Peanut growth degree prediction method, system, medium and equipment

Abstract

The invention discloses a peanut growth degree prediction method and system, a medium and equipment, and relates to the technical field of bioengineering, and the method comprises the steps: obtaining remote sensing data and peanut phenological information at a plurality of sampling points of a peanut field, and taking the remote sensing data and the peanut phenological information as a peanut data set; respectively inputting the peanut data set into the trained random forest model and the trained K nearest neighbor model to obtain a first classification prediction value and a second classification prediction value; fusing the first classification prediction value and the second classification prediction value through a feature fusion model, and obtaining features of a fusion result by using an attention mechanism for the fusion result; and inputting the features of the fusion result into the trained multi-target regression model to predict the biomass and maturity of the peanuts, and completing the prediction of the growth degree of the peanuts. The method can better predict the performance and the ability to learn from a plurality of tasks, thereby reducing the overfitting.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bioengineering, and in particular to a peanut growth degree prediction method, system, medium and device. BACKGROUND

[0002] The crop harvesting time has different degrees of influence on the yield and quality of different crops, and the current traditional crop maturity prediction method cannot meet the needs of crop production management optimization. Satellite remote sensing technology can be used to predict the maturity of crops planted in large areas, which has more practical significance for large-area planting of food crops. On the one hand, it can improve crop yield and quality by optimizing harvesting time; on the other hand, it can be used for regional crop harvesting equipment scheduling guidance to support the scheduling of harvesting machinery under the conditions of large-scale agricultural production.

[0003] As a global important economic crop, the precise prediction of peanut growth and maturity is of great significance for optimizing crop management and breeding plans. Traditional peanut maturity determination relies on manual operation of peanut profile boards to classify pods and combine with peanut maturity index. This method is not only labor-intensive and subjective, but also requires a large number of intensive sampling, because the maturity rate at different locations in the field varies. At the same time, direct measurement of peanut biomass is costly and destructive. Therefore, it is necessary to use alternative non-destructive methods to estimate the in-field peanut maturity of different production fields, rather than randomly selecting the peanut maturity of plants. By combining artificial intelligence technology with remote sensing technology, a new method for predicting peanut maturity index is proposed, which can effectively reduce the subjectivity associated with direct measurement methods.

[0004] Although existing research has used satellite or unmanned aerial vehicle sensors to detect changes in crop spectral reflectance to assess crop growth and yield, traditional regression tasks use a set of input variables to predict a single continuous variable, which lacks better prediction performance and the ability to learn from multiple tasks, leading to overfitting results. SUMMARY

[0005] The embodiments of the present application provide a peanut growth degree prediction method, system, medium and device, which have better prediction performance and the ability to learn from multiple tasks, thereby reducing overfitting.

[0006] The embodiments of the present application provide a peanut growth degree prediction method, which comprises the following steps: Obtain remote sensing data and phenology information of peanuts at multiple sampling points in the peanut field as a peanut data set; The peanut dataset is input into the trained random forest model and the trained K nearest neighbor model respectively to obtain first classification prediction values and second classification prediction values; the first classification prediction values and the second classification prediction values are fused through a feature fusion model, and the features of the fusion results are obtained by using an attention mechanism; and the features of the fusion results are input into a trained multi-target regression model to predict the peanut biomass and maturity, and the growth degree of the peanuts is predicted.

[0007] Further, the peanut phenology information specifically includes the following steps: A monitoring area of a peanut field is selected, and the area is divided into a plurality of square grids; and soil grids with significant differences in soil characteristics are dynamically screened out for sample collection according to soil apparent conductivity and bare soil images; From 90 days after sowing, peanut biomass samples are collected in each grid every week until harvest; the peanut biomass includes aboveground biomass and underground biomass; The hull-scrape method and a peanut contour plate are used to evaluate the maturity of the underground biomass, and a peanut maturity index is further obtained, so that ground data collection is completed and used as the peanut phenology information.

