Crop growth prediction method and system based on multispectral unmanned aerial vehicle monitoring

Abstract

The present application belongs to the technical field of crop growth prediction, and particularly relates to a crop growth prediction method and system based on multispectral unmanned aerial vehicle monitoring. The method firstly configures growth state data of crops in standard growth environment in different growth stages, and collects multi-dimensional data of real crops in different soil water content and pest status in different growth stages. The contribution of each vegetation index to crop growth is determined through a factor analysis algorithm, and the high-spectral data is reduced in dimension through a principal component analysis algorithm, fused with the multispectral data, and a multispectral resolution feature space is constructed. Secondly, the LiDAR data and the vegetation index are combined, and a crop real plant height growth fitting function is obtained through simulation simulation. Thirdly, the real growth vegetation index space is calculated, and the standard growth stage real growth state space is obtained. Finally, the calculated space and function are input into the model constructed by the reinforcement learning algorithm for training, so as to realize the accurate prediction of the crop growth state.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of crop growth prediction, and particularly relates to a crop growth prediction method and system based on multispectral unmanned aerial vehicle monitoring. BACKGROUND

[0002] Crop growth monitoring and prediction are crucial for precision agriculture management, which can timely understand the crop growth conditions and take reasonable agricultural measures such as irrigation, fertilization, pest control, etc., thereby optimizing resource utilization, improving crop yield and quality, and reducing environmental pollution and production costs. In recent years, multispectral remote sensing technology has been widely used in crop monitoring due to its own advantages. However, to achieve accurate crop growth prediction based on multispectral unmanned aerial vehicle monitoring, there are still many challenges. On the one hand, the processing and analysis of multispectral data require complex algorithms and models to extract effective crop growth information. On the other hand, the spectral response characteristics of different crops in different growth environments differ, and specific prediction models need to be established.

[0003] A crop growth prediction method, device, equipment and medium are disclosed in Chinese Patent No. CN114972684A. The method includes determining a lossless two-dimensional image data set of an original three-dimensional point cloud data set based on an image backtracking algorithm; fusing the original three-dimensional point cloud data set, the lossless two-dimensional image data set and meteorological factor data, and inversely calculating an optimal crop growth parameter multi-dimensional deep learning model by combining the spatial and temporal specificity of the growth parameters of different crops to evaluate the growth parameter information of different crops and obtain the spatial and temporal distribution of the growth parameters of different crops.

[0004] A crop growth prediction method and system are disclosed in Chinese Patent No. CN111582554B. The method includes obtaining normalized vegetation indices corresponding to the current year and each historical year; fitting and reconstructing the normalized vegetation indices corresponding to the current year and each historical year to obtain a current NDVI fitting curve and each historical NDVI fitting curve; performing normalization processing on the current NDVI fitting curve and each historical NDVI fitting curve using a DTW algorithm to obtain a current NDVI growth curve and each historical NDVI growth curve; calculating the shortest distance between the current NDVI growth curve and each historical NDVI growth curve, and taking the year corresponding to the historical NDVI growth curve with the smallest distance as the best NDVI growth matching year to predict the growth of the target crop in the current year.

[0005] The prior art has the following problems: the existing algorithms only predict the overall growth state of crops by using single NDVI and image data, without considering that the NDVI corresponding to different growth stages of crops is different, and without considering the specific influence of water and diseases and pests on the growth state, resulting in low prediction result accuracy, and the present application provides a crop growth prediction method and system based on multispectral unmanned aerial vehicle monitoring. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a crop growth prediction method and system based on multispectral unmanned aerial vehicle monitoring, which first configures growth state data of crops in different stages of the growth period under standard growth conditions, and collects multi-dimensional data of real crops under different soil water content and disease and pest conditions in different growth stages, determines the contribution of each vegetation index to crop growth by a factor analysis algorithm, and uses a principal component analysis algorithm to reduce the dimension of hyperspectral data, and fuses the multispectral data to construct a multispectral resolution feature space, secondly, combines LiDAR data and vegetation index, and obtains a crop real plant height growth fitting function through simulation, thirdly, calculates a real growth vegetation index space and obtains a standard growth state space in different stages of the growth period, and finally inputs the calculated space and function into a model constructed by a reinforcement learning algorithm for training to realize accurate prediction of the growth state of crops.

[0007] To achieve the above object, the present application provides the following technical scheme:

[0008] The crop growth prediction method based on multispectral unmanned aerial vehicle monitoring comprises:

[0009] S1, configure growth state data of crops in different stages of the growth period under standard growth conditions, and obtain multispectral data, crop LiDAR data and hyperspectral data of real crops under different soil water content and disease and pest conditions in different growth stages;

[0010] S2, obtain a multi-vegetation index space, and calculate the growth simulation contribution of each vegetation index to the corresponding type of crop in different growth stages according to a factor analysis algorithm to obtain a matching vegetation index for each growth stage of the corresponding type of crop;

[0011] S3, input the hyperspectral data under different soil water content and disease and pest conditions into a principal component analysis algorithm for dimension reduction, and fuse the dimension-reduced hyperspectral data with the corresponding multispectral data to obtain a multispectral resolution feature space in different growth stages;

[0012] S4. Utilize the multispectral resolution feature space of different growth stages to construct vegetation index water and pest and disease regulation factors corresponding to different growth stages. At the same time, through crop LiDAR data and water and pest and disease regulation factors, a simulation algorithm is used to obtain the actual plant height growth fitting function of crops under different growth states.

