Rice yield estimation method for optimizing RF model through improved whale optimization algorithm

Abstract

The invention discloses a rice yield estimation method for optimizing an RF model through an improved whale optimization algorithm, and belongs to the technical field of rice yield estimation. The problem that in the prior art, a traditional rice yield prediction method is low in prediction result precision is solved. The method comprises the following steps: acquiring multispectral image data of rice in different growth periods by using an unmanned aerial vehicle, and measuring and calculating rice growth parameters and average yield after a rice mature period; performing normalization preprocessing on the multispectral image data, and constructing a sample data set; the whale optimization algorithm is improved, the improved whale optimization algorithm is constructed to optimize the RF model, model parameters are set, a WOA-RF model is obtained, the sample data set is input into the WOA-RF model for training and screening to obtain an optimal rice yield estimation model, and the rice yield of the target area is estimated. According to the method, the unmanned aerial vehicle remote sensing image and the rice growth parameters are combined, the rice yield estimation precision is effectively improved, and the method can be applied to rice yield prediction.

Description

Technical Field

[0001] The invention relates to a rice yield estimation method, in particular to a rice yield estimation method using an improved whale optimization algorithm to optimize a RF model, and belongs to the technical field of rice yield estimation. Background Art

[0002] In recent years, with the continuous increase of global population, the continuous degradation of cultivated land and the outflow of rural labor and other related factors, the growth rate of grain production has gradually slowed down, and the supply and demand of grain has been in a tight balance for a long time. At the same time, as an important food crop, rice yield is one of the ultimate indicators of crop production. How to timely and accurately estimate the rice yield per unit area is of great and guiding significance for ensuring high and stable yields.

[0003] At present, most rice yield estimation methods use multivariate linear fitting to construct yield estimation models. The yield estimation model fitting correlation is too high. Too many factors will cause the model to run more slowly and easily fall into local minima. The prediction results are not accurate, making it difficult to ensure the accuracy of yield prediction and difficult to apply on a large scale in the field of precision agriculture.

[0004] In summary, there is a need for a rice yield estimation method that combines UAV remote sensing images with information such as rice growth parameters and an improved whale optimization algorithm to optimize the RF model. Summary of the invention

[0005] A brief overview of the present invention is provided below in order to provide a basic understanding of certain aspects of the present invention. It should be understood that this overview is not an exhaustive overview of the present invention. It is not intended to identify key or important parts of the present invention, nor is it intended to limit the scope of the present invention. Its purpose is merely to present certain concepts in a simplified form as a prelude to a more detailed description discussed later.

[0006] In view of this, in order to solve the problem of low prediction accuracy of traditional rice yield estimation methods in the prior art, the present invention provides a rice yield estimation method using an improved whale optimization algorithm to optimize the RF model.

[0007] The technical solution is as follows: A rice yield estimation method for optimizing the RF model using an improved whale optimization algorithm, comprising the following steps:

[0008] S1. Use drones to obtain multispectral image data of rice at different growth stages, and calculate rice growth parameters and average yield after rice maturity;

[0009] S2. Perform normalization preprocessing on the multispectral image data and construct a sample data set;

[0010] S3. Improve the whale optimization algorithm and construct an improved whale optimization algorithm;

[0011] S4. Based on the improved whale optimization algorithm, the RF model is optimized, the model parameters are set, and the WOA-RF model is obtained. The sample data set is input into the WOA-RF model for training, and the optimal rice yield estimation model is screened;

[0012] S5. Use the optimal rice yield estimation model to estimate the rice yield in the target area, generate a rice yield estimation map, and complete the rice yield estimation.

[0013] Furthermore, in S1, the drone is equipped with a multispectral camera, and the rice growth parameters include SPAD values ​​of rice leaves, aboveground biomass, plant height and leaf area index.

[0014] Furthermore, the step S2 specifically includes the following steps:

[0015] S21. Use environmental visual analysis software to pre-process the multispectral image data to obtain an orthophoto map, which is then converted into a vegetation index map to obtain a reflectance data set;

[0016] S22. Using the band mathematical tool of the environmental visual analysis software, based on the reflectance data set, combined with the calculation formula of the vegetation index, band calculation is performed to obtain the average return on investment value of the vegetation index map of each sample area as the corresponding vegetation index data set of different growth periods in the region;

[0017] S23. Based on the multispectral image data, the growth monitoring module, disaster monitoring module and soil moisture monitoring module in the ecological monitoring system are used for inversion to obtain the raster data of the growth level of rice, disaster level and soil moisture;

[0018] S24. Construct a sample dataset based on the reflectance dataset, the corresponding vegetation index datasets of different growth periods in the region and the raster data.

