Sugarcane yield prediction method and system based on machine learning

Abstract

The invention discloses a sugarcane yield prediction method and system based on machine learning, and relates to the technical field of intelligent agriculture. Through comprehensive application of technical means of data acquisition, preprocessing, feature extraction, model construction, training, prediction, decision support and the like, accurate prediction of the sugarcane yield is realized. According to the method, multi-source data is combined, key features are extracted by using an advanced machine learning algorithm, and a high-performance prediction model is constructed. Through historical data training and model parameter optimization, prediction accuracy and stability are improved. Meanwhile, the system provides a user-friendly interactive interface, and collects user feedback to continuously improve the model. The method is suitable for the field of agriculture, especially for sugarcane yield prediction, solves the problem that traditional methods such as experience estimation and statistical model prediction are inaccurate, realizes high-precision prediction through technologies such as machine learning, and has important significance for promoting agricultural sustainable development.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart agriculture, and in particular relates to a sugarcane yield prediction method and system based on machine learning. Background Art

[0002] With the development of agricultural informatization, precision agriculture has become an important trend in modern agriculture. As an important sugar crop, the prediction of sugarcane yield and quality is of great significance to sugar production. As one of the important economic crops in southern my country, sugarcane is widely used in beverage, sugar making and other industries. However, sugarcane growth is affected by many factors, such as meteorological conditions, soil environment, crop genotype, etc., which makes it difficult to accurately predict sugarcane yield. Traditional prediction methods mostly rely on manual experience and simple statistics, which are difficult to meet the high-precision requirements of modern agricultural production. In recent years, machine learning (ML) and deep learning (DL) technologies have been widely used in crop yield prediction because they can identify nonlinear patterns in massive data sets. However, the existing methods for predicting sugarcane yield and quality mostly focus on a single data source or algorithm, lacking comprehensiveness and systematicity. Summary of the invention

[0003] The present invention aims to provide a sugarcane yield prediction method and system based on machine learning, which can achieve accurate prediction of sugarcane yield by collecting multi-source data, extracting key features, and constructing and optimizing prediction models, and provide decision support for agricultural production management.

[0004] In order to solve the above technical problems, the present invention is achieved through the following technical solutions:

[0005] The present invention is a sugarcane yield prediction method based on machine learning, comprising the following steps:

[0006] Step 1: Data collection and preprocessing: Collect crop gene data, drone hyperspectral image data, satellite remote sensing data, ground observation data, meteorological data, and historical yield data during the sugarcane growth cycle, classify and store the collected data, and preprocess the relevant data one by one, merge the multi-source data, and form a multidimensional data set;

[0007] Step 2: Feature extraction: Extract key features closely related to sugarcane yield and quality from the preprocessed data, use feature selection algorithm to further select feature subsets that contribute most to the prediction results and perform dimensionality reduction processing;

[0008] Step 3: Model selection and construction: According to different prediction targets, different feature subsets are selected and appropriate machine learning algorithms or algorithm combinations are used to construct prediction models. The machine learning algorithms include but are not limited to: gradient boosting regression tree, partial least squares regression, support vector machine regression, BP neural network, long short-term memory network, deep extreme learning machine, multivariate linear regression, stepwise multivariate regression, generalized linear model, generalized boosting model and random forest algorithm.

[0009] Step 4: Model training and tuning: Use historical data to train the model and tune the model parameters through cross-validation;

[0010] Step 5: Model prediction, integration and fusion: Use the trained prediction model to predict sugarcane yield, and use ensemble learning or model fusion methods to integrate the prediction results of multiple different prediction models to improve the accuracy and stability of the prediction;

[0011] Step 6: Forecast report generation: According to the selected integrated strategy, the final sugarcane yield forecast value is calculated, and according to the forecast results and influencing factors analysis, corresponding agricultural management suggestions and optimization measures are proposed;

[0012] Step 7: Result verification and evaluation: Verify and evaluate the prediction results by comparing with the actual production data, and use a variety of evaluation indicators to verify and evaluate the accuracy and reliability of the prediction model;

[0013] Step 8. Continuous optimization and improvement: Establish a feedback mechanism to collect user feedback on the prediction results, and continuously optimize and improve the model based on the feedback. At the same time, track the development of new technologies and new methods, and introduce new technologies into the prediction model in a timely manner to improve performance.