[0008] Further, the remote sensing data specifically includes the following steps: High-resolution satellite data is used to extract reflectivity of specific spectral bands at each location on the sampling day, and a plurality of vegetation indices are calculated; The spectral bands of the satellite data include blue, green, red and near-infrared, and the spatial resolution is better than 3 meters; Spectral reflectivity is extracted from a 2-meter buffer zone of each sampling point, and average reflectivity is calculated; image data with a cloud cover of less than 5% is selected as the remote sensing data.

[0009] The embodiment of the present application provides a peanut growth degree prediction system, which comprises: A data acquisition module is configured to acquire remote sensing data and peanut phenology information at a plurality of sampling points in a peanut field as a peanut dataset; A model prediction module is configured to input the peanut dataset into a trained random forest model and a trained K nearest neighbor model respectively to obtain first classification prediction values and second classification prediction values; the first classification prediction values and the second classification prediction values are fused through a feature fusion model, and the features of the fusion results are obtained by using an attention mechanism; and the features of the fusion results are input into a trained multi-target regression model to predict the peanut biomass and maturity, and the growth degree of the peanuts is predicted.

[0010] The embodiment of the present application provides a computer device, which comprises a memory and a processor; the memory stores a computer program, and the processor implements the above-mentioned peanut growth degree prediction method when executing the computer program.

[0011] This invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the peanut growth prediction method described above.

[0012] This invention provides a method, system, medium, and device for predicting peanut growth levels. Compared with existing technologies, its advantages are as follows: The combination of random forest model and K nearest neighbor model with attention mechanism can fully learn the multi-source data features of remote sensing data and peanut phenological information, and establish a nonlinear relationship between multi-source data features and peanut biomass and maturity through multi-objective regression model. By using multi-source data features as input variables, multiple output variables can be accurately predicted, preventing overfitting. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the process structure provided in an embodiment of the present invention; Figure 2 This is a structural diagram of the multi-objective regression model provided in an embodiment of the present invention; Figure 3 The flowchart illustrates the use of the multi-objective regression model provided in this embodiment of the invention. Detailed Implementation

[0014] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0015] like Figure 1 As shown in the figure, this invention provides a method for predicting the growth level of peanuts, including the following steps: Step 1: Obtain remote sensing data and peanut phenological information from multiple sampling points in the peanut field to create a peanut dataset.

[0016] Step 2: Input the peanut dataset into the trained random forest model and the trained K-nearest neighbor model respectively to obtain the first classification prediction value and the second classification prediction value; fuse the first classification prediction value and the second classification prediction value through a feature fusion model, and use an attention mechanism to obtain the features of the fusion result; input the features of the fusion result into the trained multi-objective regression model to predict peanut biomass and maturity, and complete the prediction of peanut growth degree.

[0017] 1. Data acquisition and processing.

[0018] Select monitoring areas, and carry out data collection work during the peanut growing season. Data collection includes ground data and remote sensing data.

[0019] Ground data: Use ARCGIS professional tools to divide each monitoring area into multiple square grids. According to the apparent soil conductivity and bare soil image, select 20 grids with significant soil characteristics differences to collect data, and use the positioning tool to record the specific sampling point position. From 90 days after sowing, collect peanut biomass samples in a 1.0 meter long area in each grid every week until harvest (140 days after sowing), and subdivide the peanut biomass into aboveground biomass (leaves and stems) and underground biomass (pods). At the same time, use the hull-scrape method and peanut contour plate to manually assess the maturity of 100 pod samples extracted from the 1.0 meter biomass sample, and calculate the peanut maturity index (PMI).

[0020] .

[0021] In the formula, N is the number of mature pods, and T is the total number of pods. At the same time, record the phenological stages of peanuts in detail, such as sowing period, seedling period, flowering period, podding period, and maturation period.

[0022] Remote sensing data: The spectral bands of multispectral satellite data should at least cover blue (455-515 nm), green (500-590 nm), red (590-670 nm), and near-infrared (780-860 nm), with a spatial resolution better than 5 meters and a format of 16-bit GeoTiff. Select image data with less than 5% cloud cover. Process the obtained remote sensing data through radiation calibration, atmospheric correction, geometric correction, orthorectification, image fusion, etc., then use the research area range vector boundary to crop the fusion result, and finally obtain a multispectral image with geographic coordinates and high resolution.