[0019] Furthermore, the step S3 specifically includes the following steps:

[0020] S31. Initialize algorithm parameters, set the whale population size to N, and the maximum number of iterations of the whale optimization algorithm to t max , the constant coefficient is b, the random number is l, and the current global optimal position vector is x best(t) According to the chaotic reverse learning initialization strategy, the initialization whale population (x1, x2, x3..., x N ,i=1,2,3…,N);

[0021] In S31, the chaos reverse learning initialization strategy process is as follows:

[0022] The D-dimensional vector y is randomly generated by the whale optimization algorithm, y={y j ,j=1,2,3....,D}, D is the vector dimension, and the chaotic mapping function y is obtained according to the mapping function Ten i+1,j , y ij is the component of the jth mapping sequence on the i-th row, and generates the chaotic sequence y j ,y j ={y ij ,i=1,2,3....,N};

[0023] The chaotic sequence y j Mapped to the original space, we get the chaotic initialization population x, x = {x i ,i=1,2,3....,N}, where x i is the i-th individual in the population, x i ={x ij , i=1,2,3....,D}, where x ij is the coordinate value of the jth dimension of the i-th population individual, x ij =x min,j +y i,j *(x max,j -x min,j ), where x min,j For x ij The search upper bound, x max,j For x ij Search lower bound of ;

[0024] According to the chaotic initialization population x, the reverse population ox is calculated, ox = {ox i ,i=1,2,3....,N}, where ox i is the i-th reverse population individual, ox i ={ox ij ,j=1,2,3....,D},ox ij is the coordinate value of the jth dimension of the i-th reverse population individual, ox ij =(x max,j +x min,j -x ij );

[0025] Merge the chaotic initialization population x with the reverse population ox to obtain a new population {xUox i}, calculate the objective function value of the new population and sort it, and select N individuals whose fitness values ​​meet the preset conditions as the initial whale population;

[0026] S32. Calculate the fitness value of each individual in the initialized whale population {F(xi ),i=1,2,3....,N}, record the individual positions that currently meet the preset conditions;

[0027] S33. Calculate the value of the convergence factor α in the whale optimization algorithm, Among them, α 初 is the initial value of the convergence factor α, α 终 is the final value of the convergence factor α, y (t-1) is the chaotic sequence generated by the self-logistic mapping function, y (t) ∈(-1,1), t is the current iteration number, t max is the maximum number of iterations;

[0028] S34. Calculate the value of the inertia weight ω in the whale optimization algorithm, Among them, ω 初 is the initial value of the inertia weight ω, ω 终 is the final value of the inertia weight ω, y (t) ∈(-1,1) is the chaotic sequence generated by the self-logistic mapping function;

[0029] S35. Perform chaotic perturbation collaborative update based on convergence factor α and inertia weight ω, and set the current individual position update formula and spiral update position formula;

[0030] In S35, the current individual position update formula is expressed as:

[0031] x (t+1) =ω*x best(t) -A*|C*x best(t) -x (t) |

[0032] Among them, x (t+1) is the individual position updated by the current individual position formula, x best(t) is the individual optimal position, A is the coefficient vector, C is a random number uniformly distributed in (0,2), x (t) is the current individual position;

[0033] The spiral update position formula is expressed as:

[0034] x (t+1) =ω*x best(t) +D*e bl *cos(2πl)

[0035] Where l is a random number in [-1,1], and b is a logarithmic spiral shape constant;

[0036] S36. Update the coefficient vectors A, l and C through steps S33 and S34;

[0037] S37. Introduce a random parameter p in the whale optimization algorithm, and the value range of p is {0,1}. If the random parameter p is less than 0.5, enter global exploration. If the value of the coefficient vector of the whale optimization algorithm is |A|<1, the current position is updated according to the current individual position update formula, and the current individual position is updated by the encirclement mechanism. If the value of the coefficient vector of the whale optimization algorithm is |A|>1, the search mechanism is updated according to the search mechanism update formula to update the whale individual position.

