CN-121999366-A - Winter wheat regional yield intelligent estimation method and system integrating crop mechanism model and machine learning

CN121999366ACN 121999366 ACN121999366 ACN 121999366ACN-121999366-A

Abstract

The invention provides an intelligent estimation method and system for winter wheat regional yield by combining a crop mechanism model and machine learning, and relates to the technical field of agricultural remote sensing and crop growth simulation. The method utilizes multi-temporal satellite remote sensing images to identify the winter wheat range through object-oriented classification to construct winter wheat masks and remote sensing features thereof, selects sample points in the masks, inverts leaf area index time sequences through a radiation transmission model and drives a crop growth model, obtains single-point simulation yield through ensemble Kalman filtering assimilation, and forms a training sample library. And carrying out growth classification according to the normalized vegetation index accumulated values of the booting stage and the flowering stage, selecting representative sample points to construct a training sample set, establishing a winter wheat yield estimation model and applying the model to all pixels to obtain yield space distribution and administrative unit yield, and realizing high-precision rapid estimation of winter wheat regional yield.

Inventors

MA ZHANLIN
LIU YUHONG
LIU ZHAN
CHANG LI
Yin Shouqiang
WEN FENG
LU CHUNYANG
ZHAO ZHANHUI
ZHANG YINGLEI

Assignees

河南城建学院

Dates

Publication Date: 20260508
Application Date: 20260127

Claims (10)