[0023] 2. Vegetation index calculation.

[0024] In order to more effectively reflect the growth state of crops, it is necessary to calculate multiple vegetation indices. Commonly used vegetation indices include normalized difference vegetation index (NDVI), nonlinear vegetation index (NLI), green normalized difference vegetation index (GNDVI), modified nonlinear index (MNLI), soil-adjusted vegetation index (SAVI), and spectral band (SB) as part of the characteristic variable. Use ENVI software to complete reflectance data extraction and vegetation index calculation. The vegetation index calculation method is as follows:

[0025] .

[0026] .

[0027] .

[0028] .

[0029] .

[0030] In the formula, RED is the red band emissivity; NIR is the near-infrared band emissivity; Green is the green band emissivity; L is the soil adjustment coefficient, generally 0.5.

[0031] 3. Data extraction and integration.

[0032] With the sampling point as the center, a 5-meter buffer zone is set. Through professional geographic information processing software, the spectral reflectance data and vegetation index values of the 5-meter buffer zone of each sampling point are extracted from the processed satellite images. Since the buffer zone contains multiple pixels, the average value of the spectral reflectance of these pixels is calculated to represent the spectral reflectance characteristics of the sampling point. Ensure that each sampling point has both the extracted average spectral reflectance and the in-situ data of peanut biomass, maturity, etc. collected at the point, to build a complete data set, presented in table form. In the table, each row represents a data record of a sampling point, containing remote sensing data (spectral reflectance, VI value) and in-situ data (biomass, maturity) of the point.

[0033] 4. Data standardization: In order to make different characteristics of data comparable and improve the effect of model training, it is necessary to standardize the remote sensing data. StandardScaler method is used to standardize the data.

[0034] The first step is to calculate the mean and standard deviation of each feature (such as vegetation index, spectral band reflectance, etc.).

[0035] The second step is to transform the data according to the calculated mean and standard deviation according to the rules of StandardScale. For each data point of each feature, use the formula to calculate, where is the original data point, is the mean of the feature, is the standard deviation of the feature, is the standardized data point.

[0036] 5. Data set division: divide the processed data into training set and validation set according to certain proportion, such as 80% as training set for model learning and 20% as validation set for model performance evaluation.

[0037] 6. Multi-objective regression model construction and training.

[0038] A multi-target regression model was trained to predict ground and underground peanut biomass and peanut maturity index (PMI) using vegetation indices, spectral bands, and phenology information as input variables. Before the data was input into the trained model, the attention module was used to calculate the attention weight of each feature.

[0039] Random Forest (RF) algorithm: The first step is to determine the model framework, the second step is to initialize the model parameters using the RF tool in the Scikit-learn library, set the initial values such as the number of decision trees and the maximum depth. The third step is to input the training set data into the model, and the model constructs multiple decision trees based on the training data. Each decision tree selects a part of the features and samples for splitting during training, and learns the complex relationship between the input features (vegetation indices, spectral bands, etc.) and multiple target variables (peanut biomass and maturity, etc.) in this way. The fourth step is to optimize the hyperparameters using the Grid Search CV method, and the hyperparameter range of the RF algorithm includes: n estimators (1-30), max depth (1-30), min samples split (2, 4, 6), min samples leaf (1-2), min weight fraction leaf (0.1-0.5).

[0040] K-Nearest Neighbors (KNN) algorithm: The first step is to set the basic parameters of the model, determine the initial values of the parameters such as the number of neighbors k and the leaf node size. The second step is to calculate the distance between each sample in the training set and other samples. When there is a new data point to be predicted, the KNN algorithm finds the k nearest neighbors of the point in the training set, and according to the target variable values of these k nearest neighbor samples, it predicts the target variable values of the new data point by averaging or weighted averaging. The third step is to optimize the hyperparameters using the Grid Search CV method, and the hyperparameter range of the KNN algorithm includes: n neighbors (1-20), leaf size (1-30), p (1, 2).