[0038] In S37, the search mechanism update formula is expressed as:

[0039] x (t+1) =x rand(t) -A|C*x rand(t) -x (t) |

[0040] Among them, x rand(t) is the position vector of a randomly selected individual whale;

[0041] S38. If the random parameter p ≥ 0.5, enter local development, perform spiral mechanism update on the individual position of the whale, and update the current individual position according to the spiral update position formula;

[0042] S39. Add 1 to the number of iterations of the current whale optimization algorithm, and determine whether the current number of iterations is equal to the maximum number of iterations or the optimal value. If the maximum number of iterations is reached, proceed to step S310, otherwise return to step S32 to continue iterating;

[0043] S310. After the iteration is completed, the whale optimization algorithm improved by the above steps outputs the optimal solution.

[0044] Furthermore, the specific implementation method of step S4 includes the following steps:

[0045] S41. Based on the optimal solution output by the improved whale optimization algorithm, set the RF model parameters, optimize the RF model, and obtain the WOA-RF model;

[0046] S42. Based on the sample data set, the normalized data is used as training samples and input into the WOA-RF model for training, and the optimal WOA-RF model after training is selected as the optimal rice yield estimation model.

[0047] Furthermore, in S5, the determination coefficient R is used 2 , Corrected Root Mean Square Error The mean absolute percentage error (MAPE) is combined with the optimal rice yield estimation model to evaluate the rice yield in the target area and generate a rice yield estimation map, where RMSE is the root mean square error, is the sample average of rice yield.

[0048] The beneficial effects of the present invention are as follows: the present invention provides a rice yield estimation method for optimizing the RF model by using an improved whale optimization algorithm, using unmanned aerial vehicle remote sensing image data and rice growth parameters as sample data, combined with the actual measured value of rice yield, introducing a chaotic optimization operator into the standard whale optimization algorithm to perform a reverse learning initialization method to generate a population, thereby increasing population diversity, and introducing an inertia weight to cooperate with a convergence factor to jointly adjust the global exploration ability and local development ability of the population, thereby accelerating the convergence speed, improving the robustness, and improving the evolvability; the present invention optimizes the RF model by setting the RF model parameters, thereby ensuring the global optimal parameter combination, thereby realizing rice yield estimation, and at the same time making the optimal rice yield estimation model of the present invention have a higher estimation accuracy;

[0049] The WOA-RF model proposed in the present invention is compared with the multivariate linear regression model and the random forest model. Referring to Table 1, the comparison table of rice prediction results of the three models shows that the determination coefficient R of the WOA-RF model in the validation set is 2 Reached 0.97, corrected root mean square error The maximum value of mean absolute error percentage MAPE was only 12.96%. The determination coefficient R of WOA-RF model was higher than that of multivariate linear regression model. 2 The WOA-RF model improved by 49.2% and 38.6% compared with the random forest model. The WOA-RF model reduced the corrected root mean square error by 80% compared with the multivariate linear regression model and by 73.3% compared with the random forest model.

[0050] Table 1

[0051]

[0052] Since the coefficient of determination R 2 The closer it is to 1, the better the fit between the model prediction value and the true value, the best fitting coefficient, and the corrected root mean square error. When it is closest to 0, it means that the predicted value is closer to the true value, and the lower the mean absolute error percentage MAPE is, the more accurate the prediction accuracy is. It can be seen that the prediction accuracy of the WOA-RF model of the present invention is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0054] Figure 1 A flowchart of a rice yield estimation method using an improved whale optimization algorithm to optimize the RF model;

[0055] Figure 2 A schematic diagram of an embodiment of a method for estimating rice yield by optimizing RF model using an improved whale optimization algorithm;

[0056] Figure 3 This is an estimated rice yield chart. DETAILED DESCRIPTION

[0057] In order to make the technical solutions and advantages of the embodiments of the present invention more clearly understood, the exemplary embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than an exhaustive list of all the embodiments. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.

[0058] refer to Figure 1-Figure 3 The present embodiment is described in detail, a rice yield estimation method for optimizing the RF model using an improved whale optimization algorithm, specifically comprising the following steps:

[0059] S1. Use drones to obtain multispectral image data of rice at different growth stages, and calculate rice growth parameters and average yield after rice maturity;

[0060] S2. Perform normalization preprocessing on the multispectral image data and construct a sample data set;

[0061] S3. Improve the whale optimization algorithm and construct an improved whale optimization algorithm;

[0062] S4. Based on the improved whale optimization algorithm, the RF model is optimized, the model parameters are set, and the WOA-RF model is obtained. The sample data set is input into the WOA-RF model for training, and the optimal rice yield estimation model is screened;

[0063] S5. Use the optimal rice yield estimation model to estimate the rice yield in the target area, generate a rice yield estimation map, and complete the rice yield estimation.

