Soil salinization type remote sensing inversion method based on ensemble learning

Abstract

The invention provides a soil salinization type remote sensing inversion method based on ensemble learning, and relates to the field of remote sensing technology and machine learning, and the method comprises the steps: collecting the field sampling data of each ion concentration in the soil of a target region; constructing a comprehensive data set through field sampling data and remote sensing image data; constructing an integrated learning regression model and training through the comprehensive data set; acquiring to-be-measured remote sensing image data; utilizing the trained integrated learning regression model to predict to-be-measured remote sensing image data, and generating soil ion concentration proportion data of the research area; and performing threshold division on the soil ion concentration proportion data to obtain different salinization types, and generating a two-dimensional space-time distribution diagram of the soil salinization types. The generalization ability and the inversion precision of the soil salinization inversion model are remarkably enhanced by adopting an integrated learning algorithm; through satellite-ground data fusion, the cost and time consumption of large-scale salinization monitoring are reduced.

Description

Technical Field

[0001] The present application relates to the fields of remote sensing technology and machine learning, and in particular to a remote sensing inversion method for soil salinization type based on ensemble learning. Background Art

[0002] Soil salinization, also known as soil salinization, refers to the phenomenon that salt in the soil rises to the surface due to capillary action and accumulates in the surface soil as water evaporates. This process not only seriously damages the physical and chemical properties of the soil, but also hinders plant growth, thus having a negative impact on the entire ecosystem and may also have an adverse impact on social economy and agricultural benefits.

[0004] Traditional soil salinization monitoring methods mainly rely on ground sampling and laboratory chemical analysis. Although these methods have certain guarantees in terms of accuracy, they still have major limitations: first, ground sampling requires a lot of manpower and time investment, especially in vast and remote areas, which makes monitoring time-consuming and costly; second, ground sampling can usually only obtain limited point data, which is difficult to fully reflect the spatial distribution characteristics of soil salinization in a large area. In addition, traditional soil salinization monitoring usually relies on simple linear regression algorithms or machine learning algorithms. Although these models can provide accurate predictions in some cases, their information mining and generalization capabilities are limited. A single model may perform well on a specific data set, but its generalization ability may be insufficient when facing new or different data sets. At the same time, the model complexity is limited, and a single model may encounter problems of computational complexity and overfitting when processing high-dimensional data or complex patterns. Summary of the invention

[0005] The purpose of the present invention is to provide a soil salinization type remote sensing inversion method based on ensemble learning in order to solve the technical problem that the traditional soil salinization inversion model using empirical methods or single machine learning methods cannot meet the needs of current practical applications.

[0006] The above-mentioned purpose of the present application is achieved through the following technical solutions: S1: Collect field sampling data of the concentration of various ions in the soil of the target area; the field sampling data includes: soil ion concentration data and geographical location data; S2: Acquire remote sensing image data of the target area; S3: Build a comprehensive dataset through field sampling data and remote sensing image data; S4: Through ensemble learning technology, an ensemble learning regression model is constructed and trained with a comprehensive dataset; S5: Obtain the remote sensing image data to be measured in the study area; S6: Use the trained ensemble learning regression model to predict the remote sensing image data to generate soil ion concentration ratio data of the study area; S7: Threshold division is performed on the soil ion concentration ratio data of the study area to obtain different salinization types and generate a two-dimensional spatiotemporal distribution map of soil salinization types.

[0007] Optionally, step S3 includes: S31: preprocessing the remote sensing image data and soil ion concentration data; S32: Determine the band value of the corresponding pixel point in the remote sensing image data according to the geographical location data; S33: Matching the band value of the pixel point with the soil ion concentration data to obtain a matching result; S34: Calculate salinization index data by matching the results and combining the band combination formula; S35: Calculate the milligram equivalent ratio of the required related ions according to the soil ion concentration data to obtain the soil ion concentration ratio; and obtain the corresponding soil salinization type according to the soil ion concentration ratio; S36: Match the salinization index data with the corresponding soil salinization types to construct a comprehensive dataset for salinization type inversion.

[0008] Optionally, step S31 includes: The preprocessing of the remote sensing image data includes: radiation calibration and atmospheric correction; The preprocessing of the soil ion concentration data includes: outlier removal, missing value filling and normalization.

[0009] Optionally, step S4 includes: S41: dividing the comprehensive data set into a training set and a test set; S42: Pearson correlation analysis was performed between soil salinization types and salinization index data in the comprehensive data set to determine sensitive band values; S43: Through the stacking strategy in ensemble learning technology, an ensemble learning regression model for inverting soil salinization types is constructed; S44: using the sensitive band value as an input factor of an integrated learning regression model, using the soil salinization type corresponding to the sensitive band as an output factor of the integrated learning regression model, and training the integrated learning regression model.