[0013] S5. Multiply the vegetation index moisture and pest regulation factors corresponding to different growth stages with the vegetation index matching each growth stage to obtain a real growth vegetation index space, and multiply the real growth vegetation index space with the growth state data of the growth period in the standard growth environment to obtain the real growth state space of the standard growth period in each stage;

[0014] S6. Input the real growth state space of the standard staged growth period and the real plant height growth fitting function of crops under different growth states into the reinforcement prediction and adjustment model constructed by the reinforcement learning algorithm for training, obtain the trained reinforcement prediction and adjustment model, and predict the crop growth state through the crop growth data obtained in real time.

[0015] Specifically, the steps for obtaining the matching vegetation index for each growth stage include:

[0016] S201. Extracting the band reflectance data required for calculating each vegetation index by acquiring multispectral data of real crops at different growth stages, different soil moisture contents, and pest and disease conditions, calculating different vegetation indices, and constructing a standardized vegetation index matrix containing N vegetation indices;

[0017] S202: Set each category of crops to include m growth stages, and select q crop samples from each stage. The sample space of

[0018] S203, according to The sample space and standardized vegetation index matrix are input into the factor analysis algorithm to obtain N vegetation indices. The covariance matrix C of the sample points;

[0019] S204 , calculating the eigenvector and eigenvalue of the corresponding vegetation index using the characteristic equation according to the obtained covariance matrix C.

[0020] Specifically, the steps for obtaining the matching vegetation index for each growth stage also include:

[0021] S205. Calculate the contribution rate and factor loading matrix of the nth vegetation index to the kth type of crop at the mth growth stage based on the eigenvector and eigenvalue of the corresponding vegetation index;

[0022] S206, set a contribution rate threshold, according to the contribution rate of the N vegetation indexes obtained for the crop in the corresponding class in the m growth stage and the contribution rate threshold, the vegetation index corresponding to the cumulative contribution rate greater than the contribution rate threshold is retained as the vegetation index corresponding to the current growth stage of the crop;

[0023] S207, according to the factor loading matrix corresponding to the retained vegetation index, the contribution degree of the corresponding vegetation index is calculated by the square sum of the corresponding elements in the factor loading matrix;

[0024] S208, the contribution degree and the contribution rate of the corresponding vegetation index are used to calculate the weighting coefficient of the corresponding vegetation index, and the weighting coefficient of the vegetation index is obtained by using the calculated weighting coefficient of the vegetation index and the vegetation index.

[0025] Specifically, the construction steps of the water and disease and pest adjustment factors include:

[0026] S401, according to the S3 process, select samples corresponding to different water states from the multispectral resolution feature space containing m growth stages, and obtain the growth evaluation score under the corresponding water state by the evaluation algorithm; wherein p represents p soil water content levels;

[0027] S402, according to the selected samples, calculate the corresponding vegetation index under different water states;

[0028] S403, according to the corresponding vegetation index under different water states and the growth evaluation score under the corresponding water state, obtain the regression function of different water states on crop growth by the multi-source linear regression algorithm;

[0029] S404, the coefficient of the regression function of crop growth is taken as the adjustment factor of water on crop growth, and the water adjustment factor sequence is obtained;

[0030] S405, input the water adjustment factor sequence into the genetic algorithm for optimization search, and obtain the optimal water adjustment factor sequence corresponding to different growth periods.

[0031] Specifically, the construction steps of the water and disease and pest adjustment factors also include:

[0032] S406, according to the S3 process, select samples corresponding to different disease states and samples corresponding to different pest states from the multispectral resolution feature space containing m growth stages, and obtain the growth evaluation score under the corresponding disease state and pest state by the evaluation algorithm; wherein u represents u type disease state samples, and v represents v type pest state samples;

[0033] S407, repeat the process of S402-S405 to obtain the optimal disease adjustment factor sequence and the optimal pest adjustment factor sequence in different disease states and pest states corresponding to different growth stages;

[0034] S408, construct a water and pest regulation factor space according to the optimal water adjustment factor sequence, the optimal disease adjustment factor sequence and the optimal pest adjustment factor sequence.

[0035] Specifically, the construction of the real plant height growth fitting function includes:

[0036] S409, according to the LiDAR data of different growth stages of crops and the growth state data of crops in different stages under standard growth environment, a three-dimensional model of standard height corresponding to different growth stages of crops is constructed by a three-dimensional algorithm;

[0037] S410, according to the three-dimensional model of standard height corresponding to different growth stages of crops and the water and pest regulation factor space corresponding to different growth stages, a real plant height growth fitting function corresponding to different growth stages of crops of the corresponding type is obtained by a simulation algorithm and a nonlinear fitting function;

[0038] S411, according to the real plant height data and the real plant height growth fitting function prediction data, the prediction error is obtained, and the prediction error is fed back to the simulation algorithm for adjustment, and the real plant height growth fitting function with a prediction error less than a configured prediction error threshold is obtained as the real plant height growth fitting function corresponding to the growth stage of the current type of crop.

[0039] Specifically, the construction and training steps of the reinforcement prediction adjustment model include:

[0040] S601, constructing an input state at the current time according to the obtained water and pest regulation factor space, the growth state of crops in different stages under standard growth environment and the real plant height growth fitting function;

[0041] S602, setting a trigger information at the current time according to the input state at the current time, specifically:

[0042] If the current crop is the kth type of crop in the mth growth stage, the action execution information is triggered, that is, the water and pest regulation factors of the kth type of crop in the mth growth stage are called from the water and pest regulation factor space, the plant height growth fitting function and the corresponding comprehensive matching vegetation index are adjusted and corrected, and the corrected plant height growth fitting function and the corresponding comprehensive matching vegetation index are embedded into the corresponding time sequence prediction sub-model to predict the growth state of the current crop;

[0043] S603, constructing a current time execution action according to the current time trigger information, and constructing a reward function according to the error between the prediction result and the real growth state, the plant height growth fitting function and the corresponding comprehensive matching vegetation index correction loss;

[0044] S604, inputting the current time input state, the current time trigger information, the current time execution action and the reward function into the reinforcement prediction adjustment model for training, obtaining the trained reinforcement prediction adjustment model, and outputting the current time corresponding crop growth state.