[0064] Furthermore, in S1, the drone is equipped with a multispectral camera, and the rice growth parameters include SPAD values ​​of rice leaves, aboveground biomass, plant height and leaf area index.

[0065] Furthermore, the step S2 specifically includes the following steps:

[0066] S21. Use the Environmental Visual Analysis software (ENVI) to pre-process the multispectral image data, i.e., radiation correction, geometric correction, image stitching, image cropping, etc., to obtain the orthophoto map, which is then converted into a vegetation index map to obtain the reflectance data set;

[0067] S22. Using the Band Math tool of the environmental visual analysis software, based on the reflectance data set and combined with the calculation formula of the vegetation index, band calculation was performed to obtain the average return on investment (ROI) value of the vegetation index map of each sample area as the corresponding vegetation index data set of different growth periods in the region;

[0068] Specifically, the vegetation index includes the normalized difference vegetation index NDVI, the ratio vegetation index RVI, the difference vegetation index DVI, the soil adjusted vegetation index SAVI, the optimized soil adjusted vegetation index OSAVI, the green band normalized vegetation index GNDVI, the two-band enhanced vegetation index EVI2, and the secondary improved soil adjusted vegetation index MSAVI2;

[0069] The normalized difference vegetation index NDVI is expressed as:

[0070] NDVI = (NIR – RED) / (NIR + RED)

[0071] Among them, NIR is the reflectivity of the near-infrared band, and RED is the reflectivity of the red band;

[0072] The ratio vegetation index RVI is expressed as:

[0073] RVI=NIR / Red

[0074] The difference vegetation index DVI is expressed as:

[0075] DVI=NIR-Red

[0076] The soil adjusted vegetation index SAVI is expressed as:

[0077] SAVI=(1+L)*(NIR-Red) / (NIR+Red+L)

[0078] Where, L = 0.5;

[0079] The optimized soil adjusted vegetation index OSAVI is expressed as:

[0080] OSAVI=(1+0.16)*(NIR-Red) / (NIR+Red+0.16)

[0081] The green band normalized difference vegetation index GNDVI is expressed as:

[0082] GNDVI=(NIR-Green) / (NIR+Green)

[0083] Among them, GREEN is the reflectivity of the green wave band;

[0084] The two-band enhanced vegetation index EVI2 is expressed as:

[0085] EVI2=2.5*(NIR-Red) / (NIR+2.4*Red+1)*(1-Atb)

[0086] Among them, Atb is the correction factor for atmospheric interference;

[0087] The secondary improved soil adjusted vegetation index MSAVI2 is expressed as:

[0088] MSAVI2=(2*NIR+1–sqrt((2*NIR+1)2–8*(NIR–RED))) / 2

[0089] S23. Based on the multispectral image data, the growth monitoring module, disaster monitoring module and soil moisture monitoring module in the ecological monitoring system are used for inversion to obtain the raster data of the growth level of rice, disaster level and soil moisture;

[0090] S24. Based on the reflectance dataset, the corresponding vegetation index datasets of different growth periods in the region and the raster data, a sample dataset is constructed, and the sample dataset is normalized as the input dataset and target dataset of the subsequent rice yield estimation model.

[0091] Furthermore, the step S3 specifically includes the following steps:

[0092] S31. Initialize algorithm parameters, set the whale population size to N, and the maximum number of iterations of the whale optimization algorithm to t max , the constant coefficient is b, the random number is l, and the current global optimal position vector is x best(t) According to the chaotic reverse learning initialization strategy, the initialization whale population (x1, x2, x3..., x N ,i=1,2,3…,N);

[0093] In S31, the chaos reverse learning initialization strategy process is as follows:

[0094] The D-dimensional vector y is randomly generated by the whale optimization algorithm, y={y j ,j=1,2,3....,D}, D is the vector dimension, and the chaotic mapping function y is obtained according to the mapping function Ten i+1,j , y ij is the component of the jth mapping sequence on the i-th row, and generates the chaotic sequence y j ,y j ={y ij ,i=1,2,3....,N};