[0014] As a preferred technical solution of the present invention, the data collection and preprocessing steps specifically include:

[0015] Collection of UAV hyperspectral image data: During the sugarcane growth cycle, the hyperspectral image data of the sugarcane field is obtained by using a UAV equipped with a hyperspectral camera to capture the spectral information of the sugarcane growth process. The UAV hyperspectral image data includes but is not limited to canopy coverage, canopy height, normalized vegetation index, and overgreening index;

[0016] Acquisition of satellite remote sensing data: Using satellite remote sensing technology, remote sensing image data of sugarcane fields is regularly collected to obtain sugarcane growth information over a wider range;

[0017] Collection of ground observation data: by collecting actual measurements of key sugarcane phenotypes in sugarcane fields, including but not limited to sugarcane growth parameters, yield-related indicators, and soil and environmental parameters;

[0018] Meteorological data collection: Collect meteorological data during the sugarcane growth cycle, including but not limited to temperature, precipitation and light, to analyze the impact of meteorological factors on sugarcane yield;

[0019] Data classification storage: Classify and store various types of collected data for subsequent processing;

[0020] Data preprocessing: Clean the collected data, remove outliers, fill in missing values ​​and other preprocessing operations to improve data quality;

[0021] Multi-source data fusion: The pre-processed multi-source data are fused to form a multidimensional data set containing a variety of information, providing data support for subsequent feature extraction and model building. At the same time, advanced spatiotemporal interpolation methods are used to deeply fuse spatiotemporal data to capture the complex relationship between spatial distribution and temporal evolution during sugarcane growth.

[0022] As a preferred technical solution of the present invention, the following machine learning algorithm is selected in the data preprocessing stage to process the data collected by different collection methods:

[0023] For UAV hyperspectral image data, a convolutional neural network combined with a deep learning framework is used for feature extraction to fully utilize the spectral and spatial information in the hyperspectral image and improve the accuracy of identifying the growth status of sugarcane.

[0024] For satellite remote sensing data, long short-term memory networks or recurrent neural networks combined with attention mechanisms are used to capture long-term dependencies and changes at important time points in time series data to accurately predict sugarcane growth trends;

[0025] For continuous variables in ground observation data, including but not limited to soil moisture and nutrient content, support vector machine regression or random forest regression is used to handle the nonlinearity and interaction of these data and improve the accuracy of yield prediction;

[0026] For categorical variables in meteorological data, including but not limited to rainfall type and wind direction, decision trees or gradient boosting trees are used for classification processing to accurately determine the impact of meteorological conditions on sugarcane yield;

[0027] For time series meteorological data, including but not limited to daily rainfall and daily average temperature, the time series analysis algorithm ARIMA or the seasonal decomposition time series prediction model SARI MA is combined with a deep learning model to capture the seasonal and cyclical changes of meteorological data and improve the accuracy of the prediction.

[0028] As a preferred technical solution of the present invention, in the feature extraction stage, a combined model (ConvLSTM) of a convolutional neural network (CNN) and a recurrent neural network (RNN) is used to extract spatiotemporal features according to the characteristics of spatiotemporal data, so as to effectively extract the spatiotemporal characteristics of the sugarcane growth process.

[0029] As a preferred technical solution of the present invention, in the feature extraction stage, an unsupervised learning method is used to further reduce the dimension of the features to reduce the computational complexity and retain key information, while exploring the potential structures and patterns in the data.

[0030] As a preferred technical solution of the present invention, in the model selection and construction step, for different feature subsets, the following machine learning algorithms are selected for one-to-one correspondence:

[0031] For spectral features, gradient boosting regression tree was selected to build the prediction model;

[0032] For spatial features, support vector machine regression was selected to build a prediction model;

[0033] For environmental characteristics, multiple linear regression was selected to build a prediction model.

[0034] As a preferred technical solution of the present invention, in the model training step, the model is trained using historical data, and the model parameters are tuned by cross-validation, which specifically includes:

[0035] The historical data is divided into training set, validation set and test set in a ratio of 70%:15%:15%;

[0036] Use the training set to train the model;

[0037] Use the validation set to validate the trained model and adjust the model parameters based on the validation results;

[0038] Validate the model using a test set independent of the training set to evaluate the performance of the model in real-world applications

[0039] Repeat the above process until the model performance reaches the optimal level.

[0040] As a preferred technical solution of the present invention, it also includes predicting the sugarcane yield in a specific area by adopting a transfer learning strategy, using a model trained in other similar areas as a pre-training model, and fine-tuning with a small amount of local data to quickly adapt to the prediction needs of the new area, thereby improving the prediction efficiency and accuracy.

[0041] As a preferred technical solution of the present invention, in the model prediction, integration and fusion steps, the integrated learning method includes but is not limited to stacking, lifting and bagging;

[0042] In the result verification and evaluation step, the evaluation indicators include mean square error, root mean square error, determination coefficient and absolute percentage error.

[0043] The present invention also provides a sugarcane yield prediction system based on machine learning, which is used to implement the above method. The system comprises:

[0044] Data collection module: responsible for automatically collecting relevant data during the sugarcane growth cycle;

[0045] Data preprocessing module: receives data from the data acquisition module, performs preprocessing operations, and fuses multi-source data to form a data set containing multi-dimensional information on crop genetic information, growth status, and environmental parameters, providing a high-quality data foundation for subsequent analysis;

[0046] Feature extraction and model selection module: used to automatically extract key features from preprocessed data using advanced machine learning algorithms, and select appropriate machine learning algorithms or algorithm combinations to build prediction models based on feature extraction results;

[0047] Model training and optimization module: Use historical data to train the model and tune the model parameters through cross-validation and grid search strategies;

[0048] Prediction and decision support module: Use the trained model to predict sugarcane yield, generate detailed prediction reports based on the prediction results and influencing factors, and propose corresponding agricultural management suggestions and optimization measures. It also supports integrated learning and model fusion strategies to improve the accuracy and stability of prediction results.

