Soil entropy condition prediction method and system combined with multiple models

Abstract

The invention provides a soil entropy prediction method and system combined with multiple models in the technical field of smart agriculture, and the method comprises the steps: S1, building a soil entropy prediction model based on a multi-modal feature extraction layer, a heterogeneous model fusion layer and an integrated prediction output layer, and setting a loss function of the soil entropy prediction model; s2, acquiring a large amount of historical monitoring data, preprocessing and labeling the historical monitoring data, and then constructing a data set; s3, dividing the data set into a training set, a verification set and a test set based on a K-fold cross validation method, and training, verifying and testing the soil entropy condition prediction model through the training set, the verification set and the test set; and S4, deploying the soil entropy condition prediction model passing the test, and performing soil entropy condition prediction through the deployed soil entropy condition prediction model. The method has the advantages that the generalization ability and accuracy of soil entropy condition prediction are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of smart agriculture technology, and in particular to a soil entropy prediction method and system combining multiple models. Background Art

[0002] Soil moisture refers to the content, distribution and changes of water in the soil. It is a key environmental indicator reflecting soil drought conditions and directly affects crop growth. If soil moisture can be predicted, irrigation can be carried out in the corresponding time period, which not only saves water resources but also ensures the healthy growth of crops.

[0003] The prediction of soil moisture conditions has traditionally been dominated by monitoring and statistical models. For example, in 2007, Wang Yiming and others systematically sorted out international progress and proposed a "sky-ground" collaborative monitoring framework, which achieved high-precision data collection through thermal inertia mode satellite remote sensing and ground sensors (TDR, SWR), laying the technical foundation; after 2010, the intelligent transformation accelerated, and Xu Liping and others constructed a drought prediction model based on BP neural network, combining meteorological data (precipitation, evaporation, temperature) with measured soil moisture values, and achieved a breakthrough of average absolute error of -0.9; in 2019, Liu Xiaolu and others realized the innovation of multi-source heterogeneous data fusion, integrating MOD IS remote sensing data (vegetation index, surface temperature), TRMM precipitation data and meteorological station information were used to create a comprehensive drought index (CDI). For the first time, remote sensing spatial information was coupled with ground meteorological parameters, revealing that sunshine, precipitation and air humidity are key drought-causing factors, and the reliability of the model was verified by soil moisture. In the 2020s, deep learning promoted the upgrade of precise prediction. Yang Jingfeng and others developed the Elman neural network framework, integrating 18 factors of meteorological automatic stations and multi-layer soil parameters to achieve breakthroughs in long, medium and short-term prediction accuracy. The universality of the model was verified in the northern arid areas, marking the deep integration of spatiotemporal scale expansion and algorithm optimization.

[0004] However, traditional methods have the following technical defects in the field of soil moisture prediction: First, early monitoring systems rely on a single data source (such as satellite remote sensing or ground sensors), which has problems such as insufficient temporal and spatial resolution and weak multi-source heterogeneous data fusion capabilities, resulting in poor adaptability to complex surface conditions; Second, prediction methods based on statistical models (such as thermal inertia models) do not adequately analyze the dynamic coupling relationship between nonlinear meteorological factors and soil parameters, making it difficult to achieve cross-regional generalization applications; Third, traditional neural networks (such as BP networks) are limited by shallow structures, and the temporal and spatial correlation features between long-sequence meteorological factors and multi-layer soil parameters are insufficiently extracted, especially in extreme weather conditions, where the prediction accuracy is significantly reduced.

[0005] Therefore, how to provide a soil entropy prediction method and system that combines multiple models to improve the generalization ability and accuracy of soil entropy prediction has become a technical problem that needs to be solved urgently. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a soil entropy prediction method and system combining multiple models to improve the generalization ability and accuracy of soil entropy prediction.

[0007] In a first aspect, the present invention provides a soil entropy prediction method combining multiple models, comprising the following steps:

[0008] Step S1: creating a soil entropy prediction model based on the multimodal feature extraction layer, the heterogeneous model fusion layer, and the integrated prediction output layer, and setting the loss function of the soil entropy prediction model;

[0009] The multimodal feature extraction layer is used to extract features from the monitoring data to obtain splicing features; the heterogeneous model fusion layer is used to perform feature mining and fusion on the splicing features to obtain fusion features; the integrated prediction output layer is used to output the soil entropy prediction results based on the fusion features;

[0010] Step S2: Acquire a large amount of historical monitoring data, pre-process and annotate each of the historical monitoring data to construct a data set;

[0011] Step S3, dividing the data set into a training set, a validation set, and a test set based on the K-fold cross-validation method, and training, validating, and testing the soil entropy prediction model using the training set, the validation set, and the test set;

[0012] Step S4: deploying the soil entropy prediction model that has passed the test, and performing soil entropy prediction using the deployed soil entropy prediction model.

[0013] Furthermore, in step S1, the multimodal feature extraction layer is constructed based on a structured parameter processing module, an image processing module, and a feature splicing module; the structured parameter processing module is used to extract structured high-order features from the structured data of the monitoring data through a LASSO sub-model and a KNN sub-model; the image processing module is used to extract multi-scale image texture features from the remote sensing image of the monitoring data through a CNN sub-model; and the feature splicing module is used to splice the structured high-order features and the multi-scale image texture features to obtain spliced ​​features;

[0014] The heterogeneous model fusion layer is constructed based on a tree model channel, a kernel method channel, a linear channel, a decision channel and a dynamic feature fusion module; the tree model channel is used to extract nonlinear interaction sub-features from each splicing feature through a LightGBM sub-model, an XGBoost sub-model and an RF sub-model; the kernel method channel is used to extract high-dimensional space sub-features from each splicing feature through the kernel function of the LSVM sub-model; the linear channel is used to extract linear sub-features from each splicing feature through a linear regression sub-model; the decision channel is used to extract segmentation sub-features from each splicing feature through a decision tree sub-model; the dynamic feature fusion module is used to dynamically fuse nonlinear interaction sub-features, high-dimensional space sub-features, linear sub-features and segmentation sub-features through gated weighting to obtain fused features;

[0015] The integrated prediction output layer is used to dynamically adjust the weights of each sub-feature in the fusion feature through a Bayesian optimizer, and output the soil entropy prediction results through a deep residual network.