[0095] The chaotic sequence y jMapped to the original space, we get the chaotic initialization population x, x = {x i ,i=1,2,3....,N}, where x i is the i-th individual in the population, x i ={x ij , j = 1, 2, 3 ...., D}, where x ij is the coordinate value of the jth dimension of the i-th population individual, x ij =x min,j +y i,j *(x max,j -x min,j ), where x min,j For x ij The search upper bound, x max,j For x ij Search lower bound of ;

[0096] According to the chaotic initialization population x, the reverse population ox is calculated, ox = {ox i ,i=1,2,3…,N}, where ox i is the i-th reverse population individual, ox i ={ox ij ,j=1,2,3....,D},ox ij is the coordinate value of the jth dimension of the i-th reverse population individual, ox ij =(x max,j +x min,j -x ij );

[0097] Merge the chaotic initialization population x with the reverse population ox to obtain a new population {xUox i}, calculate the objective function value of the new population and sort it, and select N individuals whose fitness values ​​meet the preset conditions as the initial whale population;

[0098] S32. Calculate the fitness value of each individual in the initialized whale population {F(x i ),i=1,2,3....,N}, record the individual positions that currently meet the preset conditions;

[0099] S33. Calculate the value of the convergence factor α in the whale optimization algorithm, Among them, α 初 is the initial value of the convergence factor α, α 终 is the final value of the convergence factor α, y (t) is the chaotic sequence generated by the self-logistic mapping function, y (t) ∈(-1,1), t is the current iteration number, t max is the maximum number of iterations;

[0100] S34. Calculate the value of the inertia weight ω in the whale optimization algorithm, Among them, ω 初 is the initial value of the inertia weight ω, ω 终 is the final value of the inertia weight ω, y (t) ∈(-1,1) is the chaotic sequence generated by the self-logistic mapping function;

[0101] S35. Perform chaotic perturbation collaborative update based on convergence factor α and inertia weight ω, and set the current individual position update formula and spiral update position formula;

[0102] In S35, the current individual position update formula is expressed as:

[0103] x (t+1) =ω*x best(t) -A*|C*x best(t) -x (t) |

[0104] Among them, x (t+1) is the individual position updated by the current individual position formula, x best(t) is the individual optimal position, A is the coefficient vector, C is a random number uniformly distributed in (0,2), x (t) is the current individual position;

[0105] The spiral update position formula is expressed as:

[0106] x (t+1) =ω*x best(t) +D*e bl *cos(2πl)

[0107] Wherein, l is a random number in [-1,1]. When l = -1, the artificial whale is closest to the food, and when l = 1, the artificial whale is farthest from the food. b is the logarithmic spiral shape constant.

[0108] S36. Update the coefficient vectors A, l and C through steps S33 and S34;

[0109] S37. Introduce a random parameter p in the whale optimization algorithm, and the value range of p is {0,1}. If the random parameter p is less than 0.5, enter global exploration. If the value of the coefficient vector of the whale optimization algorithm is |A|<1, the current position is updated according to the current individual position update formula, and the current individual position is updated by the encirclement mechanism. If the value of the coefficient vector of the whale optimization algorithm is |A|>1, the search mechanism is updated according to the search mechanism update formula to update the whale individual position.

[0110] In S37, the search mechanism update formula is expressed as:

[0111] x (t+1) =x rand(t) -A|C*x rand(t) -x (t) |

[0112] Among them, x rand(t) is the position vector of a randomly selected individual whale;

[0113] S38. If the random parameter p ≥ 0.5, enter local development, perform spiral mechanism update on the individual position of the whale, and update the current individual position according to the spiral update position formula;

[0114] S39. Add 1 to the number of iterations of the current whale optimization algorithm, and determine whether the current number of iterations is equal to the maximum number of iterations or the optimal value. If the maximum number of iterations is reached, proceed to step S310, otherwise return to step S32 to continue iterating;

[0115] S310. After the iteration is completed, the optimal solution output by the whale optimization algorithm improved by the above steps is input into the RF model.

[0116] Furthermore, the specific implementation method of step S4 includes the following steps:

[0117] S41. Based on the optimal solution output by the improved whale optimization algorithm, set the RF model parameters, optimize the RF model, and obtain the WOA-RF model;

[0118] S42. Based on the sample data set, the normalized data is used as training samples and input into the WOA-RF model for training, and the optimal WOA-RF model after training is selected as the optimal rice yield estimation model.