[0045] The crop growth prediction system based on multispectral unmanned aerial vehicle monitoring comprises a data acquisition module, an index matching module and a simulation module.

[0046] The data acquisition module is configured to configure growth state data of crops in different stages of growth in a standard growth environment, and to acquire multispectral data, crop LiDAR data and hyperspectral data of real crops in different soil water content and pest states in different growth stages.

[0047] The index matching module comprises an index space construction unit and a matching analysis unit.

[0048] The index space construction unit is configured to acquire different types of vegetation indexes, and to construct a multi-vegetation index space.

[0049] The matching analysis unit calculates the growth simulation contribution of each vegetation index to the corresponding type of crop in different growth stages by a factor analysis algorithm, and obtains the matching vegetation index of each growth stage of the corresponding type of crop.

[0050] The simulation module comprises a feature fusion unit, a factor construction unit and a function simulation unit.

[0051] The feature fusion unit is configured to input the hyperspectral data in different soil water content and pest states into a principal component analysis algorithm for dimension reduction, and to fuse the dimension-reduced hyperspectral data with the corresponding multispectral data to obtain a multispectral resolution feature space in different growth stages.

[0052] The factor construction unit constructs water and pest adjustment factors of corresponding vegetation indexes in different growth stages by using the multispectral resolution feature space in different growth stages.

[0053] Specifically, the system further comprises a growth state space module and a prediction module.

[0054] The growth state space module is used for multiplying the vegetation index water and the pest and disease adjustment factor corresponding to the growth stage and the vegetation index matched with each growth stage to obtain a real growth vegetation index space, and multiplying the real growth vegetation index space with standard growth environment stage-by-stage growth period growth state data to obtain a standard stage-by-stage growth period real growth state space.

[0055] The prediction module is used for inputting the standard stage-by-stage growth period real growth state space and the crop real plant height growth fitting function under different growth states into a reinforcement prediction adjustment model constructed by a reinforcement learning algorithm to train the reinforcement prediction adjustment model, and obtaining the trained reinforcement prediction adjustment model, and predicting the crop growth state by using real-time acquired crop growth data.

[0056] An electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements a crop growth state prediction method based on multi-spectral unmanned aerial vehicle monitoring when executing the computer program.

[0057] A computer readable storage medium has computer instructions stored thereon, and the computer instructions execute a crop growth state prediction method based on multi-spectral unmanned aerial vehicle monitoring when running.

[0058] Compared with the prior art, the present application has the following beneficial effects:

[0059] The present application aims at the deficiencies of the prior art, and constructs a high-precision crop growth state prediction system through multi-source data fusion and intelligent algorithms, which significantly improves the accuracy and practicality of the prediction. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 The crop growth prediction method flowchart based on multispectral unmanned aerial vehicle monitoring of the embodiment 1 of the present application;

[0061] Figure 2 The crop growth prediction system module diagram based on multispectral unmanned aerial vehicle monitoring of the embodiment 2 of the present application. DETAILED DESCRIPTION

[0062] Embodiment 1

[0063] Growth, that is, the condition and trend of crop growth. The growth of crops can be described by individual and group characteristics. A reasonable group composed of healthy individuals is a good crop area. Individual characteristics can be described by stem, leaf, root and ear characteristics. Such as plant height, branch number, leaf number, shape, color, root development, etc. Group characteristics can be described by group density, leaf area index, layout and dynamics. The existing algorithm only predicts the growth state of crops through single NDVI and image data, without considering that the corresponding NDVI of crops in different growth stages is different, and without considering the specific influence of water and pests and diseases on the growth state, resulting in low prediction result accuracy. For this purpose, please refer to Figure 1 The present application provides an embodiment: a crop growth prediction method based on multispectral unmanned aerial vehicle monitoring, comprising the following steps:

[0064] S1, configure the growth state data of different stages in the standard crop growth environment, and obtain the multispectral data, crop LiDAR data and hyperspectral data of real crops under different soil water content and pest and disease states in different growth stages;

[0065] Further, the growth state data of different stages in the standard crop growth environment in the embodiment is obtained in the laboratory by corresponding to the standard growth environment required by crops, and can also be obtained by corresponding research papers.

[0066] Further, in the embodiment, the data acquisition includes:

[0067] Standard growth environment data: by referring to professional research papers, the growth state data of wheat in different stages in the standard growth environment is obtained, such as plant height, leaf number, biomass and other indicators in different stages such as tillering stage, jointing stage, heading stage, etc.

[0068] Real crop data: wheat is planted in the test field, different soil water content (such as drought, normal, wet three states) and pest and disease states (no pests and diseases, mild pests and diseases, severe pests and diseases) are set, and multispectral cameras, LiDAR equipment and hyperspectral instruments are used to obtain related data in different growth stages.