[0041] The specific implementation of the attention module can be done in the following steps: (1) Linear transformation is performed on the input feature vector X (including spectral band reflectance, vegetation index, and peanut phenology information) to obtain the query vector Q, the key vector K, and the value vector V: , , where, , , are the learnable weight matrices.

[0042] (2) Calculate the attention score seores: where, is the dimension of the key vector, The function is used to normalize the score into a probability distribution.

[0043] (3) Calculate the weighted sum according to the attention score: The feature vector processed by the attention mechanism is input into the random forest (RF) and K nearest neighbor (KNN) algorithms for training.

[0044] 7. Model evaluation and optimal model determination.

[0045] First step model evaluation: the mean absolute error (MAE) is used as the main evaluation index to calculate the average error between the predicted value and the actual value of the model. Specifically, the predicted value and the actual observation value are substituted into the MAE calculation formula: where n is the number of data, is the predicted value of the algorithm, is the observed value.

[0046] The second step is to determine the optimal model: the prediction results of different input combinations (such as only using spectral bands, only using vegetation indices, and combining spectral bands and vegetation indices) and different algorithms (RF and KNN) are compared in detail. By calculating the MAE value under different conditions, the error degree of model prediction is accurately quantified, and then the model with the lowest MAE is selected as the final peanut biomass and maturity prediction model, ensuring that the model has high accuracy and reliability.

[0047] The specific steps of data input model and model processing data are as follows: 1. Data preparation and preprocessing: the peanut data set to be predicted is preprocessed according to the processing method of the training set data, including calculating vegetation indices using the same method, standardizing spectral reflectance data, etc.

[0048] 2. Attention mechanism processing: input the preprocessed data to be predicted into the attention module to calculate the attention weight of each feature (including spectral band reflectance, vegetation index and peanut phenology information). The specific process is to perform linear transformation on the input feature vector to obtain query vector Q, key vector K and value vector V, calculate attention score scores, and then calculate the weighted sum according to the score to obtain the feature vector processed by the attention mechanism.

[0049] 3. Model prediction: After the feature vector is processed by the attention mechanism, it is input into the trained multi-target regression model (random forest or K-nearest neighbor model). Taking the random forest model as an example, multiple decision trees in the model analyze the input features, and each decision tree processes the input data according to the rules it has learned. The final prediction value is obtained by combining the results of multiple decision trees. For the K-nearest neighbor model, when the input data point is input, the model finds the k nearest samples in the training set, and according to the target variable values (peanut biomass and maturity, etc.) of these k nearest neighbors, it predicts the target variable values of the new data point by averaging or weighted averaging, etc., to obtain the growth degree of peanuts (biomass and maturity related indicators).

[0050] 4. Model accuracy evaluation (prediction phase): After obtaining the prediction results, the performance of the model on the prediction data is evaluated again using evaluation indicators such as MAE. Compare the predicted values with the actual peanut biomass and maturity data (if there are actual measurement values) to calculate MAE. If the MAE value of the prediction result is within the acceptable range, the model performs well in the actual prediction task, and the prediction result is reliable; if the MAE value exceeds the expectation, further analysis is needed, which may be due to problems in data preprocessing, insufficient model training, or the current model is not suitable for peanut growth prediction in this area, and the model needs to be adjusted or more data needs to be collected for training and optimization.

[0051] The present application incorporates a number of innovations in the technical solution, as follows: 1. For the first time, the multi-target regression (MTR, Multi-Target Regression) model is introduced into the field of joint prediction of peanut biomass and maturity, and is combined with peanut phenology information.

[0052] The present application uses a multi-target regression model to make joint prediction of peanut biomass and maturity. By simultaneously optimizing the prediction tasks of aboveground biomass, underground biomass and flower maturity index (PMI), it effectively solves the technical bottleneck that traditional single-target regression models cannot capture the correlation between multiple variables. The peanut phenology information reflects the characteristics of peanuts at different growth stages, and its combination with the multi-target regression model enables the model to fully learn the internal relationship between phenology information and biomass, maturity, significantly improving the prediction efficiency and model generalization ability, and providing more comprehensive and accurate decision-making basis for peanut planting management.