[0119] Furthermore, in S5, in order to quantitatively evaluate the effectiveness and accuracy of the optimized WOA-RF model, the determination coefficient R 2 , Corrected Root Mean Square Error The mean absolute percentage error (MAPE) is combined with the optimal rice yield estimation model to evaluate the rice yield in the target area and generate a rice yield estimation map, where RMSE is the root mean square error, is the sample average of rice yield;

[0120] Specifically, the coefficient of determination R 2 It is expressed as:

[0121]

[0122] Among them, k i is the true value of the i-th measured sample, is the predicted value of the i-th sample, and n is the total number of samples;

[0123] Corrected RMS Error It is expressed as:

[0124]

[0125] The mean absolute percentage error (MAPE) is expressed as:

[0126]

[0127] Although the present invention has been described according to a limited number of embodiments, it will be apparent to those skilled in the art, with the benefit of the above description, that other embodiments may be envisioned within the scope of the invention thus described. In addition, it should be noted that the language used in this specification is selected primarily for readability and teaching purposes, rather than for explaining or defining the subject matter of the present invention. Therefore, many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the appended claims. The disclosure of the present invention is illustrative, not restrictive, with respect to the scope of the present invention, which is defined by the appended claims.

Claims

1. A rice yield estimation method using an improved whale optimization algorithm to optimize the RF model, characterized in that: The following steps are involved: S1. Use drones to obtain multispectral image data of rice at different growth stages, and calculate rice growth parameters and average yield after rice maturity; S2. Perform normalization preprocessing on the multispectral image data and construct a sample data set; S3. Improve the whale optimization algorithm and construct an improved whale optimization algorithm; S4. Based on the improved whale optimization algorithm, the RF model is optimized, the model parameters are set, and the WOA-RF model is obtained. The sample data set is input into the WOA-RF model for training, and the optimal rice yield estimation model is screened; S5. Use the optimal rice yield estimation model to estimate the rice yield in the target area, generate a rice yield estimation map, and complete the rice yield estimation.

2. The rice yield estimation method of the improved whale optimization algorithm and RF model according to claim 1 is characterized in that: In S1, the drone is equipped with a multispectral camera, and the rice growth parameters include SPAD value of rice leaves, aboveground biomass, plant height and leaf area index.

3. The rice yield estimation method of the improved whale optimization algorithm and RF model according to claim 2 is characterized in that: The S2 specifically includes the following steps: S21. Use environmental visual analysis software to pre-process the multispectral image data to obtain an orthophoto map, which is then converted into a vegetation index map to obtain a reflectance data set; S22. Using the band mathematical tool of the environmental visual analysis software, based on the reflectance data set, combined with the calculation formula of the vegetation index, band calculation is performed to obtain the average return on investment value of the vegetation index map of each sample area as the corresponding vegetation index data set of different growth periods in the region; S23. Based on the multispectral image data, the growth monitoring module, disaster monitoring module and soil moisture monitoring module in the ecological monitoring system are used for inversion to obtain the raster data of the growth level of rice, disaster level and soil moisture; S24. Construct a sample dataset based on the reflectance dataset, the corresponding vegetation index datasets of different growth periods in the region and the raster data.