[0016] Furthermore, the step S2 is specifically as follows:

[0017] Acquire a large amount of historical monitoring data including structured data and remote sensing images, wherein the structured data at least includes maximum light intensity, average light intensity, sunshine duration, maximum temperature, minimum temperature, average temperature, daytime average temperature, nighttime average temperature, diurnal temperature difference, maximum air humidity, minimum air humidity, average relative humidity, daytime average air humidity, nighttime average air humidity, maximum carbon dioxide concentration, minimum carbon dioxide concentration, average carbon dioxide concentration, daytime average carbon dioxide concentration, nighttime average carbon dioxide concentration, maximum air pressure, minimum air pressure, average air pressure, daytime average air pressure, nighttime average air pressure, air pressure difference, daily average PM2.5, daily average PM10, maximum TBQ global radiation, minimum TBQ global radiation, daily average TBQ global radiation, maximum soil temperature, minimum soil temperature, daily average soil temperature, soil temperature difference, maximum soil pH, minimum soil pH, average soil pH, maximum soil salinity, minimum soil salinity, average soil salinity, maximum soil electrical conductivity, minimum soil electrical conductivity, average soil electrical conductivity, rainfall, wind direction, wind speed, and sampling location;

[0018] The structured data in each of the historical monitoring data are preprocessed by at least filling missing values ​​and repairing outliers, and the remote sensing images in the historical monitoring data are preprocessed by at least noise reduction, contrast enhancement, morphological adjustment, scaling, and color space conversion; the missing value filling is based on the K-nearest neighbor filling method; and the outlier repair is based on the K-means clustering algorithm;

[0019] The soil entropy conditions of the pre-processed historical monitoring data are annotated to construct a data set.

[0020] Furthermore, the step S3 is specifically as follows:

[0021] The data set is divided into a training set, a validation set, and a test set based on the K-fold cross-validation method, and the soil entropy prediction model is trained using the training set until the loss value of the loss function is less than a preset loss threshold;

[0022] The trained soil entropy prediction model is verified using the validation set to determine whether the prediction accuracy is greater than a preset accuracy threshold. If not, the verification fails and the training set is expanded to continue training. If so, the verification passes, and:

[0023] The verified soil entropy prediction model is tested using the test set to determine whether the confidence level is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training. If so, the test passes and training ends.

[0024] Furthermore, the step S4 is specifically as follows:

[0025] The soil entropy prediction model that has passed the test is deployed to the server through containerization technology, and an API interface for calling the soil entropy prediction model is set; real-time monitoring data including structured data and remote sensing images are obtained, and after preprocessing each of the real-time monitoring data, the soil entropy prediction model is input through the API interface to obtain the soil entropy prediction result to perform soil entropy prediction.

[0026] In a second aspect, the present invention provides a soil entropy prediction system combining multiple models, comprising the following modules:

[0027] A soil entropy prediction model creation module is used to create a soil entropy prediction model based on a multimodal feature extraction layer, a heterogeneous model fusion layer, and an integrated prediction output layer, and to set a loss function for the soil entropy prediction model;

[0028] The multimodal feature extraction layer is used to extract features from the monitoring data to obtain splicing features; the heterogeneous model fusion layer is used to perform feature mining and fusion on the splicing features to obtain fusion features; the integrated prediction output layer is used to output the soil entropy prediction results based on the fusion features;

[0029] A data set construction module is used to obtain a large amount of historical monitoring data, pre-process the historical monitoring data, and construct a data set after annotating the data;

[0030] A soil entropy prediction model training module is used to divide the data set into a training set, a validation set, and a test set based on the K-fold cross-validation method, and to train, validate, and test the soil entropy prediction model using the training set, validation set, and test set;

[0031] The soil entropy prediction module is used to deploy the soil entropy prediction model that has passed the test and perform soil entropy prediction using the deployed soil entropy prediction model.

[0032] Furthermore, in the soil entropy prediction model creation module, the multimodal feature extraction layer is constructed based on the structured parameter processing module, the image processing module and the feature splicing module; the structured parameter processing module is used to extract structured high-order features from the structured data of the monitoring data through the LASSO sub-model and the KNN sub-model; the image processing module is used to extract multi-scale image texture features from the remote sensing images of the monitoring data through the CNN sub-model; the feature splicing module is used to splice the structured high-order features and the multi-scale image texture features to obtain splicing features;

[0033] The heterogeneous model fusion layer is constructed based on a tree model channel, a kernel method channel, a linear channel, a decision channel and a dynamic feature fusion module; the tree model channel is used to extract nonlinear interaction sub-features from each splicing feature through a LightGBM sub-model, an XGBoost sub-model and an RF sub-model; the kernel method channel is used to extract high-dimensional space sub-features from each splicing feature through the kernel function of the LSVM sub-model; the linear channel is used to extract linear sub-features from each splicing feature through a linear regression sub-model; the decision channel is used to extract segmentation sub-features from each splicing feature through a decision tree sub-model; the dynamic feature fusion module is used to dynamically fuse nonlinear interaction sub-features, high-dimensional space sub-features, linear sub-features and segmentation sub-features through gated weighting to obtain fused features;

[0034] The integrated prediction output layer is used to dynamically adjust the weights of each sub-feature in the fusion feature through a Bayesian optimizer, and output the soil entropy prediction results through a deep residual network.

[0035] Furthermore, the dataset construction module is specifically used to:

[0036] Acquire a large amount of historical monitoring data including structured data and remote sensing images, wherein the structured data at least includes maximum light intensity, average light intensity, sunshine duration, maximum temperature, minimum temperature, average temperature, daytime average temperature, nighttime average temperature, diurnal temperature difference, maximum air humidity, minimum air humidity, average relative humidity, daytime average air humidity, nighttime average air humidity, maximum carbon dioxide concentration, minimum carbon dioxide concentration, average carbon dioxide concentration, daytime average carbon dioxide concentration, nighttime average carbon dioxide concentration, maximum air pressure, minimum air pressure, average air pressure, daytime average air pressure, nighttime average air pressure, air pressure difference, daily average PM2.5, daily average PM10, maximum TBQ global radiation, minimum TBQ global radiation, daily average TBQ global radiation, maximum soil temperature, minimum soil temperature, daily average soil temperature, soil temperature difference, maximum soil pH, minimum soil pH, average soil pH, maximum soil salinity, minimum soil salinity, average soil salinity, maximum soil electrical conductivity, minimum soil electrical conductivity, average soil electrical conductivity, rainfall, wind direction, wind speed, and sampling location;

[0037] The structured data in each of the historical monitoring data are preprocessed by at least filling missing values ​​and repairing outliers, and the remote sensing images in the historical monitoring data are preprocessed by at least noise reduction, contrast enhancement, morphological adjustment, scaling, and color space conversion; the missing value filling is based on the K-nearest neighbor filling method; and the outlier repair is based on the K-means clustering algorithm;

[0038] The soil entropy conditions of the pre-processed historical monitoring data are annotated to construct a data set.