[0053] 2. Differentiated data collection strategy based on soil characteristics.

[0054] Innovatively combining soil apparent conductivity with bare soil image analysis, the selected monitoring area is divided into multiple square grids. Through comprehensive analysis of soil apparent conductivity and bare soil images, 20 grids with significant differences in soil characteristics are dynamically selected for sample collection. This collection strategy avoids the blindness of traditional random sampling, ensuring that the collected data set is highly representative and diverse, providing a high-precision and reliable input basis for subsequent model training, which helps to improve the accuracy of model prediction.

[0055] 3. Fusion and optimization of multispectral and vegetation index.

[0056] A collaborative input scheme of spectral bands and five specific vegetation indices (normalized difference vegetation index (NDVI), nonlinear vegetation index (NLI), green normalized difference vegetation index (GNDVI), modified nonlinear index (MNLI), and soil-adjusted vegetation index (SAVI)) is proposed. In calculating the vegetation index, through nonlinear combination and optimization of the soil adjustment coefficient (such as the L value in SAVI), the internal relationship between spectral information and vegetation growth conditions is fully explored. This innovation solves the problem of insufficient adaptability of single vegetation index to complex field environment, significantly enhances the sensitivity of the model to peanut growth dynamics, and enables the model to more accurately reflect the growth status of peanuts.

[0057] 4. Intelligent hyperparameter tuning and model lightweight design.

[0058] When building a multi-objective regression model, Grid Search CV method is used to fine-tune the hyperparameters of Random Forest (RF) and K-Nearest Neighbors (KNN) algorithms. For example, for RF algorithm, dynamically adjust n_estimators (1-30), max depth (1-30) and other parameters; for KNN algorithm, realize adaptive expansion of neighborhood, adjust n neighbors (1-20) and other parameters. At the same time, combined with Standard Scaler method for data set normalization processing, and using cross-validation technology (20% data for cross-validation), the prediction accuracy is guaranteed while the risk of overfitting is greatly reduced. Through these operations, the balance between model performance and computing efficiency is achieved, the purpose of model lightweight design is achieved, and the feasibility of the model in practical application is improved.

[0059] 5. Non-destructive prediction and agricultural management closed-loop optimization.

[0060] Through the deep integration of satellite remote sensing technology and artificial intelligence, the full-cycle non-destructive monitoring of peanut biomass and maturity is realized. Using high-resolution satellite data and advanced image processing technology, the growth information of peanut plants can be obtained without direct destruction. Combined with the prediction results of the model output, dynamic decision support can be provided for precision agriculture. For example, farmers can optimize irrigation, fertilization, and harvesting time according to the growth stage and maturity of peanuts, avoiding resource waste, and thus improving peanut yield and quality, forming a closed-loop system of "monitoring-prediction-management-efficiency", and promoting the development of peanut planting towards intelligence and efficiency.

[0061] 6. Innovative application of high-resolution satellite data.

[0062] Multispectral satellite data with a spatial resolution better than 10 meters is used. This high-resolution data can capture more subtle information in the field. During data processing, a strategy of extracting reflectance from a 5-meter buffer zone around each sampling point and calculating the average reflectance is used to effectively overcome the spatial heterogeneity problem of traditional remote sensing data in small-scale field monitoring. Through the innovative application of high-resolution data, the model's ability to capture local maturity differences in the field is significantly improved, making the prediction results more in line with actual planting conditions and providing strong support for fine peanut planting management.

[0063] Compared with existing technologies, the present invention has significant advantages in predicting the growth degree of peanuts, effectively overcoming many drawbacks of traditional methods, and playing an important role in improving prediction accuracy, reducing costs, enhancing objectivity, and optimizing agricultural management, as follows: 1. Higher prediction accuracy: Combined with random forest and K-nearest neighbor algorithm, and introducing attention mechanism to build a multi-objective regression model. This model can fully learn the characteristics of multi-source data such as spectral bands, vegetation indices, and peanut phenology information, and mine the complex nonlinear relationship between input variables and peanut biomass and maturity. Through Grid Search CV method to optimize model hyperparameters, further improve model performance, compared with traditional regression task using only one set of input variables to predict a single continuous variable, can more accurately predict the growth degree of peanuts, reduce prediction error.