[0049] User interaction and feedback module: provides a friendly user interface that allows users to input query conditions, view prediction results, and receive agricultural management suggestions. At the same time, this module is also responsible for collecting user feedback on the prediction results and continuously improving and optimizing the prediction model by establishing a feedback mechanism.

[0050] The present invention has the following beneficial effects:

[0051] Aiming at the complexity of sugarcane yield prediction and the limitations of traditional methods, the present invention proposes a sugarcane yield prediction method and system based on machine learning. The method realizes high-precision prediction of sugarcane yield by comprehensively applying multiple data sources and advanced machine learning algorithms. Specifically, the method first integrates multi-dimensional information such as crop gene data and drone hyperspectral image data, effectively improving the comprehensiveness and accuracy of the prediction; secondly, the method uses feature extraction and model selection construction technology to screen out key features closely related to sugarcane yield and build a suitable prediction model; finally, through model training and optimization, prediction and decision support and other steps, the stability and reliability of the prediction results are significantly improved. In addition, the system also supports continuous optimization and improvement, and can dynamically adjust model parameters according to user feedback and new data to adapt to the changing needs of agricultural production. In summary, the patent not only improves the accuracy and efficiency of sugarcane yield prediction, but also provides scientific decision-making support for agricultural production, which has important application value and social benefits.

[0052] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0054] Figure 1 A schematic diagram of a process of predicting sugarcane yield based on machine learning in the present invention;

[0055] Figure 2 This is a flow chart of the steps of training the model using historical data in the model training step of the present invention;

[0056] Figure 3 It is a structural schematic diagram of a sugarcane yield prediction system based on machine learning in the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] like Figure 1As shown: The present invention provides a sugarcane yield prediction method based on machine learning, comprising the following steps:

[0059] Step 1: Data collection and preprocessing: Collect crop gene data, drone hyperspectral image data, satellite remote sensing data, ground observation data, meteorological data, and historical yield data during the sugarcane growth cycle, classify and store the collected data, and preprocess the relevant data one by one, merge the multi-source data, and form a multidimensional data set;

[0060] Step 2: Feature extraction: Extract key features closely related to sugarcane yield and quality from the preprocessed data, use feature selection algorithm to further select feature subsets that contribute most to the prediction results and perform dimensionality reduction processing;

[0061] Step 3: Model selection and construction: According to different prediction targets, different feature subsets are selected and appropriate machine learning algorithms or algorithm combinations are used to construct prediction models. The machine learning algorithms include but are not limited to: gradient boosting regression tree, partial least squares regression, support vector machine regression, BP neural network, long short-term memory network, deep extreme learning machine, multivariate linear regression, stepwise multivariate regression, generalized linear model, generalized boosting model and random forest algorithm.

[0062] Step 4: Model training and tuning: Use historical data to train the model and tune the model parameters through cross-validation;

[0063] Step 5: Model prediction, integration and fusion: Use the trained prediction model to predict sugarcane yield, and use ensemble learning or model fusion methods to integrate the prediction results of multiple different prediction models to improve the accuracy and stability of the prediction;

[0064] Step 6: Forecast report generation: According to the selected integrated strategy, the final sugarcane yield forecast value is calculated, and according to the forecast results and influencing factors analysis, corresponding agricultural management suggestions and optimization measures are proposed;

[0065] Step 7: Result verification and evaluation: Verify and evaluate the prediction results by comparing with the actual production data, and use a variety of evaluation indicators to verify and evaluate the accuracy and reliability of the prediction model;

[0066] Step 8. Continuous optimization and improvement: Establish a feedback mechanism to collect user feedback on the prediction results, and continuously optimize and improve the model based on the feedback. At the same time, track the development of new technologies and new methods, and introduce new technologies into the prediction model in a timely manner to improve performance.

[0067] In this paper, we first collected a large amount of sugarcane growth-related data through various channels, which covered multi-dimensional information from the genetic level to the climate level. In the data preprocessing stage, we cleaned the data, processed missing values ​​and outliers, and used advanced data fusion technology to integrate these heterogeneous data sources into a comprehensive and consistent multidimensional data set.

[0068] In step 2, in the feature extraction phase, we use advanced statistical learning methods to extract key features that can characterize sugarcane yield and quality from complex raw data. This step is crucial for subsequent model construction because it determines whether the model can accurately capture the core factors that affect sugarcane yield, and use feature selection algorithms to further select feature subsets that contribute most to the prediction results for dimensionality reduction. This step helps improve the prediction accuracy and computational efficiency of the model;

[0069] In step 3, during the model selection and construction phase, we selected the most suitable machine learning algorithm and built the corresponding prediction model based on the specific prediction target, such as the per-acre yield or total yield of sugarcane. These algorithms include but are not limited to neural networks that can handle nonlinear relationships and ensemble learning methods that are good at processing large amounts of data, such as random forests. The selection and combination of these algorithms provide us with multiple possibilities to meet the prediction needs in different scenarios.