[0039] Furthermore, the soil entropy prediction model training module is specifically used to:

[0040] The data set is divided into a training set, a validation set, and a test set based on the K-fold cross-validation method, and the soil entropy prediction model is trained using the training set until the loss value of the loss function is less than a preset loss threshold;

[0041] The trained soil entropy prediction model is verified using the validation set to determine whether the prediction accuracy is greater than a preset accuracy threshold. If not, the verification fails and the training set is expanded to continue training. If so, the verification passes, and:

[0042] The verified soil entropy prediction model is tested using the test set to determine whether the confidence level is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training. If so, the test passes and training ends.

[0043] Furthermore, the soil entropy prediction module is specifically used to:

[0044] The soil entropy prediction model that has passed the test is deployed to the server through containerization technology, and an API interface for calling the soil entropy prediction model is set; real-time monitoring data including structured data and remote sensing images are obtained, and after preprocessing each of the real-time monitoring data, the soil entropy prediction model is input through the API interface to obtain the soil entropy prediction result to perform soil entropy prediction.

[0045] The advantages of the present invention are:

[0046] 1. A soil entropy prediction model is created through the multimodal feature extraction layer, heterogeneous model fusion layer and integrated prediction output layer, and the loss function of the soil entropy prediction model is set; then a large amount of historical monitoring data is obtained to construct a data set, and the data set is divided into a training set, a validation set and a test set based on the K-fold cross-validation method. The soil entropy prediction model is trained, verified and tested through the training set, validation set and test set, and the soil entropy prediction model that passes the test is deployed, and soil entropy prediction is performed through the deployed soil entropy prediction model; that is, the soil entropy prediction is performed based on the soil entropy prediction model pre-trained on the data set constructed based on structured data and remote sensing images. Structured data contains multidimensional data, which effectively improves the generalization ability of the soil entropy prediction model. The soil entropy prediction model combines LASSO sub-model, KNN sub-model, CNN sub-model, LightGBM sub-model, XGBoost sub-model, RF sub-model, LSVM sub-model, linear regression sub-model and decision tree sub-model, combining the advantages of each model, greatly improving the feature extraction ability, and ultimately greatly improving the generalization ability and accuracy of soil entropy prediction.

[0047] 2. Numerical features are extracted through the structured parameter processing module (LASSO+KNN), and the multi-scale texture features of remote sensing images are extracted by combining the image processing module (CNN). The meteorological, soil physical and chemical parameters and spatial image information are comprehensively utilized to comprehensively capture the multi-dimensional factors affecting soil entropy. By integrating models based on different principles such as tree models (LightGBM / XGBoost / RF), kernel methods (LSVM), linear regression, and decision trees, nonlinear interactions, high-dimensional spatial mapping, linear relationships, and decision boundary features are captured respectively, breaking through the limitations of a single model and greatly improving the accuracy of soil entropy prediction.

[0048] 3. The sub-features output by the heterogeneous models are gated and weighted through the dynamic feature fusion module, and the contribution of each channel is adaptively adjusted to solve the problem of feature redundancy or conflict. The Bayesian optimizer is used to dynamically adjust the sub-feature weights through the integrated prediction layer to ensure that the fusion process is scientific and efficient, and effectively improve the generalization ability of the soil entropy prediction model for different soil types or environmental conditions.

[0049] 4. By using K-nearest neighbor to fill missing values ​​and K-means to repair outliers in structured data, noise interference can be avoided; by performing noise reduction and contrast enhancement on remote sensing images, effective information can be enhanced; by covering multidimensional parameters such as light, temperature and humidity, air pressure, and soil physical and chemical indicators in structured data, combined with the geographic spatial information of remote sensing images, a multidimensional feature system is constructed to comprehensively reflect the soil status, thereby greatly improving the accuracy of soil entropy prediction.

[0050] 5. By dividing the data set by K-fold, we can maximize the use of limited data, reduce the deviation caused by random division, and make the verification results more statistically significant. After training, we conduct double verification on the validation set (accuracy threshold) and the test set (confidence threshold) to ensure the reliability and robustness of the model and avoid overfitting or underfitting.

[0051] 6. By using containerization to encapsulate the soil entropy prediction model and environmental dependencies, rapid transplantation and elastic expansion can be achieved to adapt to different server configuration requirements.

[0052] 7. By providing a unified API interface, it is easy to integrate with other agricultural Internet of Things systems (such as weather stations and irrigation equipment), supporting real-time data stream processing and instant feedback of prediction results.

[0053] 8. The soil entropy prediction accuracy is significantly improved through the dynamic integration of multimodal features and heterogeneous models: On the one hand, the structured parameter processing module (LASSO+KNN) and the image processing module (CNN) are used to collaboratively extract the multi-scale features of meteorological, soil physical and chemical parameters and remote sensing images, and the gated weighting mechanism and the Bayesian optimizer are combined to dynamically integrate the complementary advantages of heterogeneous models such as tree models, kernel methods, and linear regression to effectively capture complex nonlinear relationships; On the other hand, the robustness of the model is guaranteed by the K-fold cross-validation and double threshold verification mechanisms, supplemented by data preprocessing technologies such as K-nearest neighbor filling and K-means anomaly repair to improve the input quality, and the containerized deployment and standardized API interface are combined to achieve efficient project implementation. It has both modular expansion capabilities and closed-loop iteration mechanisms, and has significant application value in precision agriculture decision support and soil disaster warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0055] Figure 1 The present invention is a flow chart of a soil entropy prediction method combining multiple models.

[0056] Figure 2 It is a structural schematic diagram of a soil entropy prediction system combining multiple models of the present invention. DETAILED DESCRIPTION

[0057] The technical solution in the embodiments of the present application has the following overall idea: a soil entropy prediction model is pre-trained based on a data set constructed based on structured data and remote sensing images to predict soil entropy. The structured data contains multidimensional data, which effectively improves the generalization ability of the soil entropy prediction model. The soil entropy prediction model combines the LASSO sub-model, KNN sub-model, CNN sub-model, LightGBM sub-model, XGBoost sub-model, RF sub-model, LSVM sub-model, linear regression sub-model and decision tree sub-model, combining the advantages of each model, greatly improving the feature extraction ability, and thus improving the generalization ability and accuracy of soil entropy prediction.