[0010] Optionally, step S44 includes: S44a: Obtain each base learner and meta learner in the ensemble learning regression model; S44b: Use k-fold cross validation to randomly divide the training set into k equal-sized and mutually exclusive subsets; S44c: Iteratively train the subset through each base learner to generate k prediction results; S44d: Calculate the average value of k prediction results to obtain an average prediction result; S44e: Use the average prediction result as the input feature of the meta-learner; S44f: Use the meta-learner to predict the soil ion concentration ratio and obtain the corresponding soil salinization type; S44d: Obtain the true soil ion concentration ratio, evaluate the base learner and the meta learner through the test set, and obtain the error between the true soil ion concentration ratio and the predicted value; S44h: Adjust the weight parameters of the ensemble learning regression model according to the error.

[0011] Optionally, step S6 includes: S61: extract remote sensing image data of the study area and pre-process the remote sensing image data to be measured; S62: Obtain the sensitive band value of each pixel in the study area according to the preprocessed remote sensing image data to be measured; S63: Perform feature engineering processing on the sensitive band value of each pixel; S64: Input the sensitive band value of each pixel after feature engineering processing into the trained integrated learning regression model to generate the predicted value of the soil ion concentration ratio data of each pixel in the remote sensing image data to be measured, that is, the soil ion concentration ratio data of the study area.

[0012] A computer-readable storage medium stores instructions. When the instructions are executed, a soil salinization type remote sensing inversion method based on integrated learning is performed.

[0013] The beneficial effects of the technical solution provided by this application are: 1. The stacking strategy in ensemble learning technology is adopted. An ensemble learning regression model is constructed, which combines the prediction results of multiple basic learners. Compared with the prediction of a single basic learner, this method can provide more accurate data and significantly reduce data errors. This ensemble method not only improves the accuracy of salinization type inversion, but also enhances the generalization ability of the model, solving the problem of insufficient information mining ability and poor transferability of existing remote sensing inversion models based on machine learning.

[0014] 2. According to the geographic location data, determine the band value of the corresponding pixel point in the remote sensing image data, and match the band value of the pixel point with the soil ion concentration data. By matching the satellite-ground data within a ±5-day time window and using the precise geographic coordinates of the soil sampling points, the band value of the pixel point of the surface reflectance in the multispectral remote sensing image is obtained, which reduces the cost of obtaining the salinization inversion data set and significantly improves the efficiency of obtaining training samples.

[0015] The salinization index data is calculated by combining the matching results with the band combination formula; the salinization index data is matched with the corresponding salinization type data to construct a comprehensive data set for salinization type inversion. By implementing this dual matching strategy, not only the coverage of training samples in time and space is enhanced, but also the cost of sample collection is effectively reduced.

[0016] 3. Pearson correlation analysis was performed based on the soil ion concentration ratio of the soil ion concentration data in the comprehensive data set and the band values ​​of the remote sensing image data to select the spectral bands that are more sensitive to soil salinization. This helps to reduce the inversion error caused by spectral information redundancy, thereby not only improving the data processing efficiency, but also helping to improve the accuracy of the inverted soil salinization type. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present application will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 is a step diagram in an embodiment of the present application; Figure 2 It is a diagram of the Stacking network structure in an embodiment of the present application; Figure 3 It is a diagram of the training prediction process of the Stacking network in an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to have a clearer understanding of the technical features, purposes and effects of the present application, the specific implementation methods of the present application are now described in detail with reference to the accompanying drawings.

[0019] The embodiment of the present application provides a remote sensing inversion method for soil salinization type based on ensemble learning.

[0020] Please refer to Figure 1 , Figure 1 This is a step diagram of a soil salinization type remote sensing inversion method based on ensemble learning in an embodiment of the present application, including: S1: Collect field sampling data of the concentration of various ions in the soil of the target area; the field sampling data includes: soil ion concentration data and geographical location data; Specifically, multiple soil sampling points are set in the target area to collect soil ion concentration data and corresponding geographical location information of the soil sampling points. Soil ion concentration data should cover key soluble salt ion indicators of soil such as chloride ions and sulfate ions. Subsequently, the collected soil ion concentration data is preprocessed as necessary, including: removal of outliers, filling of missing values, and normalization of data. These steps are crucial to improving data quality and accuracy, and provide a solid foundation for the construction and analysis of subsequent models.