[0070] In step 4, the model training phase, we trained the model using historical data and tuned the model parameters through cross-validation. This step ensures that the model can better fit the data, thereby improving the accuracy of the prediction.

[0071] In step five, in the model prediction, integration and fusion stage, we used the trained prediction model to predict sugarcane yield and integrated the prediction results of multiple different prediction models using ensemble learning or model fusion methods. This step helps to improve the accuracy and stability of the prediction.

[0072] In step six, the forecast report generation stage, we not only gave the final sugarcane yield forecast value, but also analyzed the various factors affecting sugarcane growth, and based on these analyses, put forward specific agricultural management suggestions and optimization measures for reference by agricultural producers.

[0073] In step seven, the result verification and evaluation phase, we compared and verified the prediction results with the actual production data and used a variety of evaluation indicators to verify and evaluate the prediction results. This step helps to evaluate the accuracy and reliability of the prediction model and provide a basis for subsequent optimization and improvement.

[0074] Finally, in step eight, during the continuous optimization and improvement phase, we established a feedback mechanism to collect user feedback on the prediction results and continuously optimized and improved the model based on the feedback. At the same time, we also tracked the development of new technologies and new methods and introduced new technologies into the prediction model in a timely manner to improve performance. This step helps to keep the model continuously updated and optimized to adapt to changing needs and environments.

[0075] In another optional embodiment, the data collection and preprocessing steps specifically include: acquisition of drone hyperspectral image data: during the sugarcane growth cycle, hyperspectral image data of the sugarcane field is obtained by using a drone equipped with a hyperspectral camera to capture spectral information during the sugarcane growth process, and the drone hyperspectral image data includes but is not limited to canopy coverage, canopy height, normalized vegetation index, and overgreening index;

[0076] Acquisition of satellite remote sensing data: Using satellite remote sensing technology, remote sensing image data of sugarcane fields is regularly collected to obtain sugarcane growth information over a wider range;

[0077] Collection of ground observation data: by collecting actual measurements of key sugarcane phenotypes in sugarcane fields, including but not limited to sugarcane growth parameters, yield-related indicators, and soil and environmental parameters;

[0078] Meteorological data collection: Collect meteorological data during the sugarcane growth cycle, including but not limited to temperature, precipitation and light, to analyze the impact of meteorological factors on sugarcane yield;

[0079] Data classification storage: Classify and store various types of collected data for subsequent processing;

[0080] Data preprocessing: Clean the collected data, remove outliers, fill in missing values ​​and other preprocessing operations to improve data quality;

[0081] Multi-source data fusion: The pre-processed multi-source data are fused to form a multidimensional data set containing a variety of information, providing data support for subsequent feature extraction and model building. At the same time, advanced spatiotemporal interpolation methods are used to deeply fuse spatiotemporal data to capture the complex relationship between spatial distribution and temporal evolution during sugarcane growth.

[0082] In the data preprocessing stage, the following machine learning algorithms are selected to process the data collected by different collection methods:

[0083] For UAV hyperspectral image data, a convolutional neural network combined with a deep learning framework is used for feature extraction to fully utilize the spectral and spatial information in the hyperspectral image and improve the accuracy of identifying the growth status of sugarcane.

[0084] For satellite remote sensing data, long short-term memory networks or recurrent neural networks combined with attention mechanisms are used to capture long-term dependencies and changes at important time points in time series data to accurately predict sugarcane growth trends;

[0085] For continuous variables in ground observation data, including but not limited to soil moisture and nutrient content, support vector machine regression or random forest regression is used to handle the nonlinearity and interaction of these data and improve the accuracy of yield prediction;

[0086] For categorical variables in meteorological data, including but not limited to rainfall type and wind direction, decision trees or gradient boosting trees are used for classification processing to accurately determine the impact of meteorological conditions on sugarcane yield;

[0087] For time series meteorological data, including but not limited to daily rainfall and daily average temperature, the time series analysis algorithm ARIMA or the seasonal decomposition time series prediction model SARI MA is combined with a deep learning model to capture the seasonal and cyclical changes of meteorological data and improve the accuracy of the prediction.

[0088] In this embodiment, we further refined each link of data collection and preprocessing. In the acquisition of UAV hyperspectral image data, we precisely controlled the flight altitude and angle of the UAV and the parameter settings of the hyperspectral camera to ensure that the acquired hyperspectral image data can fully and accurately reflect the growth status of sugarcane. At the same time, we also carried out fine radiation calibration and atmospheric correction processing on the collected hyperspectral image data to eliminate the influence of atmospheric effects on the image data.