[0058] Please refer to Figures 1 to 2 As shown, a preferred embodiment of the soil entropy prediction method combining multiple models of the present invention includes the following steps:

[0059] Step S1: creating a soil entropy prediction model based on the multimodal feature extraction layer, the heterogeneous model fusion layer, and the integrated prediction output layer, and setting the loss function of the soil entropy prediction model;

[0060] The multimodal feature extraction layer is used to extract features from the monitoring data to obtain splicing features; the heterogeneous model fusion layer is used to perform feature mining and fusion on the splicing features to obtain fusion features; the integrated prediction output layer is used to output the soil entropy prediction results based on the fusion features;

[0061] Step S2: Acquire a large amount of historical monitoring data, pre-process and annotate each of the historical monitoring data to construct a data set;

[0062] Step S3, dividing the data set into a training set, a validation set, and a test set based on the K-fold cross-validation method, and training, validating, and testing the soil entropy prediction model using the training set, the validation set, and the test set;

[0063] Step S4: deploying the soil entropy prediction model that has passed the test, and performing soil entropy prediction using the deployed soil entropy prediction model.

[0064] Through the dynamic integration of multimodal features and heterogeneous models, the soil entropy prediction accuracy has been significantly improved: on the one hand, the structured parameter processing module (LASSO+KNN) and the image processing module (CNN) are used to collaboratively extract the multi-scale features of meteorological, soil physical and chemical parameters and remote sensing images, and the gated weighting mechanism and the Bayesian optimizer are combined to dynamically integrate the complementary advantages of heterogeneous models such as tree models, kernel methods, and linear regression to effectively capture complex nonlinear relationships; on the other hand, the robustness of the model is ensured by the K-fold cross-validation and double threshold verification mechanisms, supplemented by data preprocessing techniques such as K-nearest neighbor filling and K-means anomaly repair to improve the input quality, and the combination of containerized deployment and standardized API interfaces to achieve efficient project implementation. It has both modular expansion capabilities and closed-loop iteration mechanisms, and has significant application value in precision agriculture decision support and soil disaster warning.

[0065] In step S1, the multimodal feature extraction layer is constructed based on a structured parameter processing module, an image processing module, and a feature splicing module; the structured parameter processing module is used to extract structured high-order features from the structured data of the monitoring data through a LASSO sub-model and a KNN sub-model; the image processing module is used to extract multi-scale image texture features from the remote sensing image of the monitoring data through a CNN sub-model; and the feature splicing module is used to splice the structured high-order features and the multi-scale image texture features to obtain spliced ​​features;

[0066] The heterogeneous model fusion layer is constructed based on a tree model channel, a kernel method channel, a linear channel, a decision channel and a dynamic feature fusion module; the tree model channel is used to extract nonlinear interaction sub-features from each splicing feature through a LightGBM sub-model, an XGBoost sub-model and an RF sub-model; the kernel method channel is used to extract high-dimensional space sub-features from each splicing feature through the kernel function of the LSVM sub-model; the linear channel is used to extract linear sub-features from each splicing feature through a linear regression sub-model; the decision channel is used to extract segmentation sub-features from each splicing feature through a decision tree sub-model; the dynamic feature fusion module is used to dynamically fuse nonlinear interaction sub-features, high-dimensional space sub-features, linear sub-features and segmentation sub-features through gated weighting to obtain fused features;

[0067] The integrated prediction output layer is used to dynamically adjust the weights of each sub-feature in the fusion feature through a Bayesian optimizer, and output the soil entropy prediction results through a deep residual network.

[0068] Numerical features are extracted through the structured parameter processing module (LASSO+KNN), and the multi-scale texture features of remote sensing images are extracted by combining the image processing module (CNN). The meteorological, soil physical and chemical parameters and spatial image information are comprehensively utilized to comprehensively capture the multi-dimensional factors affecting soil entropy. By integrating models based on different principles such as tree models (LightGBM / XGBoost / RF), kernel methods (LSVM), linear regression, and decision trees, nonlinear interactions, high-dimensional spatial mapping, linear relationships, and decision boundary features are captured respectively, breaking through the limitations of a single model and greatly improving the accuracy of soil entropy prediction.

[0069] The dynamic feature fusion module is used to gate and weight the sub-features output by the heterogeneous models, adaptively adjust the contribution of each channel, and solve the problem of feature redundancy or conflict; the Bayesian optimizer is used to dynamically adjust the sub-feature weights through the integrated prediction layer to ensure that the fusion process is scientific and efficient, and effectively improve the generalization ability of the soil entropy prediction model for different soil types or environmental conditions.

[0070] The formula of the loss function is:

[0071]

[0072] Among them, L represents the loss value of the loss function; represents the mean square error loss, which is used to measure the average square difference between the predicted entropy and the true entropy; N represents the number of samples; y i Represents the true entropy of the i-th sample; Represents the predicted entropy of the i-th sample; represents the L1 regularization loss, which is used to prevent the model from overfitting; λ represents the regularization coefficient; M represents the parameter of the soil entropy prediction model; w j represents the jth model parameter; represents weight loss, which is used to ensure the reasonable distribution of weights; u represents the weight adjustment coefficient; K represents the number of sub-features; w k Represents the weight of the kth sub-feature; represents the optimal weight of the kth sub-feature.

[0073] The step S2 is specifically as follows:

[0074] Acquire a large amount of historical monitoring data including structured data and remote sensing images, wherein the structured data at least includes maximum light intensity, average light intensity, sunshine duration, maximum temperature, minimum temperature, average temperature, daytime average temperature, nighttime average temperature, diurnal temperature difference, maximum air humidity, minimum air humidity, average relative humidity, daytime average air humidity, nighttime average air humidity, maximum carbon dioxide concentration, minimum carbon dioxide concentration, average carbon dioxide concentration, daytime average carbon dioxide concentration, nighttime average carbon dioxide concentration, maximum air pressure, minimum air pressure, average air pressure, daytime average air pressure, nighttime average air pressure, air pressure difference, daily average PM2.5, daily average PM10, maximum TBQ global radiation, minimum TBQ global radiation, daily average TBQ global radiation, maximum soil temperature, minimum soil temperature, daily average soil temperature, soil temperature difference, maximum soil pH, minimum soil pH, average soil pH, maximum soil salinity, minimum soil salinity, average soil salinity, maximum soil electrical conductivity, minimum soil electrical conductivity, average soil electrical conductivity, rainfall, wind direction, wind speed, and sampling location;

[0075] The structured data in each of the historical monitoring data are preprocessed by at least filling missing values ​​and repairing outliers, and the remote sensing images in the historical monitoring data are preprocessed by at least noise reduction, contrast enhancement, morphological adjustment, scaling, and color space conversion; the missing value filling is based on the K-nearest neighbor filling method; and the outlier repair is based on the K-means clustering algorithm;

[0076] The soil entropy conditions of the pre-processed historical monitoring data are annotated to construct a data set.