[0069] S2, obtain a multi-vegetation index space, and calculate a growth simulation contribution degree of each vegetation index for a corresponding type of crop at different growth stages according to a factor analysis algorithm, to obtain a matching vegetation index for each growth stage of the corresponding type of crop;

[0070] It is known from existing data that the leaf of a green plant is a basic organ for photosynthesis. In the case of the same other conditions, the greater the leaf area, the stronger the photosynthesis, and the stronger the photosynthesis, the more dry matter accumulation, and the higher the biomass. This physiological mechanism can be confirmed indirectly from the analysis of remote sensing data through the relationship between different wave bands in the plant reflectance spectrum, such as the use of a combination of red and near-infrared wave band reflection information to form various vegetation indices. A large number of studies have shown that the normalized difference vegetation index NDVI has a good correlation with the leaf area index, biomass and chlorophyll content of plants, and is an important indicator factor of the growth state and spatial distribution density of vegetation. Therefore, in this embodiment, the contribution degrees of different vegetation index spaces to different crops at different biological stages are analyzed to obtain a matching vegetation index for each type of crop at each growth stage, so as to improve the prediction accuracy.

[0071] Further, the multi-vegetation index space in this embodiment includes a ratio vegetation index, a normalized difference vegetation index, a difference vegetation index, an enhanced vegetation index, a soil-adjusted vegetation index, a green normalized difference vegetation index, a triangular vegetation index, a vertical vegetation index, and a red edge vegetation index.

[0072] Further, in this embodiment, a vegetation index matrix is constructed: wave band reflectance data is extracted from multi-spectral data, and various vegetation indices such as a ratio vegetation index (RVI), a normalized difference vegetation index (NDVI), and a difference vegetation index (DVI) are calculated to construct a standardized vegetation index matrix containing N=10 vegetation indices. For example, when calculating NDVI, assuming that the near-infrared wave band reflectance is 0.6 and the red light wave band reflectance is 0.2, then NDVI is 0.5.

[0073] Further, the step of obtaining a matching vegetation index for each growth stage in this embodiment includes:

[0074] S201, through the obtained multi-spectral data of real crops at different growth stages and under different soil water content and pest conditions, wave band reflectance data required for calculating various vegetation indices is extracted, different vegetation indices are calculated, and a standardized vegetation index matrix containing N vegetation indices is constructed;

[0075] S202, set m growth stages for each type of crop, and select q crop samples from each stage to form a sample space ;

[0076] S203, according to The sample space and the normalized vegetation index matrix are input into a factor analysis algorithm to obtain a covariance matrix C of N vegetation indexes at N sample points.

[0077] S204, according to the obtained covariance matrix C, the characteristic vector and the characteristic value corresponding to the vegetation index are calculated through the characteristic equation;

[0078] S205, according to the characteristic vector and the characteristic value corresponding to the vegetation index, the contribution rate of the nth vegetation index to the kth type of crop at the mth growth stage and the factor loading matrix are calculated;

[0079] S206, a contribution rate threshold is set, and according to the contribution rate of the N vegetation indexes to the corresponding crops at the mth growth stage and the contribution rate threshold, the vegetation index corresponding to the cumulative contribution rate greater than the contribution rate threshold is reserved as the vegetation index corresponding to the current growth stage of the crop;

[0080] S207, according to the factor loading matrix corresponding to the reserved vegetation index, the contribution degree of the corresponding vegetation index is calculated through the square sum of the corresponding elements in the factor loading matrix;

[0081] S208, the contribution degree and the contribution rate of the corresponding vegetation index are used to calculate the weighting coefficient of the corresponding vegetation index, and the calculated vegetation index weighting coefficient and the vegetation index are used to obtain the comprehensive matching vegetation index corresponding to the current growth stage of the crop.

[0082] ​The process realizes accurate acquisition of the vegetation index matched with different growth stages of crops through a fine factor analysis algorithm, and significantly improves the accuracy and reliability of the prediction of the growth state of crops. Specifically, first, the band reflectance is extracted from the real crop multispectral data, and a variety of vegetation indexes are calculated to construct a normalized vegetation index matrix, ensuring the consistency and comparability of the data. Second, by setting the sample space and inputting the factor analysis algorithm, the covariance matrix and eigenvectors and eigenvalues are obtained, which not only considers the diversity of different growth stages, but also quantifies the importance of each vegetation index, solving the problem that the single NDVI in the prior art cannot adapt to different growth stages. Third, by calculating the contribution rate and factor loading matrix, and selecting the key vegetation index according to the contribution rate threshold, it is ensured that the selected vegetation index has the highest explanatory power, especially by calculating the contribution degree through the element square sum in the factor loading matrix, which further enhances the scientificity and rationality of the selection. Finally, by using the vegetation index weighting coefficient and the contribution degree, the comprehensive matching vegetation index is obtained, realizing the best vegetation index matching for each growth stage. This fine matching method not only improves the prediction accuracy, but also dynamically reflects the specific influence of water and pests and diseases, greatly improving the adaptability and practicality of the model. In summary, through the systematic application of the factor analysis algorithm, the process solves the problem that the single NDVI in the prior art cannot adapt to different growth stages, and realizes high-precision vegetation index matching.

[0083] S3, inputting the hyperspectral data of different soil water content and pest and disease states into a principal component analysis algorithm for dimension reduction, and fusing the dimension-reduced hyperspectral data with the corresponding multispectral data to obtain a multispectral resolution feature space of different growth stages;

[0084] In this embodiment, the rich spectral information of the hyperspectral data is integrated into the multispectral image to enhance the spectral resolution and feature expression capability of the image. For example, the dimension-reduced hyperspectral data is fused with the multispectral data through principal component analysis or the like, or the mapping relationship between the hyperspectral and multispectral data is automatically learned by using a deep learning algorithm to generate a fused high-resolution, high-spectral-resolution image.