[0089] In the acquisition of satellite remote sensing data, we selected satellite remote sensing data with high spatial resolution and temporal resolution to ensure that we can obtain detailed spatial information and time series information of the sugarcane fields. In addition, we also performed geometric correction and radiometric calibration on the satellite remote sensing data to improve the accuracy and availability of the data.

[0090] In the collection of ground observation data, we collected key environmental parameters such as soil moisture and nutrient content, as well as sugarcane growth parameters and yield-related indicators in real time through a sensor network deployed in the sugarcane field. These data provide us with more detailed field-scale information and help reveal the microscopic mechanism of sugarcane growth.

[0091] In the process of collecting meteorological data, we not only collected data on conventional meteorological elements such as temperature, precipitation, and light, but also collected meteorological data on wind speed, humidity, and other dimensions to comprehensively analyze the impact of meteorological factors on sugarcane yield. At the same time, we also performed quality control on the meteorological data to remove abnormal and missing data.

[0092] In the data preprocessing stage, we selected appropriate machine learning algorithms for processing different types of collected data. For drone hyperspectral image data, we used convolutional neural networks (CNN) combined with deep learning frameworks for feature extraction, effectively capturing the spectral and spatial information in hyperspectral images. For satellite remote sensing data, we used long short-term memory networks (LSTM) or recurrent neural networks (RNN) combined with attention mechanisms to successfully capture long-term dependencies and changes in important time points in time series data. For continuous variables in ground observation data, we used support vector machine regression (SVR) or random forest regression (RFR) to process, effectively handling the nonlinearity and interaction of data. For categorical variables in meteorological data, we used decision trees (DT) or gradient boosting trees (GBT) for classification processing, accurately determining the impact of meteorological conditions on sugarcane yield. For time series meteorological data, we used time series analysis algorithms ARIMA or seasonal decomposition time series prediction models SARIMA combined with deep learning models to successfully capture seasonal and cyclical changes in meteorological data.

[0093] Through the above meticulous data collection and preprocessing work, we provide high-quality data support for subsequent feature extraction and model building. These preprocessed multi-source data will be used to build a more accurate and reliable sugarcane yield prediction model.

[0094] In another optional embodiment, in the feature extraction stage, a combined model (ConvLSTM) of a convolutional neural network (CNN) and a recurrent neural network (RNN) is used to extract spatiotemporal features according to the characteristics of spatiotemporal data, so as to effectively extract spatiotemporal features during sugarcane growth. In the feature extraction stage, an unsupervised learning method is used to further reduce the dimension of the features to reduce the computational complexity and retain key information, while exploring the potential structures and patterns in the data.

[0095] In this embodiment, in the feature extraction stage, we fully utilized the advantages of CNN and RNN, and deeply explored the spatiotemporal characteristics of sugarcane growth through the ConvLSTM model. This model combines the spatial feature extraction capability of CNN and the time series processing capability of RNN, enabling us to more comprehensively capture the dynamic changes of sugarcane at different growth stages.

[0096] In the process of feature extraction, we also introduced unsupervised learning methods, such as principal component analysis (PCA) or autoencoder (AE), to further reduce the dimensionality of the extracted features. These methods can reduce the dimension of the data and the computational complexity while retaining key information, providing more efficient data input for subsequent model construction.

[0097] In addition, unsupervised learning methods can also help us discover potential structures and patterns in data, which is of great significance for understanding the internal mechanism of sugarcane growth. Through these methods, we can reveal the hidden laws in the growth process of sugarcane and provide a more solid theoretical basis for the construction of prediction models. By adopting the ConvLSTM model and unsupervised learning methods, we can not only effectively extract the spatiotemporal characteristics of sugarcane growth, but also reduce the computational complexity of the model and improve the efficiency and accuracy of prediction. In addition, this feature extraction method helps to discover potential patterns in data and provide a deeper understanding and insight into sugarcane planting. Through these advanced data processing and analysis technologies, this embodiment provides strong technical support for the intelligent management and development of the sugarcane planting industry.

[0098] In another optional embodiment, in the model selection and construction step, for different feature subsets, the following machine learning algorithms are selected for one-to-one correspondence:

[0099] For spectral features, gradient boosting regression tree was selected to build the prediction model;

[0100] For spatial features, support vector machine regression was selected to build a prediction model;

[0101] For environmental characteristics, multiple linear regression was selected to build a prediction model.

[0102] In this embodiment, in the model selection and construction steps, we specifically selected the corresponding machine learning algorithm for model construction based on the different characteristics of the feature subsets.

[0103] For the spectral features extracted from UAV hyperspectral image data and satellite remote sensing data, since they usually contain a large number of continuous variables and there may be complex nonlinear relationships between these variables, we chose the Gradient Boosting Regression Tree to build a prediction model. The Gradient Boosting Regression Tree can effectively handle nonlinear problems, has a good fitting ability for complex patterns in spectral features, and is also very robust to interactions between features.