[0077] By using K-nearest neighbor to fill missing values ​​and K-means to repair outliers in structured data, noise interference can be avoided; by performing noise reduction and contrast enhancement on remote sensing images, effective information can be enhanced; by covering multidimensional parameters such as light, temperature and humidity, air pressure, and soil physical and chemical indicators in structured data, combined with the geographic spatial information of remote sensing images, a multidimensional feature system is constructed to comprehensively reflect the soil status, thereby greatly improving the accuracy of soil entropy prediction.

[0078] The step S3 is specifically as follows:

[0079] The data set is divided into a training set, a validation set, and a test set based on the K-fold cross-validation method, and the soil entropy prediction model is trained using the training set until the loss value of the loss function is less than a preset loss threshold;

[0080] The trained soil entropy prediction model is verified using the validation set to determine whether the prediction accuracy is greater than a preset accuracy threshold. If not, the verification fails and the training set is expanded to continue training. If so, the verification passes, and:

[0081] The verified soil entropy prediction model is tested using the test set to determine whether the confidence level is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training. If so, the test passes and training ends.

[0082] By K-folding the data set, we can maximize the use of limited data, reduce the deviation caused by random division, and make the verification results more statistically significant. After training, we conduct double tests on the validation set (accuracy threshold) and the test set (confidence threshold) to ensure the reliability and robustness of the model and avoid overfitting or underfitting.

[0083] The step S4 is specifically as follows:

[0084] The soil entropy prediction model that has passed the test is deployed to the server through containerization technology, and an API interface for calling the soil entropy prediction model is set; real-time monitoring data including structured data and remote sensing images are obtained, and after preprocessing each of the real-time monitoring data, the soil entropy prediction model is input through the API interface to obtain the soil entropy prediction result to perform soil entropy prediction.

[0085] By using containerization to encapsulate the soil entropy prediction model and environmental dependencies, rapid transplantation and elastic expansion can be achieved to adapt to different server configuration requirements.

[0086] By providing a unified API interface, it facilitates integration with other agricultural IoT systems (such as weather stations and irrigation equipment), supporting real-time data stream processing and instant feedback of prediction results.

[0087] A preferred embodiment of the soil entropy prediction system combining multiple models of the present invention includes the following modules:

[0088] A soil entropy prediction model creation module is used to create a soil entropy prediction model based on a multimodal feature extraction layer, a heterogeneous model fusion layer, and an integrated prediction output layer, and to set a loss function for the soil entropy prediction model;

[0089] The multimodal feature extraction layer is used to extract features from the monitoring data to obtain splicing features; the heterogeneous model fusion layer is used to perform feature mining and fusion on the splicing features to obtain fusion features; the integrated prediction output layer is used to output the soil entropy prediction results based on the fusion features;

[0090] A data set construction module is used to obtain a large amount of historical monitoring data, pre-process the historical monitoring data, and construct a data set after annotating the data;

[0091] A soil entropy prediction model training module is used to divide the data set into a training set, a validation set, and a test set based on the K-fold cross-validation method, and to train, validate, and test the soil entropy prediction model using the training set, validation set, and test set;

[0092] The soil entropy prediction module is used to deploy the soil entropy prediction model that has passed the test and perform soil entropy prediction using the deployed soil entropy prediction model.

[0093] Through the dynamic integration of multimodal features and heterogeneous models, the soil entropy prediction accuracy has been significantly improved: on the one hand, the structured parameter processing module (LASSO+KNN) and the image processing module (CNN) are used to collaboratively extract the multi-scale features of meteorological, soil physical and chemical parameters and remote sensing images, and the gated weighting mechanism and the Bayesian optimizer are combined to dynamically integrate the complementary advantages of heterogeneous models such as tree models, kernel methods, and linear regression to effectively capture complex nonlinear relationships; on the other hand, the robustness of the model is ensured by the K-fold cross-validation and double threshold verification mechanisms, supplemented by data preprocessing techniques such as K-nearest neighbor filling and K-means anomaly repair to improve the input quality, and the combination of containerized deployment and standardized API interfaces to achieve efficient project implementation. It has both modular expansion capabilities and closed-loop iteration mechanisms, and has significant application value in precision agriculture decision support and soil disaster warning.

[0094] In the soil entropy prediction model creation module, the multimodal feature extraction layer is constructed based on a structured parameter processing module, an image processing module, and a feature splicing module; the structured parameter processing module is used to extract structured high-order features from the structured data of the monitoring data through a LASSO sub-model and a KNN sub-model; the image processing module is used to extract multi-scale image texture features from the remote sensing images of the monitoring data through a CNN sub-model; the feature splicing module is used to splice the structured high-order features and the multi-scale image texture features to obtain splicing features;

[0095] The heterogeneous model fusion layer is constructed based on a tree model channel, a kernel method channel, a linear channel, a decision channel and a dynamic feature fusion module; the tree model channel is used to extract nonlinear interaction sub-features from each splicing feature through a LightGBM sub-model, an XGBoost sub-model and an RF sub-model; the kernel method channel is used to extract high-dimensional space sub-features from each splicing feature through the kernel function of the LSVM sub-model; the linear channel is used to extract linear sub-features from each splicing feature through a linear regression sub-model; the decision channel is used to extract segmentation sub-features from each splicing feature through a decision tree sub-model; the dynamic feature fusion module is used to dynamically fuse nonlinear interaction sub-features, high-dimensional space sub-features, linear sub-features and segmentation sub-features through gated weighting to obtain fused features;

[0096] The integrated prediction output layer is used to dynamically adjust the weights of each sub-feature in the fusion feature through a Bayesian optimizer, and output the soil entropy prediction results through a deep residual network.

[0097] Numerical features are extracted through the structured parameter processing module (LASSO+KNN), and the multi-scale texture features of remote sensing images are extracted by combining the image processing module (CNN). The meteorological, soil physical and chemical parameters and spatial image information are comprehensively utilized to comprehensively capture the multi-dimensional factors affecting soil entropy. By integrating models based on different principles such as tree models (LightGBM / XGBoost / RF), kernel methods (LSVM), linear regression, and decision trees, nonlinear interactions, high-dimensional spatial mapping, linear relationships, and decision boundary features are captured respectively, breaking through the limitations of a single model and greatly improving the accuracy of soil entropy prediction.

[0098] The dynamic feature fusion module is used to gate and weight the sub-features output by the heterogeneous models, adaptively adjust the contribution of each channel, and solve the problem of feature redundancy or conflict; the Bayesian optimizer is used to dynamically adjust the sub-feature weights through the integrated prediction layer to ensure that the fusion process is scientific and efficient, and effectively improve the generalization ability of the soil entropy prediction model for different soil types or environmental conditions.