[0085] S4, using the multispectral resolution feature space of different growth stages to construct the water and pest and disease adjustment factors corresponding to different growth stages, and simultaneously, through the crop LiDAR data and the water and pest and disease adjustment factors, a real plant height growth fitting function of the crops under different growth states is obtained through a simulation algorithm;

[0086] Further, the construction steps of the water and pest and disease adjustment factors include:

[0087] S401, selecting m vegetation indexes from the multispectral resolution feature space containing m growth stages according to the S3 process corresponding to different water conditions, and obtain the growth evaluation score under the corresponding water condition through the evaluation algorithm; p represents p soil water content levels; the selection process is selected by a person skilled in the art according to the expert experience method, which will not be described further here;

[0088] S402, according to the selected sample, the vegetation index corresponding to different water conditions is calculated; the vegetation index of the selected sample in this embodiment is calculated by using the standard vegetation index calculation formula corresponding to the growth stage obtained in S201-S208;

[0089] S403, according to the vegetation index corresponding to different water conditions and the growth evaluation score under the corresponding water condition, a regression function of crop growth under different water conditions is obtained through a multi-source linear regression algorithm;

[0090] S404, the coefficient of the regression function of crop growth is taken as a water regulation factor of crop growth, and a water regulation factor sequence is obtained;

[0091] S405, the water regulation factor sequence is input into a genetic algorithm for optimization search, and an optimal water regulation factor sequence corresponding to different growth periods is obtained;

[0092] Further, as an example of the above process:

[0093] Constructing a water regulation factor:

[0094] Selecting samples and evaluation: from a multi-spectral resolution feature space containing m=5 growth stages, 30 samples are selected for drought, normal and wet three water conditions, and the growth evaluation score under the corresponding water condition is obtained through the evaluation algorithm (such as comprehensive scoring based on crop growth indicators), assuming that the drought state score is 60, the normal state score is 80, and the wet state score is 70.

[0095] Calculate the vegetation index: according to the selected sample, the vegetation index corresponding to different water conditions is calculated.

[0096] Regression calculation and sequence generation: through a multi-source linear regression algorithm, taking the vegetation index as the independent variable and the growth evaluation score as the dependent variable, a regression function of crop growth under different water conditions is obtained. Assuming that the regression function is , then the coefficient 0.5 of the regression function is taken as a water regulation factor of crop growth, and a water regulation factor sequence is obtained, x is the independent variable of the regression function.

[0097] S406, according to the S3 process, select different disease states and corresponding to different pest states, and through the evaluation algorithm, growth evaluation scores under corresponding disease states and pest states are obtained; wherein, u represents u type disease state samples, and v represents v type pest state samples;

[0098] S407, the processes of S402-S405 are repeated to obtain optimal disease adjustment factor sequences and optimal pest adjustment factor sequences under different disease states and pest states corresponding to different growth stages;

[0099] S408, according to the optimal water adjustment factor sequence, the optimal disease adjustment factor sequence and the optimal pest adjustment factor sequence, a water and disease and pest adjustment factor space is constructed.

[0100] This process constructs accurate water and disease and pest adjustment factors through principal component analysis (PCA) dimension reduction and multi-source linear regression algorithm combined with genetic algorithm optimization, significantly improving the accuracy and adaptability of crop growth state prediction. Specifically, first, the hyperspectral data of different soil water contents and disease and pest states are subjected to PCA dimension reduction and fused with multispectral data to generate a multispectral resolution feature space, ensuring efficient data processing and information retention. Second, for samples under different water conditions, growth evaluation scores are obtained through the evaluation algorithm, and the regression function of water on crop growth is calculated using the multi-source linear regression algorithm. The regression coefficient is used as the water adjustment factor. The introduction of genetic algorithm further optimizes the water adjustment factor sequence to ensure its optimal performance at each growth stage. This process not only quantifies the specific impact of water on crop growth, but also improves the robustness and accuracy of the model. Third, optimal disease adjustment factor sequences and optimal pest adjustment factor sequences are obtained, taking into account the impact of disease and pests on crop growth. Finally, the optimal water, disease and pest adjustment factors are integrated to construct a complete water and disease and pest adjustment factor space, achieving comprehensive regulation of the crop growth environment. Through systematic data analysis and optimization algorithms, the problem of single factor not being able to fully reflect crop growth conditions in existing technologies is solved, providing more accurate and dynamic adjustment factors, significantly improving the accuracy of crop growth state prediction.

[0101] Further, the construction steps of the real plant height growth fitting function in this embodiment include:

[0102] S409, according to the LiDAR data of different growth stages of crops and the growth state data of crops under standard growth environment, a three-dimensional model of the standard height corresponding to different growth stages of crops is constructed through a three-dimensional algorithm;

[0103] S410, according to the corresponding standard height three-dimensional model of different growth stages of crops and the corresponding water and pest regulation factor space of different growth stages, through a simulation algorithm and a nonlinear fitting function, a real plant height growth fitting function corresponding to different growth stages of the corresponding type of crops is obtained;

[0104] S411, according to the real plant height data and the real plant height growth fitting function prediction data, the prediction error is obtained, and the prediction error is fed back to the simulation algorithm for adjustment, and the real plant height growth fitting function with a prediction error less than a configured prediction error threshold is obtained as the real plant height growth fitting function of the corresponding growth stage of the current type of crops.

[0105] This process constructs an accurate real plant height growth fitting function through three-dimensional modeling, simulation algorithm and error feedback mechanism, significantly improves the accuracy and reliability of crop growth state prediction; first, the LiDAR data of different growth stages of crops and the growth state data of different stages of growth period under standard growth environment are used to construct a standard height three-dimensional model of different growth stages of crops through a three-dimensional algorithm. This model provides an accurate structure reference for subsequent simulation, ensuring the accuracy of the simulation results.

