulation model formed by coupling the leaf optical model and the canopy radiative transfer model. S43. The leaf optical model is used to describe the absorption, reflection and transmission process of incident radiation by plant leaves. By introducing leaf structural parameters and physiological parameters, a physical relationship between leaf spectral characteristics and physiological state is established. S44. The canopy radiative transfer model is used to describe the propagation, absorption, reflection and scattering process of solar radiation in the alfalfa canopy. By introducing leaf optical characteristics, canopy structure parameters and solar altitude angle, sensor observation angle and azimuth angle, the spectral reflectance characteristics of alfalfa canopy in different bands are simulated. S45. Based on the alfalfa canopy structure and growth period characteristics, a reasonable range of input parameters for the radiative transfer model is set, and then the radiative transfer model outputs a continuous spectrum. S46. The continuous spectrum output by the radiative transfer model is convolved with the center wavelength and response function of the UAV multispectral camera to obtain multispectral reflectance data consistent with the actual UAV image, and a predicted multispectral sample is generated. S47. The predicted multispectral samples are used to cover the spectral variation range of alfalfa at different growth stages, different canopy densities, and different leaf water content states.

6. The alfalfa yield calculation method according to claim 1, characterized in that: Step S50 further includes: the alfalfa hay yield calculation training dataset is constructed from measured samples and predicted samples. The alfalfa hay yield calculation training dataset contains the hay yield of alfalfa quadrats cut in the field, as well as the corresponding spectral information. The alfalfa hay yield calculation training dataset is compared with the multispectral remote sensing image data taken for subsequent alfalfa yield measurement, thereby predicting the alfalfa hay yield.

7. The alfalfa yield calculation method according to claim 1, characterized in that: Step S60 further includes: the machine learning algorithm includes random forest algorithm, gradient boosting algorithm, support vector machine algorithm or a combination thereof, inputting data corresponding to the relationship between the UAV multispectral features and alfalfa hay yield, and completing the prediction of alfalfa hay yield through machine learning; The spatial distribution results of alfalfa hay yield are used for alfalfa harvesting operation planning or refined field management.

Images