[0021] S2: Acquire remote sensing image data of the target area; Specifically, in this application, remote sensing image data within a time window of ±5 days from the sampling time is obtained, and the time window can be adjusted according to the actual environment. The quality requirements of multispectral remote sensing images used for soil salinization inversion are relatively high. On the premise that the date and coverage are met, it should also be ensured that: the cloud cover is less than 10%, the weather is clear, and there are no clouds over the target area. Based on open data platforms such as NASA's Earth Observation Dataset and ESA's Copernicus, a detailed search is performed through keywords, geographic location, time range, band configuration, and cloud cover to screen out multispectral remote sensing image data that meet specific needs. After finding data that meets the conditions, use the preview function provided by the data source to check the thumbnail or partial sample of the image to further verify whether it meets expectations. Once it is confirmed that the data meets the requirements, you can select the download option and set the download parameters as needed, such as data format and resolution.

[0022] S3: Build a comprehensive dataset through field sampling data and remote sensing image data; Specifically, remote sensing image data and field sampling data are combined to perform precise data matching and construct a data set for salinization inversion.

[0023] Step S3 includes: S31: preprocessing the remote sensing image data and soil ion concentration data; Step S31 includes: The preprocessing of the remote sensing image data includes: radiation calibration and atmospheric correction; The preprocessing of the soil ion concentration data includes: outlier removal, missing value filling and normalization.

[0024] Specifically, the steps of preprocessing remote sensing image data include: converting the digital value of remote sensing image data into radiation value; obtaining the atmospheric transmission model; correcting the remote sensing image data according to the atmospheric transmission model and the atmospheric information data in the remote sensing image data. Preprocessing operations such as radiation calibration and atmospheric correction are performed on the remote sensing image data to improve the accuracy of data processing.

[0025] Radiometric calibration: Convert the digital values ​​of the original remote sensing image into reliable radiometric values. Usually, the radiometric reference targets in the image are used for calibration to ensure consistency in comparisons between different images and accuracy in subsequent physical quantity calculations.

[0026] Atmospheric correction: Use atmospheric transmission models and atmospheric information in remote sensing images to correct images to reduce the impact of atmospheric scattering and absorption on the spectrum of ground objects and improve the accuracy of remote sensing data.

[0027] S32: Determine the band value of the corresponding pixel point in the remote sensing image data according to the geographical location data; S33: Matching the band value of the pixel point with the soil ion concentration data to obtain a matching result; S34: Calculate salinization index data by matching the results and combining the band combination formula; S35: Calculate the milligram equivalent ratio of the required related ions according to the soil ion concentration data to obtain the soil ion concentration ratio; and obtain the corresponding soil salinization type according to the soil ion concentration ratio; Specifically, the soil ion concentration ratio is matched to different soil salinization types. The judgment of salinization type refers to the classification standard in the book "Basic Knowledge of Saline Soil", as shown in Table 1. The calculation formula of the milligram equivalent ratio of the required relevant ions is as follows:

[0028] Where A represents the milligram equivalent (mg eq) of each component, m represents the mass of each component in milligrams (mg), and M represents the relative molecular mass of each component.

[0029] Table 1 Salinization types

[0030] S36: Match the salinization index data with the corresponding soil salinization types to construct a comprehensive dataset for salinization type inversion.

[0031] In this application, the band combination formula includes: salt index and vegetation index, etc.; Some salt index calculation methods are as follows:

[0032]

[0033]

[0034]

[0035]

[0036]

[0037]

[0038]

[0039]

[0040]

[0041]

[0042] The calculation method of some vegetation indices is as follows:

[0043]

[0044]

[0045]

[0046]

[0047]

[0048] Among them, Red, Green, Blue, NIR, and SWIR1 are the red, green, blue, near infrared, and short-wave infrared bands of multispectral remote sensing images, respectively.

[0049] S4: Through ensemble learning technology, an ensemble learning regression model is constructed and trained with a comprehensive dataset; Specifically, by adopting the Stacking strategy in ensemble learning technology, an efficient regression model is constructed, which not only solves the problem of high cost of acquiring salinization inversion datasets in traditional technology, but also improves the inversion accuracy and generalization ability of the model.

[0050] Step S4 includes: S41: dividing the comprehensive data set into a training set and a test set; S42: Pearson correlation analysis was performed between soil salinization types and salinization index data in the comprehensive data set to determine sensitive band values; S43: Through the stacking strategy in ensemble learning technology, an ensemble learning regression model for inverting soil salinization types is constructed; S44: using the sensitive band value as an input factor of an integrated learning regression model, using the soil salinization type corresponding to the sensitive band as an output factor of the integrated learning regression model, and training the integrated learning regression model.