[0104] For the spatial features extracted from UAV hyperspectral image data and satellite remote sensing data, since they often involve spatial location information and spatial correlation, we chose Support Vector Regression to build a prediction model. Support Vector Regression can not only handle linear problems, but also map data to high-dimensional space through kernel functions to handle nonlinear problems, and has a natural advantage in processing spatial data.

[0105] For the environmental characteristics in meteorological data and ground observation data, since they usually contain multiple independent variables, and there may be a linear relationship between these independent variables and the dependent variable, we chose Multivariate Linear Regression to build a prediction model. Multivariate Linear Regression can effectively handle the relationship between multiple independent variables and dependent variables, and the model is simple and easy to explain.

[0106] For time series meteorological data, such as temperature and precipitation, time series analysis algorithms, such as the autoregressive integrated moving average model (ARIMA) or the seasonal decomposition time series forecasting model (SARIMA), can be selected to build forecasting models. These algorithms can capture seasonal and trend changes in time series data, which helps to improve the model's forecasting accuracy for meteorological factors.

[0107] Through the above targeted model selection and construction, we can give full play to the advantages of various machine learning algorithms and improve the accuracy and stability of sugarcane yield prediction.

[0108] like Figure 2 As shown: In another optional embodiment, in the model training step, the model is trained using historical data, and the model parameters are tuned by cross-validation, which specifically includes the following steps:

[0109] Step 1: Divide the historical data into training set, validation set and test set in a ratio of 70%:15%:15%;

[0110] Step 2: Use the training set to train the model;

[0111] Step 3: Use the validation set to validate the trained model and adjust the model parameters based on the validation results;

[0112] Step 4: Use a test set independent of the training set to validate the model to evaluate the performance of the model in practical applications;

[0113] Step 5: Repeat the above process until the model performance reaches the optimal level.

[0114] In this embodiment, in the model training step, we adopted a rigorous data partitioning and model tuning process to ensure that the model can perform optimally in practical applications.

[0115] First, we divide the historical data into training set, validation set and test set in a ratio of 70%:15%:15%. This division ratio ensures that the model has enough data for training and retains enough independent data to verify the generalization ability of the model.

[0116] Next, we use the training set to perform preliminary training on the model. In this stage, the model will try to learn the patterns in the data and gradually approach the true function.

[0117] Then, we use the validation set to validate the trained model. The validation set provides a data environment different from the training set for the model to be tested. By observing the performance of the model on the validation set, we can evaluate the generalization ability of the model and adjust the parameters of the model as needed.

[0118] After that, we use a test set that is completely independent of the training set to perform a final validation of the model. The purpose of this step is to ensure that the model not only performs well on the training and validation sets, but also maintains stable performance on completely unseen data.

[0119] Finally, we repeat the above process and continuously adjust the model parameters until the model achieves optimal performance on the training set, validation set, and test set. This process may involve multiple iterations, but it is through this repeated optimization that we can ensure that the model has the best predictive ability in practical applications.

[0120] In another optional embodiment, the prediction method further includes, for the prediction of sugarcane yield in a specific area, adopting a transfer learning strategy, using a model trained in other similar areas as a pre-trained model, and fine-tuning with a small amount of local data to quickly adapt to the prediction needs of the new area, thereby improving the prediction efficiency and accuracy. In the model prediction, integration and fusion steps, the integrated learning method includes but is not limited to stacking, lifting and bagging; in the result verification and evaluation step, the evaluation indicators include mean square error, root mean square error, determination coefficient and absolute percentage error.

[0121] In this embodiment, transfer learning is a machine learning method that uses knowledge learned from one field (source field) to help learning tasks in another field (target field). This method not only improves prediction efficiency, but also improves prediction accuracy. In order to cope with the prediction bias that may be caused by differences in data distribution between different regions, we added an adaptive learning mechanism on the basis of transfer learning, which can automatically detect and adjust model parameters to better adapt to the data characteristics and laws unique to the new region. By combining the fast startup capability of transfer learning and the dynamic adjustment capability of adaptive learning, our prediction method can achieve higher prediction accuracy in a shorter time while maintaining good adaptability to the new environment.

[0122] In the model prediction, integration and fusion steps, the stability and reliability of the prediction results are further improved by combining the advantages of multiple prediction models. We particularly emphasize the diversity and complementarity of the models. In addition to using traditional ensemble learning methods such as stacking, boosting and bagging, we also explore hybrid model strategies, that is, organically combining different types of prediction models (such as statistical-based models, machine learning models and deep learning models). This hybrid model can make full use of the advantages of each model and reduce the possible deviations and limitations of a single model, thereby further improving the accuracy and stability of the prediction.

[0123] In the result verification and evaluation steps, we selected a variety of evaluation indicators to comprehensively measure the performance of the prediction model. These indicators include mean square error (MSE), root mean square error (RMSE), determination coefficient (R 2 ) and absolute percentage error (MAPE). Through these evaluation indicators, we can more accurately understand the prediction accuracy and reliability of the model, thereby providing more accurate guidance and suggestions for agricultural production.