[0106] Secondly, combined with the standard height three-dimensional model and the water and pest regulation factor space, through a simulation algorithm and a nonlinear fitting function, a real plant height growth fitting function corresponding to different growth stages of the corresponding type of crops is obtained. This method of comprehensively considering the influence of multiple factors not only improves the explanatory power of the model, but also dynamically reflects the changes in the actual growth environment.

[0107] S5, multiply the vegetation index water and pest regulation factor corresponding to different growth stages and the matching vegetation index of each growth stage to obtain a real growth vegetation index space, and multiply the real growth vegetation index space with the growth state data of the standard growth environment under the growth period of different stages to obtain a standard real growth state space of different stages of growth period;

[0108] The vegetation index water and pest regulation factor corresponding to different growth stages and the matching vegetation index of each growth stage are multiplied to obtain a real growth vegetation index space. For example, in the jointing stage, assuming that the comprehensive matching vegetation index is 0.6 and the water regulation factor is 0.8, then the multiplication of the two values in the real growth vegetation index space of this stage is 0.48.

[0109] S6, input the standard phased growth period real growth state space and the crop real plant height growth fitting function under different growth states into the reinforcement prediction adjustment model constructed by the reinforcement learning algorithm for training, obtain the trained reinforcement prediction adjustment model, and predict the crop growth state through the real-time acquired crop growth data. Further, the prediction results in the embodiment include predicted heading stage plant height, leaf number, and final crop ear number, etc.

[0110] Further, the construction and training steps of the reinforcement prediction adjustment model in the embodiment include:

[0111] S601, construct the current time input state according to the acquired water and pest and disease adjustment factor space, the growth state under the standard growth environment and the real plant height growth fitting function in the phased growth period;

[0112] S602, set the current time trigger information according to the current time input state, specifically:

[0113] If the current time crop is the kth type crop in the mth growth stage, trigger the action execution information, that is, from the water and pest and disease adjustment factor space, call the water and pest and disease adjustment factor of the kth type crop in the mth growth stage, adjust and correct the plant height growth fitting function and the corresponding comprehensive matching vegetation index, and embed the corrected plant height growth fitting function and the corresponding comprehensive matching vegetation index into the corresponding time sequence prediction sub-model, and predict the growth state of the current crop;

[0114] S603, construct the current time action according to the current time trigger information, and construct the reward function according to the error between the prediction result and the real growth state, the correction loss of the plant height growth fitting function and the corresponding comprehensive matching vegetation index;

[0115] S604, input the current time input state, the current time trigger information, the current time action and the reward function into the reinforcement prediction adjustment model for training, obtain the trained reinforcement prediction adjustment model, and output the corresponding crop growth state at the current time.

[0116] The process realizes high-precision dynamic prediction of the growth state of crops through a reinforcement prediction adjustment model constructed by a reinforcement learning algorithm, and significantly improves the scientificity and efficiency of agricultural management. Specifically, first, the vegetation index water and pest and disease adjustment factors at different growth stages are multiplied by the matching vegetation index to obtain a real growth vegetation index space, and further multiplied by the growth state data at the standard growth environment to generate a standard growth state space at the growth stage; this process ensures that the vegetation index and growth state at each growth stage are accurately adjusted to reflect the actual growth conditions. Secondly, the reinforcement prediction adjustment model constructed by the reinforcement learning algorithm is trained in combination with the standard growth state space at the growth stage and the real plant height growth fitting function, which not only considers the influence of multiple factors, but also dynamically adjusts through the real-time acquisition of crop growth data, improves the accuracy and adaptability of the prediction, especially when the crop is at a specific growth stage, the model will call the corresponding water and pest and disease adjustment factors to modify the plant height growth fitting function and the comprehensive matching vegetation index, and is embedded into the time series prediction sub-model for prediction. This closed-loop feedback mechanism continuously adjusts the model parameters by comparing the prediction results with the error of the real growth state, to ensure that the prediction accuracy is always better than the set threshold.

[0117] Embodiment 2

[0118] Please refer to Figure 2 Another embodiment provided by the present application is a crop growth prediction system based on multi-spectral unmanned aerial vehicle monitoring, comprising a data acquisition module, an index matching module, a simulation module, a growth state space module and a prediction module.

[0119] The data acquisition module is configured to configure growth state data at the growth stage in the standard crop growth environment, and acquire multi-spectral data, crop LiDAR data and hyperspectral data of real crops under different soil water content and pest and disease states at different growth stages.

[0120] The index matching module is configured to acquire a matching vegetation index for each growth stage of a corresponding type of crop; the index matching module comprises an index space construction unit and a matching analysis unit.

[0121] The index space construction unit is configured to acquire different types of vegetation indexes to construct a multi-vegetation index space.

[0122] The matching analysis unit calculates the growth simulation contribution degree of each type of vegetation index to the different growth stages of the corresponding type of crop through a factor analysis algorithm to obtain a matching vegetation index for each growth stage of the corresponding type of crop.

[0123] The simulation module obtains a real plant height growth fitting function under different growth states through a simulation algorithm.

[0124] The simulation module comprises a feature fusion unit, a factor construction unit and a function simulation unit.

[0125] The feature fusion unit is configured to input different soil water content and pest and disease state hyperspectral data into a principal component analysis algorithm for dimension reduction, and fuse the dimension-reduced hyperspectral data with corresponding multispectral data to obtain multispectral resolution feature spaces at different growth stages.