[0099] The formula of the loss function is:

[0100]

[0101] Among them, L represents the loss value of the loss function; represents the mean square error loss, which is used to measure the average square difference between the predicted entropy and the true entropy; N represents the number of samples; y i Represents the true entropy of the i-th sample; Represents the predicted entropy of the i-th sample; represents the L1 regularization loss, which is used to prevent the model from overfitting; λ represents the regularization coefficient; M represents the parameter of the soil entropy prediction model; w j represents the jth model parameter; represents weight loss, which is used to ensure the reasonable distribution of weights; u represents the weight adjustment coefficient; K represents the number of sub-features; w k Represents the weight of the kth sub-feature; represents the optimal weight of the kth sub-feature.

[0102] The dataset construction module is specifically used for:

[0103] Acquire a large amount of historical monitoring data including structured data and remote sensing images, wherein the structured data at least includes maximum light intensity, average light intensity, sunshine duration, maximum temperature, minimum temperature, average temperature, daytime average temperature, nighttime average temperature, diurnal temperature difference, maximum air humidity, minimum air humidity, average relative humidity, daytime average air humidity, nighttime average air humidity, maximum carbon dioxide concentration, minimum carbon dioxide concentration, average carbon dioxide concentration, daytime average carbon dioxide concentration, nighttime average carbon dioxide concentration, maximum air pressure, minimum air pressure, average air pressure, daytime average air pressure, nighttime average air pressure, air pressure difference, daily average PM2.5, daily average PM10, maximum TBQ global radiation, minimum TBQ global radiation, daily average TBQ global radiation, maximum soil temperature, minimum soil temperature, daily average soil temperature, soil temperature difference, maximum soil pH, minimum soil pH, average soil pH, maximum soil salinity, minimum soil salinity, average soil salinity, maximum soil electrical conductivity, minimum soil electrical conductivity, average soil electrical conductivity, rainfall, wind direction, wind speed, and sampling location;

[0104] The structured data in each of the historical monitoring data are preprocessed by at least filling missing values ​​and repairing outliers, and the remote sensing images in the historical monitoring data are preprocessed by at least noise reduction, contrast enhancement, morphological adjustment, scaling, and color space conversion; the missing value filling is based on the K-nearest neighbor filling method; and the outlier repair is based on the K-means clustering algorithm;

[0105] The soil entropy conditions of the pre-processed historical monitoring data are annotated to construct a data set.

[0106] By using K-nearest neighbor to fill missing values ​​and K-means to repair outliers in structured data, noise interference can be avoided; by performing noise reduction and contrast enhancement on remote sensing images, effective information can be enhanced; by covering multidimensional parameters such as light, temperature and humidity, air pressure, and soil physical and chemical indicators in structured data, combined with the geographic spatial information of remote sensing images, a multidimensional feature system is constructed to comprehensively reflect the soil status, thereby greatly improving the accuracy of soil entropy prediction.

[0107] The soil entropy prediction model training module is specifically used for:

[0108] The data set is divided into a training set, a validation set, and a test set based on the K-fold cross-validation method, and the soil entropy prediction model is trained using the training set until the loss value of the loss function is less than a preset loss threshold;

[0109] The trained soil entropy prediction model is verified using the validation set to determine whether the prediction accuracy is greater than a preset accuracy threshold. If not, the verification fails and the training set is expanded to continue training. If so, the verification passes, and:

[0110] The verified soil entropy prediction model is tested using the test set to determine whether the confidence level is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training. If so, the test passes and training ends.

[0111] By K-folding the data set, we can maximize the use of limited data, reduce the deviation caused by random division, and make the verification results more statistically significant. After training, we conduct double tests on the validation set (accuracy threshold) and the test set (confidence threshold) to ensure the reliability and robustness of the model and avoid overfitting or underfitting.

[0112] The soil entropy prediction module is specifically used for:

[0113] The soil entropy prediction model that has passed the test is deployed to the server through containerization technology, and an API interface for calling the soil entropy prediction model is set; real-time monitoring data including structured data and remote sensing images are obtained, and after preprocessing each of the real-time monitoring data, the soil entropy prediction model is input through the API interface to obtain the soil entropy prediction result to perform soil entropy prediction.

[0114] By using containerization to encapsulate the soil entropy prediction model and environmental dependencies, rapid transplantation and elastic expansion can be achieved to adapt to different server configuration requirements.

[0115] By providing a unified API interface, it facilitates integration with other agricultural IoT systems (such as weather stations and irrigation equipment), supporting real-time data stream processing and instant feedback of prediction results.

[0116] In summary, the advantages of the present invention are:

[0117] 1. A soil entropy prediction model is created through the multimodal feature extraction layer, heterogeneous model fusion layer and integrated prediction output layer, and the loss function of the soil entropy prediction model is set; then a large amount of historical monitoring data is obtained to construct a data set, and the data set is divided into a training set, a validation set and a test set based on the K-fold cross-validation method. The soil entropy prediction model is trained, verified and tested through the training set, validation set and test set, and the soil entropy prediction model that passes the test is deployed, and soil entropy prediction is performed through the deployed soil entropy prediction model; that is, the soil entropy prediction is performed based on the soil entropy prediction model pre-trained on the data set constructed based on structured data and remote sensing images. Structured data contains multidimensional data, which effectively improves the generalization ability of the soil entropy prediction model. The soil entropy prediction model combines LASSO sub-model, KNN sub-model, CNN sub-model, LightGBM sub-model, XGBoost sub-model, RF sub-model, LSVM sub-model, linear regression sub-model and decision tree sub-model, combining the advantages of each model, greatly improving the feature extraction ability, and ultimately greatly improving the generalization ability and accuracy of soil entropy prediction.

[0118] 2. Numerical features are extracted through the structured parameter processing module (LASSO+KNN), and the multi-scale texture features of remote sensing images are extracted by combining the image processing module (CNN). The meteorological, soil physical and chemical parameters and spatial image information are comprehensively utilized to comprehensively capture the multi-dimensional factors affecting soil entropy. By integrating models based on different principles such as tree models (LightGBM / XGBoost / RF), kernel methods (LSVM), linear regression, and decision trees, nonlinear interactions, high-dimensional spatial mapping, linear relationships, and decision boundary features are captured respectively, breaking through the limitations of a single model and greatly improving the accuracy of soil entropy prediction.