[0051] In this application, 80% of the salinization inversion dataset can be extracted as a training set, and the remaining 20% ​​can be used as a test set for the model. The specific extraction ratio can be modified according to the actual application scenario.

[0052] Specifically, the Stacking strategy in ensemble learning technology is used to build an ensemble learning regression model to fit the nonlinear relationship between satellite remote sensing reflectivity and soil ion concentration ratio, thereby improving the accuracy and generalization of the learning regression model.

[0053] refer to Figure 2 , Figure 2 It is a Stacking network structure diagram of the soil salinization type inversion method based on satellite remote sensing and integrated learning in the embodiment of the present application.

[0054] refer to Figure 3 , Figure 3 It is a training prediction process diagram of the Stacking network of the soil salinization type inversion method based on satellite remote sensing and integrated learning in the embodiment of the present application.

[0055] Step S44 includes: S44a: Obtain each base learner and meta learner in the ensemble learning regression model; S44b: Use k-fold cross validation to randomly divide the training set into k equal-sized and mutually exclusive subsets; S44c: Iteratively train the subset through each base learner to generate k prediction results; S44d: Calculate the average value of k prediction results to obtain an average prediction result; S44e: Use the average prediction result as the input feature of the meta-learner; S44f: Use the meta-learner to predict the soil ion concentration ratio and obtain the corresponding soil salinization type; S44d: Obtain the true soil ion concentration ratio, evaluate the base learner and the meta learner through the test set, and obtain the error between the true soil ion concentration ratio and the predicted value; S44h: Adjust the weight parameters of the ensemble learning regression model according to the error.

[0056] In this application, base learners include but are not limited to: Random Forest, GBDT, SVR, Xgboost, ExtraTrees, Decision Tree and MLP; meta learners include but are not limited to: Linear Regression.

[0057] Specifically, the k-fold cross-validation technique is used to avoid overfitting of regression models in ensemble learning. First, the training data set is randomly divided into k subsets of equal size and non-overlapping. In the next k rounds of iterative training, each round of training uses the data of k-1 subsets for model training, while predicting the remaining subset and the entire test set. This process is repeated k times to ensure that every sample in the training set is predicted.

[0058] After all iterations are completed, the k prediction results of the test set are averaged to obtain a comprehensive prediction result. This result will be used as the input feature of the meta-learner for training and testing. If there are T basic learners involved in training and prediction in this process, the algorithm will generate T groups of training sample features and T groups of test sample features. Ultimately, the meta-learner will use the prediction results of these basic learners to determine the prediction of the ion concentration ratio in salinized soil.

[0059] Calculate the error between the predicted value and the true value. This application uses the coefficient of determination ( ) and root mean square error (RMSE) are used to evaluate the prediction accuracy of the ensemble learning regression model. RMSE, also known as standard error, is very sensitive to extreme values ​​(either too high or too low) in the test data, so it is an effective tool to measure the prediction accuracy of the model. The lower the RMSE value, the smaller the difference between the model's prediction results and the actual values, that is, the higher the prediction accuracy. The formula for calculating the root mean square error is as follows:

[0060] The coefficient of determination (usually expressed as ) is an important indicator in statistics, which is used to measure the proportion of independent variables in explaining the changes in dependent variables. This indicator reflects the degree of fit of the statistical model to the data, that is, the quality of the model's predictive ability. The value of is between 0 and 1. The closer the value is to 1, the stronger the model's ability to explain the data and the better the fitting effect. The specific calculation formula is as follows:

[0061] in, and Represent the predicted value and the actual value respectively, N is the number of actual values ​​in the test set, is the average of the actual values.

[0062] S5: Obtain the remote sensing image data to be measured in the study area; S6: Use the trained ensemble learning regression model to predict the remote sensing image data to generate soil ion concentration ratio data of the study area; Step S6 includes: S61: extract remote sensing image data of the study area and pre-process the remote sensing image data to be measured; S62: Obtain the sensitive band value of each pixel in the study area according to the preprocessed remote sensing image data to be measured; S63: Perform feature engineering processing on the sensitive band value of each pixel; S64: Input the sensitive band value of each pixel after feature engineering processing into the trained integrated learning regression model to generate the predicted value of the soil ion concentration ratio data of each pixel in the remote sensing image data to be measured, that is, the soil ion concentration ratio data of the study area.