1. An intelligent estimation method for winter wheat regional yield by combining a crop mechanism model and machine learning is characterized by comprising the following steps: Acquiring a multi-temporal Sentinel satellite active and passive remote sensing image covering the whole growth period of winter wheat, extracting vegetation index, texture characteristics and spectral band characteristics, and identifying the planting range of the winter wheat by adopting an object-oriented classification method to obtain a winter wheat mask and multi-temporal remote sensing characteristics corresponding to the winter wheat mask; Selecting sample points in the winter wheat mask, inverting multi-temporal remote sensing characteristics of each sample point based on a radiation transmission model to obtain leaf area index time sequence of each sample point, taking the leaf area index time sequence, meteorological parameters, soil parameters and field management parameters together as input to drive a crop growth model, and calibrating model parameters related to the maximum value of the leaf area index by using a parameter optimization method according to the maximum value of the target leaf area index to obtain a calibrated crop growth model; Inputting leaf area index time sequences of all the sample points as assimilation variables into the calibrated crop growth model, sequentially assimilating by adopting a set Kalman filtering method, updating state variables of the crop growth model and parameters related to yield, outputting single-point simulation yield of all the sample points, and constructing a yield sample library containing positions of the sample points, the leaf area index time sequences and the single-point simulation yield; According to the winter wheat booting period normalized vegetation index accumulated value and the flowering period normalized vegetation index accumulated value, performing growth classification on pixels in the winter wheat mask, selecting representative sample points from each growth grade, and pairing the multi-temporal remote sensing characteristics of the representative sample points with the corresponding single-point simulation output to obtain a machine learning training sample set; The machine learning training sample set is used as input, a random forest regression algorithm is adopted to establish an regional winter wheat yield estimation model, multi-temporal remote sensing features are used as model input, and the single-point simulation yield is used as model output, so that the regional winter wheat yield estimation model can represent the corresponding relation between the remote sensing features and the yield under the mechanism constraint of a crop growth model; The regional winter wheat yield estimation model is applied to all pixels in the winter wheat mask, the multi-temporal remote sensing characteristic of each pixel is taken as input, a winter wheat yield spatial distribution result with the spatial resolution of 10m is obtained, and area weighting statistics is carried out on an administrative unit on the basis of the yield spatial distribution result, so that an administrative unit-scale winter wheat yield estimation result is obtained.
2. The intelligent estimation method for winter wheat regional yield by combining a crop mechanism model and machine learning according to claim 1, wherein the object-oriented classification method adopts a segmentation algorithm based on simple non-iterative clustering to segment the multi-temporal Sentinel satellite active and passive remote sensing images, uses Sentinel radar polarization characteristics, vegetation indexes and the texture characteristics obtained by gray level co-occurrence matrix calculation as input characteristics, and adopts a supervision classification algorithm based on random forest to classify winter wheat and non-winter wheat, so that the overall accuracy of winter wheat planting range identification is not lower than 95%.
3. The intelligent estimation method for winter wheat regional yield by fusing a crop mechanism model and machine learning according to claim 1, wherein in the step of acquiring multi-temporal sentel satellite active and passive remote sensing images covering the whole growth period of winter wheat, a multi-scene image with cloud content less than 10% is selected for time sequence mean synthesis in a winter wheat turning-green jointing stage for an optical image, and a radar time sequence image is acquired in a month mean synthesis mode for a radar image so as to ensure the time continuity and the space consistency of the multi-temporal remote sensing features.
4. The intelligent estimation method for winter wheat regional yield by combining a crop mechanism model and machine learning according to claim 1 is characterized in that the radiation transmission model is a unified radiation transmission model of blades and crowns, multispectral reflectivity of each sample point on each observation date is used as input, forward modeling is carried out on a leaf area index and a crowned structural parameter related to the leaf area index, a direct search type optimization algorithm is adopted in a leaf area index inversion process by the parameter optimization method, and the leaf area index time sequence of each sample point on each observation date is obtained by solving with the minimum error between the forward modeling reflectivity and the corresponding wave band reflectivity in the multi-time phase remote sensing feature as a target.
5. The intelligent estimation method for winter wheat regional yield by combining a crop mechanism model and machine learning according to claim 1, wherein when the parameter optimization method is used for calibrating crop growth model parameters related to the maximum value of the leaf area index, firstly, a group of crop growth model parameters sensitive to the maximum value of the leaf area index are screened out based on a global sensitivity analysis method, the group of crop growth model parameters sensitive to the maximum value of the leaf area index are used as parameters to be optimized, a particle swarm optimization algorithm is adopted for searching parameter combinations of the parameters to be optimized, and root mean square error between leaf area index time sequences obtained by simulating the calibrated crop growth model and leaf area index time sequences obtained by inversion of a radiation transmission model is minimized.
6. The intelligent estimation method for winter wheat regional yield by combining a crop mechanism model and machine learning according to claim 1, wherein the ensemble kalman filtering method uses leaf area index as an external observed quantity when the leaf area index time sequence of each sample point is input into a calibrated crop growth model as an assimilation variable, updates state variables of the crop growth model and crop growth model parameters related to yield at each time node with remote sensing observation, outputs single-point simulated yield of each sample point, and constructs a yield sample library comprising positions of the sample points, leaf area index time sequence and the single-point simulated yield.
7. The intelligent estimation method for winter wheat regional yield by combining a crop mechanism model and machine learning according to claim 1, wherein the method for estimating winter wheat regional yield by integrating the crop mechanism model and the machine learning is characterized by classifying the pixels in the winter wheat mask according to the normalized vegetation index integrated value of the winter wheat booting period and the normalized vegetation index integrated value of the flowering period, selecting representative sample points from the growth levels, and comprising the following steps: Taking the booting stage normalized vegetation index accumulated value and the flowering stage normalized vegetation index accumulated value as growth potential evaluation indexes, and dividing the growth potential into four grades of excellent growth potential, better growth potential, general growth potential and worse growth potential by adopting a natural break point grading method; the representative sample points are selected in each growth class according to a spatially uniform distribution principle.
8. The intelligent estimation method for winter wheat regional yield by combining a crop mechanism model and machine learning according to claim 1, wherein the number of samples of the machine learning training sample set is not less than six hundred twenty five, the representative sample points cover winter wheat planting regions of each growth level and each year in the winter wheat mask, and the number of samples of each growth level is distributed according to a preset proportion, so that the regional winter wheat yield estimation model established by adopting a random forest regression algorithm has stable generalization capability under the conditions of multiple years and multiple growth levels.
9. The intelligent estimation method for winter wheat regional yield by combining a crop mechanism model and machine learning according to claim 1, wherein in the step of applying the regional winter wheat yield estimation model to all pixels in the winter wheat mask, the single-point simulation yield is obtained by operating the crop growth model assimilated by the ensemble kalman filter method only at the sample points, the crop growth model and the ensemble kalman filter assimilation are not operated any more in all pixel ranges in the winter wheat mask, and the multi-temporal remote sensing characteristics of each pixel are input into the regional winter wheat yield estimation model to obtain a winter wheat yield spatial distribution result, and the winter wheat yield spatial distribution result is output as a regional winter wheat yield estimation result.
10. An intelligent estimation system for winter wheat regional yield by combining a crop mechanism model and machine learning, which is characterized by comprising the following components: The winter wheat mask construction unit is used for acquiring a multi-temporal Sentinel satellite active and passive remote sensing image covering the whole growth period of winter wheat, extracting vegetation index, texture characteristics and spectral band characteristics, and identifying the winter wheat planting range by adopting an object-oriented classification method to obtain a winter wheat mask and multi-temporal remote sensing characteristics corresponding to the winter wheat mask; The system comprises a leaf area index inversion and crop growth model calibration unit, a parameter optimization method and a parameter optimization unit, wherein the leaf area index inversion and crop growth model calibration unit is used for selecting sample points in the winter wheat mask, inverting multi-temporal remote sensing characteristics of each sample point based on a radiation transmission model to obtain leaf area index time sequences of each sample point, taking the leaf area index time sequences, meteorological parameters, soil parameters and field management parameters together as input to drive a crop growth model, and calibrating model parameters related to the maximum value of the leaf area index according to a target maximum value of the leaf area index by using the parameter optimization method to obtain a calibrated crop growth model; The leaf area index assimilation and single-point simulation yield generation unit is used for inputting the leaf area index time sequence of each sample point as an assimilation variable into the calibrated crop growth model, sequentially assimilating by adopting a set Kalman filtering method, updating the state variable of the crop growth model and parameters related to yield, outputting the single-point simulation yield of each sample point, and constructing a yield sample library containing the position of the sample point, the leaf area index time sequence and the single-point simulation yield; The growth classification and representative sample point construction unit is used for classifying the growth of pixels in the winter wheat mask according to the normalized vegetation index accumulated value of the winter wheat in the booting stage and the normalized vegetation index accumulated value of the flowering stage, selecting representative sample points from the growth levels, and pairing the multi-temporal remote sensing characteristics of the representative sample points with the corresponding single-point simulation output to obtain a machine learning training sample set; The regional yield estimation model training unit is used for taking the machine learning training sample set as input, adopting a random forest regression algorithm to establish a regional winter wheat yield estimation model, taking multi-temporal remote sensing characteristics as model input, taking the single-point simulation yield as model output, and enabling the regional winter wheat yield estimation model to represent the corresponding relation between the remote sensing characteristics and the yield under the mechanism constraint of a crop growth model; the regional yield spatial expression and statistics output unit is used for applying the regional winter wheat yield estimation model to all pixels in the winter wheat mask, taking multi-temporal remote sensing characteristics of each pixel as input, obtaining a winter wheat yield spatial distribution result with the spatial resolution of 10m, and carrying out area weighted statistics on the administrative unit on the basis of the yield spatial distribution result to obtain an administrative unit-scale winter wheat yield estimation result.