[0119] 3. The sub-features output by the heterogeneous models are gated and weighted through the dynamic feature fusion module, and the contribution of each channel is adaptively adjusted to solve the problem of feature redundancy or conflict. The Bayesian optimizer is used to dynamically adjust the sub-feature weights through the integrated prediction layer to ensure that the fusion process is scientific and efficient, and effectively improve the generalization ability of the soil entropy prediction model for different soil types or environmental conditions.

[0120] 4. By using K-nearest neighbor to fill missing values ​​and K-means to repair outliers in structured data, noise interference can be avoided; by performing noise reduction and contrast enhancement on remote sensing images, effective information can be enhanced; by covering multidimensional parameters such as light, temperature and humidity, air pressure, and soil physical and chemical indicators in structured data, combined with the geographic spatial information of remote sensing images, a multidimensional feature system is constructed to comprehensively reflect the soil status, thereby greatly improving the accuracy of soil entropy prediction.

[0121] 5. By dividing the data set by K-fold, we can maximize the use of limited data, reduce the deviation caused by random division, and make the verification results more statistically significant. After training, we conduct double verification on the validation set (accuracy threshold) and the test set (confidence threshold) to ensure the reliability and robustness of the model and avoid overfitting or underfitting.

[0122] 6. By using containerization to encapsulate the soil entropy prediction model and environmental dependencies, rapid transplantation and elastic expansion can be achieved to adapt to different server configuration requirements.

[0123] 7. By providing a unified API interface, it is easy to integrate with other agricultural Internet of Things systems (such as weather stations and irrigation equipment), supporting real-time data stream processing and instant feedback of prediction results.

[0124] 8. The soil entropy prediction accuracy is significantly improved through the dynamic integration of multimodal features and heterogeneous models: On the one hand, the structured parameter processing module (LASSO+KNN) and the image processing module (CNN) are used to collaboratively extract the multi-scale features of meteorological, soil physical and chemical parameters and remote sensing images, and the gated weighting mechanism and the Bayesian optimizer are combined to dynamically integrate the complementary advantages of heterogeneous models such as tree models, kernel methods, and linear regression to effectively capture complex nonlinear relationships; On the other hand, the robustness of the model is guaranteed by the K-fold cross-validation and double threshold verification mechanisms, supplemented by data preprocessing technologies such as K-nearest neighbor filling and K-means anomaly repair to improve the input quality, and the containerized deployment and standardized API interface are combined to achieve efficient project implementation. It has both modular expansion capabilities and closed-loop iteration mechanisms, and has significant application value in precision agriculture decision support and soil disaster warning.

[0125] Although the specific embodiments of the present invention are described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A soil entropy prediction method combining multiple models, characterized by: The steps include: Step S1: creating a soil entropy prediction model based on the multimodal feature extraction layer, the heterogeneous model fusion layer, and the integrated prediction output layer, and setting the loss function of the soil entropy prediction model; The multimodal feature extraction layer is used to extract features from the monitoring data to obtain splicing features; the heterogeneous model fusion layer is used to perform feature mining and fusion on the splicing features to obtain fusion features; the integrated prediction output layer is used to output the soil entropy prediction results based on the fusion features; Step S2: Acquire a large amount of historical monitoring data, pre-process and annotate each of the historical monitoring data to construct a data set; Step S3, dividing the data set into a training set, a validation set, and a test set based on the K-fold cross-validation method, and training, validating, and testing the soil entropy prediction model using the training set, the validation set, and the test set; Step S4: deploying the soil entropy prediction model that has passed the test, and performing soil entropy prediction using the deployed soil entropy prediction model.

2. The soil entropy prediction method combining multiple models according to claim 1, wherein: In step S1, the multimodal feature extraction layer is constructed based on a structured parameter processing module, an image processing module, and a feature splicing module; the structured parameter processing module is used to extract structured high-order features from the structured data of the monitoring data through a LASSO sub-model and a KNN sub-model; the image processing module is used to extract multi-scale image texture features from the remote sensing image of the monitoring data through a CNN sub-model; and the feature splicing module is used to splice the structured high-order features and the multi-scale image texture features to obtain spliced ​​features; The heterogeneous model fusion layer is constructed based on a tree model channel, a kernel method channel, a linear channel, a decision channel and a dynamic feature fusion module; the tree model channel is used to extract nonlinear interaction sub-features from each splicing feature through a LightGBM sub-model, an XGBoost sub-model and an RF sub-model; the kernel method channel is used to extract high-dimensional space sub-features from each splicing feature through the kernel function of the LSVM sub-model; the linear channel is used to extract linear sub-features from each splicing feature through a linear regression sub-model; the decision channel is used to extract segmentation sub-features from each splicing feature through a decision tree sub-model; the dynamic feature fusion module is used to dynamically fuse nonlinear interaction sub-features, high-dimensional space sub-features, linear sub-features and segmentation sub-features through gated weighting to obtain fused features; The integrated prediction output layer is used to dynamically adjust the weights of each sub-feature in the fusion feature through a Bayesian optimizer, and output the soil entropy prediction results through a deep residual network.

3. The soil entropy prediction method combining multiple models according to claim 1, wherein: The step S2 is specifically as follows: Acquire a large amount of historical monitoring data including structured data and remote sensing images, wherein the structured data at least includes maximum light intensity, average light intensity, sunshine duration, maximum temperature, minimum temperature, average temperature, daytime average temperature, nighttime average temperature, diurnal temperature difference, maximum air humidity, minimum air humidity, average relative humidity, daytime average air humidity, nighttime average air humidity, maximum carbon dioxide concentration, minimum carbon dioxide concentration, average carbon dioxide concentration, daytime average carbon dioxide concentration, nighttime average carbon dioxide concentration, maximum air pressure, minimum air pressure, average air pressure, daytime average air pressure, nighttime average air pressure, air pressure difference, daily average PM2.5, daily average PM10, maximum TBQ global radiation, minimum TBQ global radiation, daily average TBQ global radiation, maximum soil temperature, minimum soil temperature, daily average soil temperature, soil temperature difference, maximum soil pH, minimum soil pH, average soil pH, maximum soil salinity, minimum soil salinity, average soil salinity, maximum soil electrical conductivity, minimum soil electrical conductivity, average soil electrical conductivity, rainfall, wind direction, wind speed, and sampling location; The structured data in each of the historical monitoring data are preprocessed by at least filling missing values ​​and repairing outliers, and the remote sensing images in the historical monitoring data are preprocessed by at least noise reduction, contrast enhancement, morphological adjustment, scaling, and color space conversion; the missing value filling is based on the K-nearest neighbor filling method; and the outlier repair is based on the K-means clustering algorithm; The soil entropy conditions of the pre-processed historical monitoring data are annotated to construct a data set.