[0064] 2. Reduce labor and economic costs: Traditional peanut maturity determination relies on manual operation of peanut profile board to classify pods and combine with peanut maturity index, which not only requires a lot of effort, but also requires a large number of intensive sampling; direct measurement of peanut biomass is costly and destructive. The present invention uses a non-destructive method to obtain data through satellite remote sensing and combines artificial intelligence technology for prediction, reducing the workload of manual sampling and direct measurement, reducing labor costs and losses caused by measurement on peanuts, and saving economic costs.

[0065] 3. Reducing subjectivity: Traditional manual flower maturity determination is highly subjective, and different personnel may have different judgment standards. The present application uses multi-source data and objective model algorithms for prediction, reducing the interference of human factors and making the prediction results more objective and reliable.

[0066] 4. Providing more comprehensive information: The biomass and maturity of peanuts can be predicted simultaneously, providing more comprehensive peanut growth information for growers. Growers can accurately grasp the growth status of peanuts based on this information and make scientific and reasonable planting management decisions, such as precise irrigation, fertilization, and harvesting time arrangement, to achieve closed-loop optimization of agricultural management and improve peanut yield and quality.

[0067] 5. Suitable for large-scale planting management: Using satellite remote sensing technology, the growth degree of peanuts planted in large areas can be predicted. This is of great significance for large-scale agricultural production and can provide strong guidance for the scheduling of regional crop harvesting equipment, reasonably arranging harvesting machinery, and improving agricultural production efficiency.

[0068] The embodiment of the present application provides a peanut growth degree prediction system, comprising: A data acquisition module is used to acquire remote sensing data and phenological information of peanuts at multiple sampling points in a peanut field as a peanut data set.

[0069] A model prediction module is used to input the peanut data set into a trained random forest model and a trained K-nearest neighbor model to obtain first and second classification prediction values; the first and second classification prediction values are fused through a feature fusion model, and the features of the fusion results are obtained using an attention mechanism; the features of the fusion results are input into a trained multi-objective regression model to predict the biomass and maturity of peanuts, and the prediction of the growth degree of peanuts is completed.

[0070] The embodiment of the present application provides a computer device, comprising a memory and a processor; the memory stores a computer program, and the processor implements the steps of a peanut growth degree prediction method when executing the computer program.

[0071] The embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of a peanut growth degree prediction method.

[0072] A specific implementation is as follows: The embodiment discloses a peanut growth degree prediction method, and the specific steps are as follows: S1, sow peanuts in an experimental field, and perform irrigation, fertilization and other agricultural operations according to local conventional planting management specifications. At the same time, record the sowing date of the peanuts and the starting time of each subsequent phenological stage in detail.

[0073] S2, at the specified time node, use professional sampling instruments to collect peanut biomass samples in the selected grid, ensure the integrity and representativeness of the samples. At the same time, carefully collect pod samples for maturity assessment to prevent damage. Record the phenological stage of peanuts at each sampling time.

[0074] S3, obtain satellite image data, and perform radiation calibration, atmospheric correction and other preprocessing operations through professional image processing software to ensure the accuracy of reflectance data. Then extract spectral band reflectance and calculate vegetation index according to the set method.

[0075] S4, input the processed data into the attention module, calculate the attention weight of each feature (including spectral band reflectance, vegetation index and phenological information of peanut), and get the feature vector after attention mechanism processing.

[0076] S5, divide the processed data into training set and validation set according to the proportion of 80% and 20%, write code in Scikit-learn library environment to build multi-objective regression model of RF and KNN algorithm, set the corresponding hyperparameter range, run the training and cross-validation process, and record the MAE value under different parameter combinations. In the training process, make full use of hyperparameter intelligent optimization technology, constantly adjust the parameters to optimize the model performance, and consider the influence of phenological information on hyperparameters.