[0126] The factor construction unit is configured to construct vegetation index water and pest and disease adjustment factors at different growth stages based on the multispectral resolution feature spaces at different growth stages.

[0127] The function simulation unit is configured to obtain crop real plant height growth fitting functions at different growth states by a simulation algorithm based on crop LiDAR data and the water and pest and disease adjustment factors.

[0128] The growth state space module is configured to multiply the vegetation index water and pest and disease adjustment factors at the same growth stage and the matched vegetation index at each growth stage to obtain a real growth vegetation index space, and multiply the real growth vegetation index space with growth state data at different growth stages in a standard growth environment to obtain a standard real growth state space at different growth stages.

[0129] The prediction module is configured to input the standard real growth state space at different growth stages and the crop real plant height growth fitting functions at different growth states into a reinforcement prediction adjustment model constructed by a reinforcement learning algorithm for training, obtain a trained reinforcement prediction adjustment model, and predict crop growth states by real-time acquired crop growth data.

[0130] Embodiment 3

[0131] An electronic device comprising a memory and a processor, the memory storing a computer program, and the processor implementing a crop growth prediction method based on multispectral unmanned aerial vehicle monitoring when executing the computer program.

[0132] A computer readable storage medium having computer instructions stored thereon, the computer instructions being executed to perform a crop growth prediction method based on multispectral unmanned aerial vehicle monitoring.

[0133] The embodiments of the present application are described above with reference to the drawings, but the present application is not limited to the specific embodiments described above, which are merely illustrative rather than restrictive, and a person of ordinary skill in the art can make changes, modifications, replacements and variations to the above-described embodiments without departing from the purpose of the present application and the scope protected by the claims, which are all within the protection of the present application.

Claims

1. A method for crop growth prediction based on multispectral drone monitoring, characterized in that, The method comprises the following steps: S1, configure growth state data of crops in different growth stages in a standard growth environment, and obtain multispectral data, crop LiDAR data and hyperspectral data of real crops under different soil water content and pest and disease states in different growth stages; S2, obtain a multi-vegetation index space, and calculate the growth simulation contribution of each vegetation index to the corresponding type of crop in different growth stages according to a factor analysis algorithm to obtain a matching vegetation index for each growth stage of the corresponding type of crop; S3, input the hyperspectral data under different soil water content and pest and disease states into a principal component analysis algorithm for dimension reduction, and fuse the dimension-reduced hyperspectral data with the corresponding multispectral data to obtain a multispectral resolution feature space in different growth stages; S4, use the multispectral resolution feature space in different growth stages to construct water and pest and disease adjustment factors of the corresponding vegetation index in different growth stages, and obtain a real crop height growth fitting function in different growth states through simulation and simulation algorithm based on the crop LiDAR data and the water and pest and disease adjustment factors; S5, multiply the water and pest and disease adjustment factors of the corresponding vegetation index in different growth stages and the matching vegetation index in each growth stage to obtain a real growth vegetation index space, and multiply the real growth vegetation index space with the growth state data in the standard growth environment to obtain a standard real growth state space in different growth stages; S6, input the standard real growth state space in different growth stages and the real crop height growth fitting function in different growth states into a reinforcement prediction adjustment model constructed by a reinforcement learning algorithm for training to obtain a trained reinforcement prediction adjustment model, and predict the growth state of the crop through real-time acquisition of crop growth data.

2. The crop health prediction method based on multispectral drone monitoring as claimed in claim 1, wherein, The step of obtaining the matching vegetation index in each growth stage comprises: S201, extract the band reflectance data required for calculating each vegetation index from the obtained multispectral data of real crops in different growth stages under different soil water content and pest and disease states, calculate different vegetation indexes, and construct a standardized vegetation index matrix containing N vegetation indexes; S202, set m growth stages for each type of crop, and select q crop samples from each stage to form a sample space ; S203、According to The sample space and the normalized vegetation index matrix of N vegetation indexes on the sample points are input into the factor analysis algorithm, and a covariance matrix C of S204, according to the obtained covariance matrix C, the characteristic equation is used to calculate the eigenvectors and eigenvalues of the corresponding vegetation index.

3. The crop health prediction method based on multispectral drone monitoring as claimed in claim 2, wherein, The step of obtaining the matching vegetation index in each growth stage further comprises: S205, according to the eigenvectors and eigenvalues of the corresponding vegetation index, the contribution rate of the nth vegetation index to the kth type of crop in the mth growth stage and the factor loading matrix are calculated; S206, set a contribution rate threshold, and according to the contribution rate of the N vegetation indexes to the crops in the mth growth stage in the corresponding category and the contribution rate threshold, retain the vegetation index corresponding to the cumulative contribution rate greater than the contribution rate threshold as the vegetation index corresponding to the current growth stage of the crop; S207, according to the factor loading matrix corresponding to the retained vegetation index, the contribution degree of the corresponding vegetation index is calculated by the sum of squares of the corresponding elements in the factor loading matrix. S208, calculating the corresponding vegetation index weighting coefficient by using the contribution degree and the contribution rate of the corresponding vegetation index, and obtaining the corresponding comprehensive matching vegetation index of the current growth stage of the crop by using the calculated vegetation index weighting coefficient and the vegetation index.