4. The soil entropy prediction method combining multiple models according to claim 1, wherein: The step S3 is specifically as follows: The data set is divided into a training set, a validation set, and a test set based on the K-fold cross-validation method, and the soil entropy prediction model is trained using the training set until the loss value of the loss function is less than a preset loss threshold; The trained soil entropy prediction model is verified using the validation set to determine whether the prediction accuracy is greater than a preset accuracy threshold. If not, the verification fails and the training set is expanded to continue training. If so, the verification passes, and: The verified soil entropy prediction model is tested using the test set to determine whether the confidence level is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training. If so, the test passes and training ends.

5. The soil entropy prediction method combining multiple models according to claim 1, wherein: The step S4 is specifically as follows: The soil entropy prediction model that has passed the test is deployed to the server through containerization technology, and an API interface for calling the soil entropy prediction model is set; real-time monitoring data including structured data and remote sensing images are obtained, and after preprocessing each of the real-time monitoring data, the soil entropy prediction model is input through the API interface to obtain the soil entropy prediction result to perform soil entropy prediction.

6. A soil entropy prediction system combining multiple models, characterized by: Includes the following modules: A soil entropy prediction model creation module is used to create a soil entropy prediction model based on a multimodal feature extraction layer, a heterogeneous model fusion layer, and an integrated prediction output layer, and to set a loss function for the soil entropy prediction model; The multimodal feature extraction layer is used to extract features from the monitoring data to obtain splicing features; the heterogeneous model fusion layer is used to perform feature mining and fusion on the splicing features to obtain fusion features; the integrated prediction output layer is used to output the soil entropy prediction results based on the fusion features; A data set construction module is used to obtain a large amount of historical monitoring data, pre-process the historical monitoring data, and construct a data set after annotating the data; A soil entropy prediction model training module is used to divide the data set into a training set, a validation set, and a test set based on the K-fold cross-validation method, and to train, validate, and test the soil entropy prediction model using the training set, validation set, and test set; The soil entropy prediction module is used to deploy the soil entropy prediction model that has passed the test and perform soil entropy prediction using the deployed soil entropy prediction model.

7. The soil entropy prediction system combining multiple models according to claim 6, characterized in that: In the soil entropy prediction model creation module, the multimodal feature extraction layer is constructed based on a structured parameter processing module, an image processing module, and a feature splicing module; the structured parameter processing module is used to extract structured high-order features from the structured data of the monitoring data through a LASSO sub-model and a KNN sub-model; the image processing module is used to extract multi-scale image texture features from the remote sensing images of the monitoring data through a CNN sub-model; the feature splicing module is used to splice the structured high-order features and the multi-scale image texture features to obtain splicing features; The heterogeneous model fusion layer is constructed based on a tree model channel, a kernel method channel, a linear channel, a decision channel and a dynamic feature fusion module; the tree model channel is used to extract nonlinear interaction sub-features from each splicing feature through a LightGBM sub-model, an XGBoost sub-model and an RF sub-model; the kernel method channel is used to extract high-dimensional space sub-features from each splicing feature through the kernel function of the LSVM sub-model; the linear channel is used to extract linear sub-features from each splicing feature through a linear regression sub-model; the decision channel is used to extract segmentation sub-features from each splicing feature through a decision tree sub-model; the dynamic feature fusion module is used to dynamically fuse nonlinear interaction sub-features, high-dimensional space sub-features, linear sub-features and segmentation sub-features through gated weighting to obtain fused features; The integrated prediction output layer is used to dynamically adjust the weights of each sub-feature in the fusion feature through a Bayesian optimizer, and output the soil entropy prediction results through a deep residual network.

8. The soil entropy prediction system combining multiple models according to claim 6, characterized in that: The dataset construction module is specifically used for: Acquire a large amount of historical monitoring data including structured data and remote sensing images, wherein the structured data at least includes maximum light intensity, average light intensity, sunshine duration, maximum temperature, minimum temperature, average temperature, daytime average temperature, nighttime average temperature, diurnal temperature difference, maximum air humidity, minimum air humidity, average relative humidity, daytime average air humidity, nighttime average air humidity, maximum carbon dioxide concentration, minimum carbon dioxide concentration, average carbon dioxide concentration, daytime average carbon dioxide concentration, nighttime average carbon dioxide concentration, maximum air pressure, minimum air pressure, average air pressure, daytime average air pressure, nighttime average air pressure, air pressure difference, daily average PM2.5, daily average PM10, maximum TBQ global radiation, minimum TBQ global radiation, daily average TBQ global radiation, maximum soil temperature, minimum soil temperature, daily average soil temperature, soil temperature difference, maximum soil pH, minimum soil pH, average soil pH, maximum soil salinity, minimum soil salinity, average soil salinity, maximum soil electrical conductivity, minimum soil electrical conductivity, average soil electrical conductivity, rainfall, wind direction, wind speed, and sampling location; The structured data in each of the historical monitoring data are preprocessed by at least filling missing values ​​and repairing outliers, and the remote sensing images in the historical monitoring data are preprocessed by at least noise reduction, contrast enhancement, morphological adjustment, scaling, and color space conversion; the missing value filling is based on the K-nearest neighbor filling method; and the outlier repair is based on the K-means clustering algorithm; The soil entropy conditions of the pre-processed historical monitoring data are annotated to construct a data set.

9. The soil entropy prediction system combining multiple models according to claim 6, characterized in that: The soil entropy prediction model training module is specifically used for: The data set is divided into a training set, a validation set, and a test set based on the K-fold cross-validation method, and the soil entropy prediction model is trained using the training set until the loss value of the loss function is less than a preset loss threshold; The trained soil entropy prediction model is verified using the validation set to determine whether the prediction accuracy is greater than a preset accuracy threshold. If not, the verification fails and the training set is expanded to continue training. If so, the verification passes, and: The verified soil entropy prediction model is tested using the test set to determine whether the confidence level is greater than a preset confidence threshold. If not, the test fails and the training set is expanded to continue training. If so, the test passes and training ends.

10. The soil entropy prediction system combining multiple models according to claim 6, characterized in that: The soil entropy prediction module is specifically used for: The soil entropy prediction model that has passed the test is deployed to the server through containerization technology, and an API interface for calling the soil entropy prediction model is set; real-time monitoring data including structured data and remote sensing images are obtained, and after preprocessing each of the real-time monitoring data, the soil entropy prediction model is input through the API interface to obtain the soil entropy prediction result to perform soil entropy prediction.

Images