[0077] S6, according to the evaluation results, select the model configuration with the lowest MAE as the final model, and test and verify it on new data set to ensure the stability and generalization ability of the model. Use the model prediction results and phenological information of peanuts to provide more accurate irrigation, fertilization and harvesting time optimization suggestions for farmers, and realize closed-loop optimization of agricultural management.

[0078] Figure 2 The process of multi-objective regression model from data input to prediction result output is shown. The peanut data set first enters the data preprocessing module for cleaning, encoding and normalization processing. The processed data is input into the random forest module and the K nearest neighbor module respectively. The random forest module constructs decision tree, and the K nearest neighbor module calculates sample distance. The outputs of the two modules are fused in the feature fusion module, and then enter the attention mechanism module to calculate the weighted sum. Finally, the weighted sum feature is input into the multi-objective regression module to predict peanut biomass and maturity and output the results.

[0079] Figure 3The complete process from data acquisition to model application is shown in detail. First, the data is acquired, and then the data is sequentially cleaned, encoded, and normalized. Then, the data set is divided and the model is initialized. Random forest and K nearest neighbor models are trained respectively, and their outputs are fused and processed by an attention mechanism. Then, a multi-objective regression model is trained. Next, the hyperparameter range is defined, and cross-validation is performed by traversing the hyperparameter combinations to select the optimal hyperparameters to update the model. Then, the model performance is evaluated, and if it meets the requirements, the model is applied, otherwise it is adjusted, and the process ends.

[0080] The above embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it cannot be understood as a limitation on the scope of the patent. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method of predicting the degree of growth of peanuts, characterized by, The method comprises the following steps: Obtaining remote sensing data and peanut phenology information at multiple sampling points in a peanut field as a peanut data set; Inputting the peanut data set into a trained random forest model and a trained K-nearest neighbor model respectively to obtain first classification prediction values and second classification prediction values; Fusing the first classification prediction values and the second classification prediction values through a feature fusion model, using an attention mechanism to obtain features of the fusion results, and inputting the features of the fusion results into a trained multi-objective regression model to predict peanut biomass and maturity, thereby completing the prediction of the growth degree of peanuts.

2. The peanut growth degree prediction method of claim 1, wherein, The specific steps for obtaining the peanut phenology information include: Selecting a monitoring area in the peanut field, dividing the area into multiple square grids, and dynamically screening grids with significant soil property differences for sample collection according to soil apparent conductivity and bare soil images; Collecting peanut biomass samples in each grid every week from 90 days after sowing until harvesting; the peanut biomass includes aboveground biomass and underground biomass; Using the hull-scrape method and a peanut contour plate to assess the maturity of underground biomass and further obtain a peanut maturity index, thereby completing the collection of ground data and obtaining the peanut phenology information.

3. The peanut growth degree prediction method of claim 1, wherein, The specific steps for obtaining the remote sensing data include: Using high-resolution satellite data to extract reflectivity of specific spectral bands at each location on the sampling day and calculate multiple vegetation indices; The spectral bands of the satellite data include blue, green, red, and near-infrared, and the spatial resolution is better than 3 meters; Extracting spectral reflectivity from a 2-meter buffer zone around each sampling point and calculating the average reflectivity, and selecting image data with less than 5% cloud cover as the remote sensing data.

4. A peanut growth degree prediction system characterized by, The method comprises: A data acquisition module for obtaining remote sensing data and peanut phenology information at multiple sampling points in a peanut field as a peanut data set; A model prediction module for inputting the peanut data set into a trained random forest model and a trained K-nearest neighbor model respectively to obtain first classification prediction values and second classification prediction values; Fusing the first classification prediction values and the second classification prediction values through a feature fusion model, using an attention mechanism to obtain features of the fusion results, and inputting the features of the fusion results into a trained multi-objective regression model to predict peanut biomass and maturity, thereby completing the prediction of the growth degree of peanuts.

5. A computer device comprising: A memory and a processor; The memory stores a computer program, and the processor executes the computer program to implement the peanut growth degree prediction method of any one of claims 1-3.

6. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 5. The computer program is executed by the processor to implement the peanut growth degree prediction method of any one of claims 1-3.

Images