[0124] like Figure 3 As shown: The present invention also provides a sugarcane yield prediction system based on machine learning, which is used to implement the above method. The system includes:

[0125] Data collection module: responsible for automatically collecting relevant data during the sugarcane growth cycle;

[0126] Data preprocessing module: receives data from the data acquisition module, performs preprocessing operations, and fuses multi-source data to form a data set containing multi-dimensional information on crop genetic information, growth status, and environmental parameters, providing a high-quality data foundation for subsequent analysis;

[0127] Feature extraction and model selection module: used to automatically extract key features from preprocessed data using advanced machine learning algorithms, and select appropriate machine learning algorithms or algorithm combinations to build prediction models based on feature extraction results;

[0128] Model training and optimization module: Use historical data to train the model and tune the model parameters through cross-validation and grid search strategies;

[0129] Prediction and decision support module: Use the trained model to predict sugarcane yield, generate detailed prediction reports based on the prediction results and influencing factors, and propose corresponding agricultural management suggestions and optimization measures. It also supports integrated learning and model fusion strategies to improve the accuracy and stability of prediction results.

[0130] User interaction and feedback module: provides a friendly user interface that allows users to input query conditions, view prediction results, and receive agricultural management suggestions. At the same time, this module is also responsible for collecting user feedback on the prediction results and continuously improving and optimizing the prediction model by establishing a feedback mechanism.

[0131] In the present invention, the sugarcane yield prediction system based on machine learning is a comprehensive platform, which integrates multiple links such as data collection, preprocessing, feature extraction, model construction, training, prediction and decision support, and realizes the automation and intelligence of sugarcane yield prediction.

[0132] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0133] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A sugarcane yield prediction method based on machine learning, It is characterized in that The following steps are involved: Step 1: Data collection and preprocessing: Collect crop gene data, drone hyperspectral image data, satellite remote sensing data, ground observation data, meteorological data, and historical yield data during the sugarcane growth cycle, classify and store the collected data, and preprocess the relevant data one by one, merge the multi-source data, and form a multidimensional data set; Step 2: Feature extraction: Extract key features closely related to sugarcane yield and quality from the preprocessed data, use feature selection algorithm to further select feature subsets that contribute most to the prediction results and perform dimensionality reduction processing; Step 3: Model selection and construction: According to different prediction targets, different feature subsets are selected and appropriate machine learning algorithms or algorithm combinations are used to construct prediction models. The machine learning algorithms include but are not limited to: gradient boosting regression tree, partial least squares regression, support vector machine regression, BP neural network, long short-term memory network, deep extreme learning machine, multivariate linear regression, stepwise multivariate regression, generalized linear model, generalized boosting model and random forest algorithm. Step 4: Model training and tuning: Use historical data to train the model and tune the model parameters through cross-validation; Step 5: Model prediction, integration and fusion: Use the trained prediction model to predict sugarcane yield, and use ensemble learning or model fusion methods to integrate the prediction results of multiple different prediction models to improve the accuracy and stability of the prediction; Step 6: Forecast report generation: According to the selected integrated strategy, the final sugarcane yield forecast value is calculated, and according to the forecast results and influencing factors analysis, corresponding agricultural management suggestions and optimization measures are proposed; Step 7: Result verification and evaluation: Verify and evaluate the prediction results by comparing with the actual production data, and use a variety of evaluation indicators to verify and evaluate the accuracy and reliability of the prediction model; Step 8. Continuous optimization and improvement: Establish a feedback mechanism to collect user feedback on the prediction results, and continuously optimize and improve the model based on the feedback. At the same time, track the development of new technologies and new methods, and introduce new technologies into the prediction model in a timely manner to improve performance.

2. The sugarcane yield prediction method based on machine learning according to claim 1, It is characterized in that The data collection and preprocessing steps specifically include: Crop genetic data i Collection of historical production data; Collection of UAV hyperspectral image data: During the sugarcane growth cycle, the hyperspectral image data of the sugarcane field is obtained by using a UAV equipped with a hyperspectral camera to capture the spectral information of the sugarcane growth process. The UAV hyperspectral image data includes but is not limited to canopy coverage, canopy height, normalized vegetation index, and overgreening index; Acquisition of satellite remote sensing data: Using satellite remote sensing technology, remote sensing image data of sugarcane fields is regularly collected to obtain sugarcane growth information over a wider range; Collection of ground observation data: by collecting actual measurements of key sugarcane phenotypes in sugarcane fields, including but not limited to sugarcane growth parameters, yield-related indicators, and soil and environmental parameters; Meteorological data collection: Collect meteorological data during the sugarcane growth cycle, including but not limited to the daily maximum and minimum temperatures, precipitation and sunshine, to analyze the impact of meteorological factors on sugarcane yield; Data classification storage: Classify and store various types of collected data for subsequent processing; Data preprocessing: Clean the collected data, remove outliers, fill in missing values ​​and other preprocessing operations to improve data quality; Multi-source data fusion: The pre-processed multi-source data are fused to form a multidimensional data set containing a variety of information, providing data support for subsequent feature extraction and model building. At the same time, advanced spatiotemporal interpolation methods are used to deeply fuse spatiotemporal data to capture the complex relationship between spatial distribution and temporal evolution during sugarcane growth.