Description

Winter wheat regional yield intelligent estimation method and system integrating crop mechanism model and machine learning Technical Field The invention relates to the technical field of agricultural remote sensing and crop growth simulation, in particular to an intelligent winter wheat regional yield estimation method and system integrating a crop mechanism model and machine learning. Background Winter wheat is one of three food crops worldwide, and efficient and accurate winter wheat planting area monitoring and yield estimation provide important basis for food policy adjustment. Compared with the manual statistical method, the satellite remote sensing technology has the advantages of economy, real time, high efficiency and objectivity in aspects of regional crop identification, growth condition monitoring, yield estimation and the like, and has become an important means for acquiring regional winter wheat information. In the process of simulating winter wheat growth, the mechanical crop growth model comprehensively considers various factors such as soil, climate, nutrients, variety, field management and the like, can describe the processes such as photosynthetic production, respiration, transpiration, organ building and the like day by day, and can reflect the crop growth mechanism more than the traditional statistical model and the estimated yield model only based on simple remote sensing indexes. The existing research explores different assimilation algorithms, assimilation variables and the performances of the crop growth model in regional yield estimation by coupling the crop growth model with a remote sensing information assimilation method, wherein the mode of adopting a set Kalman filtering method to assimilate the leaf area index time sequence data to the WOFOST crop growth model is adopted, and the method has better performance in the aspect of regional crop yield simulation precision. However, such modes are mainly based on area assimilation, and when high-spatial resolution remote sensing data are assimilated, the calculation amount and calculation force are very large. On the one hand, the forward modeling of the single-point winter wheat by the crop growth model already requires a large amount of calculation, if data assimilation is introduced on a regional scale and remote sensing data with higher spatial resolution is used, although the spatial distribution of winter wheat yield can be carefully expressed, the requirement of regional assimilation on calculation power and time is high, so that the application efficiency in a large range is low. On the other hand, by taking machine learning algorithms such as multiple linear regression, random forest regression and the like as cores and establishing an empirical relation between remote sensing indexes and winter wheat yield, a good yield estimation effect is achieved in part of research areas, but a large number of yield samples are required to be collected in the field, data collected in the last 5 years are usually required for related research, the samples have strong timeliness, and the harvesting period of winter wheat is only about one week generally, so that the number of the obtainable samples is limited. The existing regional winter wheat yield estimation mode has the defects of ensuring mechanism constraint, high spatial resolution expression and yield estimation efficiency and reducing dependence on large-scale long-term yield samples, and a method for realizing rapid estimation of winter wheat regional yield by using a crop growth model and remote sensing data to carry out physical constraint on limited representative samples and combining the obtained single-point yield estimation result with regional remote sensing information is needed. Disclosure of Invention In order to overcome the defects of the prior art, the invention aims to provide the intelligent estimation method and the intelligent estimation system for the yield of the winter wheat region, which are combined with the crop mechanism model and the machine learning, by adopting crop growth model assimilation to generate a high-reliability yield sample at a small number of representative sample points and combining the machine learning to realize the high-resolution extrapolation of the region scale, the accuracy and the efficiency of the yield estimation of the winter wheat region are obviously improved. In order to achieve the above object, the present invention provides the following solutions: an intelligent estimation method for winter wheat regional yield by combining a crop mechanism model and machine learning, comprising the following steps: Acquiring a multi-temporal Sentinel satellite active and passive remote sensing image covering the whole growth period of winter wheat, extracting vegetation index, texture characteristics and spectral band characteristics, and identifying the planting range of the winter wheat by adopting an object-oriented classification