4. The crop health prediction method based on multispectral drone monitoring as claimed in claim 3, wherein, The constructing step of the water and disease and pest regulating factor includes: S401、selecting, according to the S3 process, from a multispectral resolution feature space containing m growth stages corresponding samples under p different water conditions, and obtaining a growth evaluation score under the corresponding water condition through an evaluation algorithm; p represents p soil water content levels. S402, calculating the corresponding vegetation index under different water conditions according to the selected sample; S403, obtaining the regression function of different water conditions on crop growth by a multi-source linear regression algorithm according to the corresponding vegetation index under different water conditions and the growth evaluation score under the corresponding water condition; S404, taking the coefficient of the regression function of crop growth as the water regulating factor of crop growth, and obtaining the water regulating factor sequence; S405, inputting the water regulating factor sequence into the genetic algorithm for optimization search to obtain the optimal water regulating factor sequence corresponding to different growth periods.

5. The crop health prediction method based on multispectral drone monitoring as claimed in claim 4, wherein, The constructing step of the water and disease and pest regulating factor further includes: S406、selecting, according to the S3 process, samples corresponding to u different disease states and v different pest states from the multispectral resolution feature space containing m growth stages and obtaining, through the evaluation algorithm, growth evaluation scores under the corresponding disease states and pest states; wherein u represents u types of disease state samples, and v represents v types of pest state samples.​ S407, repeating the process of S402-S405 to obtain the optimal disease regulating factor sequence and the optimal pest regulating factor sequence under different disease conditions and pest conditions corresponding to different growth periods; S408, constructing the water and disease and pest regulating factor space according to the optimal water regulating factor sequence, the optimal disease regulating factor sequence and the optimal pest regulating factor sequence.

6. The crop health prediction method based on multispectral drone monitoring as claimed in claim 5, wherein, The constructing step of the real plant height growth fitting function includes: S409, constructing the corresponding standard height three-dimensional model of different growth stages of the crop by a three-dimensional algorithm according to the LiDAR data of different growth stages of the crop and the growth state data of different growth stages under the standard growth environment of the crop; S410, obtaining the real plant height growth fitting function corresponding to different growth stages of the corresponding type of crop by a simulation algorithm and a nonlinear fitting function according to the corresponding standard height three-dimensional model of different growth stages of the crop and the water and disease and pest regulating factor space corresponding to different growth stages; S411, obtaining the prediction error by predicting the real plant height data and the real plant height growth fitting function prediction data, and feeding back the prediction error to the simulation algorithm for adjustment to obtain the real plant height growth fitting function with a prediction error less than a configured prediction error threshold as the real plant height growth fitting function of the corresponding growth stage of the current type of crop.

7. The crop health prediction method based on multispectral drone monitoring as claimed in claim 6, wherein, The constructing and training step of the reinforcement prediction adjustment model includes: S601, constructing the current time input state according to the obtained water and disease and pest regulating factor space, the growth state of different growth stages under the standard growth environment and the real plant height growth fitting function; S602, setting the current time trigger information according to the current time input state, specifically: If the current time is the kth type of crop in the mth growth stage, trigger the action execution information, that is, call the water and pest regulation factor of the kth type of crop in the mth growth stage from the water and pest regulation factor space, adjust and correct the plant height growth fitting function and the corresponding comprehensive matching vegetation index, and embed the corrected plant height growth fitting function and the corresponding comprehensive matching vegetation index into the corresponding time sequence prediction sub-model to predict the growth state of the current crop; S603, according to the current time trigger information, construct the current time execution action, and construct the reward function according to the error between the prediction result and the real growth state, the correction loss of the plant height growth fitting function and the corresponding comprehensive matching vegetation index; S604, input the current time input state, current time trigger information, current time execution action and reward function into the reinforcement prediction adjustment model for training, obtain the trained reinforcement prediction adjustment model, and output the corresponding crop growth state at the current time.

8. A crop vigor prediction system based on multispectral UAV monitoring for implementing the crop vigor prediction method based on multispectral UAV monitoring according to any one of claims 1 to 7, characterized in that, It includes: Data acquisition module, index matching module; The data acquisition module is configured to configure growth state data in different stages of growth period under standard crop growth environment, and acquire multispectral data, crop LiDAR data and hyperspectral data of real crops under different soil water content and pest states in different growth stages; The index matching module includes an index space construction unit and a matching analysis unit; The index space construction unit is configured to acquire different types of vegetation indexes to construct a multi-vegetation index space; The matching analysis unit calculates the growth simulation contribution degree of each type of vegetation index to the corresponding type of crop in different growth stages by using a factor analysis algorithm, and obtains the matching vegetation index of each growth stage of the corresponding type of crop.

9. The crop health prediction system based on multispectral drone monitoring as claimed in claim 8, wherein, The system further includes a simulation module; The simulation module includes a feature fusion unit, a factor construction unit and a function simulation unit; The feature fusion unit is configured to input the hyperspectral data under different soil water content and pest states into a principal component analysis algorithm for dimension reduction, and fuse the dimension-reduced hyperspectral data with the corresponding multispectral data to obtain a multispectral resolution feature space in different growth stages. The factor construction unit constructs water and pest regulation factors of corresponding vegetation indexes in different growth stages by using the multispectral resolution feature space in different growth stages.

10. The crop health prediction system based on multispectral drone monitoring as claimed in claim 9, wherein, The system further includes a growth state space module and a prediction module; The growth state space module is configured to multiply the water and pest regulation factors of corresponding vegetation indexes in the same growth stage and the matching vegetation index in each growth stage to obtain a real growth vegetation index space, and multiply the real growth vegetation index space with the growth state data in different stages of growth period under standard growth environment to obtain a standard real growth state space in different stages of growth period. The prediction module is configured to input a standard phased growth period real growth state space and a crop real plant height growth fitting function under different growth states into a reinforcement prediction adjustment model constructed by a reinforcement learning algorithm for training, obtain a trained reinforcement prediction adjustment model, and predict a growth state of the crop by using real-time acquired crop growth data.

Images