3. The sugarcane yield prediction method based on machine learning according to claim 1, It is characterized in that In the data preprocessing stage, the following machine learning algorithms are selected to process the data collected by different collection methods: For UAV hyperspectral image data, a convolutional neural network combined with a deep learning framework is used for feature extraction to fully utilize the spectral and spatial information in the hyperspectral image and improve the accuracy of identifying the growth status of sugarcane. For satellite remote sensing data, long short-term memory networks or recurrent neural networks combined with attention mechanisms are used to capture long-term dependencies and changes at important time points in time series data to accurately predict sugarcane growth trends; For continuous variables in ground observation data, including but not limited to soil moisture and nutrient content, support vector machine regression or random forest regression is used to handle the nonlinearity and interaction of these data and improve the accuracy of yield prediction; For categorical variables in meteorological data, including but not limited to rainfall type and wind direction, decision trees or gradient boosting trees are used for classification processing to accurately determine the impact of meteorological conditions on sugarcane yield; For time series meteorological data, including but not limited to daily rainfall and daily average temperature, the time series analysis algorithm ARIMA or the seasonal decomposition time series prediction model SARI MA is combined with a deep learning model to capture the seasonal and cyclical changes of meteorological data and improve the accuracy of the prediction.

4. The sugarcane yield prediction method based on machine learning according to claim 1, It is characterized in that In the feature extraction stage, according to the characteristics of spatiotemporal data, a combined model (ConvLSTM) of convolutional neural network (CNN) and recurrent neural network (RNN) was used to extract spatiotemporal features, so as to effectively extract the spatiotemporal characteristics of sugarcane growth process.

5. The sugarcane yield prediction method based on machine learning according to claim 1, It is characterized in that In the feature extraction stage, unsupervised learning methods are used to further reduce the dimension of features to reduce computational complexity and retain key information while exploring the potential structures and patterns in the data.

6. The sugarcane yield prediction method based on machine learning according to claim 1, It is characterized in that In the model selection and construction step, the following machine learning algorithms are selected for one-to-one correspondence with different feature subsets: For spectral features, gradient boosting regression tree was selected to build the prediction model; For spatial features, support vector machine regression was selected to build a prediction model; For environmental characteristics, multiple linear regression was selected to build a prediction model.

7. The sugarcane yield prediction method based on machine learning according to claim 1, It is characterized in that In the model training step, the model is trained using historical data, and the model parameters are tuned through cross-validation, which specifically includes the following steps: Step 1: Divide the historical data into training set, validation set and test set in a ratio of 70%:15%:15%; Step 2: Use the training set to train the model; Step 3: Use the validation set to validate the trained model and adjust the model parameters based on the validation results; Step 4: Use a test set independent of the training set to validate the model to evaluate the performance of the model in practical applications; Step 5: Repeat the above process until the model performance reaches the optimal level.

8. The sugarcane yield prediction method based on machine learning according to claim 1, It is characterized in that It also includes predicting sugarcane yield in a specific area by adopting a transfer learning strategy, using models trained in other similar areas as pre-trained models, and fine-tuning them with a small amount of local data to quickly adapt to the prediction needs of the new area, thereby improving prediction efficiency and accuracy.

9. The sugarcane yield prediction method based on machine learning according to claim 1, It is characterized in that In the model prediction, integration and fusion steps, the integrated learning method includes but is not limited to stacking, boosting and bagging; In the result verification and evaluation step, the evaluation indicators include mean square error, root mean square error, determination coefficient and absolute percentage error.

10. A sugarcane yield prediction system based on machine learning, used to implement the method according to any one of claims 1 to 9, It is characterized in that The system includes: Data collection module: responsible for automatically collecting relevant data during the sugarcane growth cycle; Data preprocessing module: receives data from the data acquisition module, performs preprocessing operations, and fuses multi-source data to form a data set containing multi-dimensional information on crop genetic information, growth status, and environmental parameters, providing a high-quality data foundation for subsequent analysis; Feature extraction and model selection module: used to automatically extract key features from preprocessed data using advanced machine learning algorithms, and select appropriate machine learning algorithms or algorithm combinations to build prediction models based on feature extraction results; Model training and optimization module: Use historical data to train the model and tune the model parameters through cross-validation and grid search strategies; Prediction and decision support module: Use the trained model to predict sugarcane yield, generate detailed prediction reports based on the prediction results and influencing factors, and propose corresponding agricultural management suggestions and optimization measures. It also supports integrated learning and model fusion strategies to improve the accuracy and stability of prediction results. User interaction and feedback module: provides a friendly user interface that allows users to input query conditions, view prediction results, and receive agricultural management suggestions. At the same time, this module is also responsible for collecting user feedback on the prediction results and continuously improving and optimizing the prediction model by establishing a feedback mechanism.

Images