[0063] S7: Threshold division is performed on the soil ion concentration ratio data of the study area to obtain different salinization types and generate a two-dimensional spatiotemporal distribution map of soil salinization types.

[0064] The present application also discloses a computer-readable storage medium storing a plurality of instructions suitable for a processor to load to execute the above-mentioned soil salinization type remote sensing inversion method based on integrated learning.

[0065] The above are only exemplary embodiments of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure.

[0066] This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art not described in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A remote sensing inversion method for soil salinization type based on ensemble learning, characterized in that: The method comprises the following steps: S1: Collect field sampling data of the concentration of various ions in the soil of the target area; the field sampling data includes: soil ion concentration data and geographical location data; S2: Acquire remote sensing image data of the target area; S3: Build a comprehensive dataset through field sampling data and remote sensing image data; S4: Through ensemble learning technology, an ensemble learning regression model is constructed and trained with a comprehensive dataset; S5: Obtain the remote sensing image data to be measured in the study area; S6: Use the trained ensemble learning regression model to predict the remote sensing image data to generate soil ion concentration ratio data of the study area; S7: Threshold division is performed on the soil ion concentration ratio data of the study area to obtain different salinization types and generate a two-dimensional spatiotemporal distribution map of soil salinization types.

2. The soil salinization type remote sensing inversion method based on ensemble learning as claimed in claim 1, characterized in that: Step S3 includes: S31: preprocessing the remote sensing image data and soil ion concentration data; S32: Determine the band value of the corresponding pixel point in the remote sensing image data according to the geographical location data; S33: Matching the band value of the pixel point with the soil ion concentration data to obtain a matching result; S34: Calculate salinization index data by matching the results and combining the band combination formula; S35: Calculate the milligram equivalent ratio of the required related ions according to the soil ion concentration data to obtain the soil ion concentration ratio; and obtain the corresponding soil salinization type according to the soil ion concentration ratio; S36: Match the salinization index data with the corresponding soil salinization types to construct a comprehensive dataset for salinization type inversion.

3. The soil salinization type remote sensing inversion method based on ensemble learning as claimed in claim 2, characterized in that: Step S31 includes: The preprocessing of the remote sensing image data includes: radiation calibration and atmospheric correction; The preprocessing of the soil ion concentration data includes: outlier removal, missing value filling and normalization.

4. The soil salinization type remote sensing inversion method based on ensemble learning as claimed in claim 1, characterized in that: Step S4 includes: S41: dividing the comprehensive data set into a training set and a test set; S42: Pearson correlation analysis was performed between soil salinization types and salinization index data in the comprehensive data set to determine sensitive band values; S43: Through the stacking strategy in ensemble learning technology, an ensemble learning regression model for inverting soil salinization types is constructed; S44: using the sensitive band value as an input factor of an integrated learning regression model, using the soil salinization type corresponding to the sensitive band as an output factor of the integrated learning regression model, and training the integrated learning regression model.

5. The soil salinization type remote sensing inversion method based on ensemble learning as claimed in claim 4, characterized in that: Step S44 includes: S44a: Obtain each base learner and meta learner in the ensemble learning regression model; S44b: Use k-fold cross validation to randomly divide the training set into k equal-sized and mutually exclusive subsets; S44c: Iteratively train the subset through each base learner to generate k prediction results; S44d: Calculate the average value of k prediction results to obtain an average prediction result; S44e: Use the average prediction result as the input feature of the meta-learner; S44f: Use the meta-learner to predict the soil ion concentration ratio and obtain the corresponding soil salinization type; S44d: Obtain the true soil ion concentration ratio, evaluate the base learner and the meta learner through the test set, and obtain the error between the true soil ion concentration ratio and the predicted value; S44h: Adjust the weight parameters of the ensemble learning regression model according to the error.

6. The soil salinization type remote sensing inversion method based on ensemble learning as claimed in claim 1, characterized in that: Step S6 includes: S61: extract remote sensing image data of the study area and pre-process the remote sensing image data to be measured; S62: Obtain the sensitive band value of each pixel in the study area according to the preprocessed remote sensing image data to be measured; S63: Perform feature engineering processing on the sensitive band value of each pixel; S64: Input the sensitive band value of each pixel after feature engineering processing into the trained integrated learning regression model to generate the predicted value of the soil ion concentration ratio data of each pixel in the remote sensing image data to be measured, that is, the soil ion concentration ratio data of the study area.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed by a computer, the method according to any one of claims 1 to 6 is executed.

Images