4. The rice yield estimation method of the improved whale optimization algorithm and RF model according to claim 3 is characterized in that: The S3 specifically includes the following steps: S31. Initialize algorithm parameters, set the whale population size to N, and the maximum number of iterations of the whale optimization algorithm to t max , the constant coefficient is b, the random number is l, and the current global optimal position vector is x best(t) , according to the chaotic reverse learning initialization strategy, the initialization whale population is obtained; In S31, the chaos reverse learning initialization strategy process is as follows: The D-dimensional vector y is randomly generated by the whale optimization algorithm, y={y j ,j=1,2,3....,D}, D is the vector dimension, and the chaotic mapping function y is obtained according to the mapping function Ten i+1,j , y ij is the component of the jth mapping sequence on the i-th row, and generates the chaotic sequence y j ,y j ={y in ,i=1,2,3…,N}; The chaotic sequence y j Mapped to the original space, we get the chaotic initialization population x, x = {x i ,i=1,2,3....,N}, where x i is the i-th individual in the population, x i ={x ij , j = 1, 2, 3 ...., D}, where x ij is the coordinate value of the jth dimension of the i-th population individual, x ij =x min,j +y i,j *(x max,j -x min,j ), where x min,j For x ij The search upper bound, x max,j For x ij Search lower bound of ; According to the chaotic initialization population x, the reverse population ox is calculated, ox = {ox i ,i=1,2,3....,N}, where ox i is the i-th reverse population individual, ox i ={ox ij ,j=1,2,3....,D},ox ij is the coordinate value of the jth dimension of the i-th reverse population individual, ox ij =(x max,j +x min,j -x ij ); Merge the chaotic initialization population x with the reverse population ox to obtain a new population {xUox i }, calculate the objective function value of the new population and sort it, and select N individuals whose fitness values ​​meet the preset conditions as the initial whale population; S32. Calculate the fitness value of each individual in the initialized whale population {F(x i ),i=1,2,3....,N}, record the individual positions that currently meet the preset conditions; S33. Calculate the value of the convergence factor α in the whale optimization algorithm, Among them, α 初 is the initial value of the convergence factor α, α 终 is the final value of the convergence factor α, y (t) is the chaotic sequence generated by the self-logistic mapping function, y (t) ∈(-1,1), t is the current iteration number, t max is the maximum number of iterations; S34. Calculate the value of the inertia weight ω in the whale optimization algorithm, Among them, ω 初 is the initial value of the inertia weight ω, ω 终 is the final value of the inertia weight ω, y (t) ∈(-1,1) is the chaotic sequence generated by the self-logistic mapping function; S35. Perform chaotic perturbation collaborative update based on convergence factor α and inertia weight ω, and set the current individual position update formula and spiral update position formula; In S35, the current individual position update formula is expressed as: x (t+1) =ω*x best(t) -A*|C*x best(t) -x (t) | Among them, x (t+1) is the individual position updated by the current individual position formula, x best(t) is the individual optimal position, A is the coefficient vector, C is a random number uniformly distributed in (0,2), x (t) is the current individual position; The spiral update position formula is expressed as: x (t+1) =ω*x best(t) +D*e bl *cos(2πl) Where l is a random number in [-1,1], and b is a logarithmic spiral shape constant; S36. Update the coefficient vectors A, l and C through steps S33 and S34; S37. Introduce a random parameter p in the whale optimization algorithm, and the value range of p is {0,1}. If the random parameter p is less than 0.5, enter global exploration. If the value of the coefficient vector of the whale optimization algorithm is |A|<1, the current position is updated according to the current individual position update formula, and the current individual position is updated by the encirclement mechanism. If the value of the coefficient vector of the whale optimization algorithm is |A|>1, the search mechanism is updated according to the search mechanism update formula to update the whale individual position. In S37, the search mechanism update formula is expressed as: x (t+1) =x rand(t) -A|C*x rand(t) -x (t) | Among them, x rand(t) is the position vector of a randomly selected individual whale; S38. If the random parameter p ≥ 0.5, enter local development, perform a spiral mechanism update on the individual position of the whale, and update the current individual position according to the spiral update position formula; S39. Add 1 to the number of iterations of the current whale optimization algorithm, and determine whether the current number of iterations is equal to the maximum number of iterations or the optimal value. If the maximum number of iterations is reached, proceed to step S310, otherwise return to step S32 to continue iterating; S310. After the iteration is completed, the whale optimization algorithm improved by the above steps outputs the optimal solution.

5. The rice yield estimation method of the improved whale optimization algorithm and RF model according to claim 4 is characterized in that: The S4 specifically includes the following steps: S41. Based on the optimal solution output by the improved whale optimization algorithm, set the RF model parameters, optimize the RF model, and obtain the WOA-RF model; S42. Based on the sample data set, the normalized data is used as training samples and input into the WOA-RF model for training, and the optimal WOA-RF model after training is selected as the optimal rice yield estimation model.

6. The rice yield estimation method of the improved whale optimization algorithm and RF model according to claim 5 is characterized in that: In S5, the determination coefficient R is used. 2 , Corrected Root Mean Square Error The mean absolute percentage error (MAPE) is combined with the optimal rice yield estimation model to evaluate the rice yield in the target area and generate a rice yield estimation map, where RMSE is the root mean square error, is the sample average of rice yield.

Images