Drought monitoring method and system integrating multi-source remote sensing data

Abstract

The invention relates to a drought monitoring method integrating multi-source remote sensing data. The method comprises the following steps: S1, data collection: collecting multi-source data from multiple remote sensing platforms; s2, data preprocessing: carrying out preprocessing operation on the collected data; s3, class specific feature selection: calculating a correlation coefficient between features in each class by using a Pearson's correlation coefficient, constructing a class specific correlation matrix, understanding the distribution condition of the features in different classes and the interaction between the features by visualizing the class specific correlation matrix, and according to an analysis result of the class specific correlation matrix, selecting a class specific feature; selecting features with significant correlation in different categories as key features; s4, model construction and training: training a neural network by using the selected features and label data, introducing a CBPSR algorithm to optimize the structure and performance of the neural network in the training process, and constructing a drought monitoring model; and S5, real-time monitoring: realizing real-time monitoring and prediction of drought according to the real-time data and the drought monitoring model.

Description

Technical Field

[0001] The present invention relates to the field of drought monitoring, and in particular to a drought monitoring method and system integrating multi-source remote sensing data. Background Art

[0002] Drought disaster is a natural disaster that is prevalent worldwide and has a profound impact on human society and the natural environment. As a new type of earth observation method, remote sensing technology has the characteristics of macro, economical, and rapid, and has obvious advantages in drought monitoring. In particular, satellite remote sensing technology can obtain drought information on a large scale, quickly, dynamically, and objectively, effectively making up for the lack of spatial discontinuity and point-to-surface coverage of ground stations. In drought monitoring, multi-source remote sensing data has significant advantages. First, multi-source data can provide more comprehensive drought information. Different remote sensing platforms (such as satellites, drones, etc.) and sensors (such as optical, thermal infrared, microwave, etc.) can obtain drought-related information of different scales and resolutions, such as vegetation index, soil moisture, surface temperature, etc.

[0003] Feature selection is a key step in traditional drought monitoring methods that integrate multi-source remote sensing data. However, there is a possibility that when evaluating the overall feature importance of all categories, the features that contribute most to the overall classification performance are selected, thereby ignoring the unique role of certain features in specific categories. This may lead to poor performance of the model on specific categories. Therefore, it is particularly important to propose a drought monitoring method and system that integrates multi-source remote sensing data. Summary of the invention

[0004] The purpose of the present invention is to solve the problem in the prior art that in the traditional drought monitoring method integrating multi-source remote sensing data, feature selection is a key step. However, there is indeed a possibility that when evaluating the overall feature importance of all categories, the features that contribute most to the overall classification performance are selected, thereby ignoring the unique role of certain features in specific categories. This may lead to the disadvantage of poor performance of the model on specific categories, and a drought monitoring method and system integrating multi-source remote sensing data are provided.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A drought monitoring method integrating multi-source remote sensing data comprises the following steps: S1: Data collection: Collect multi-source data such as vegetation index and soil moisture from various remote sensing platforms such as satellites and drones; S2: Data preprocessing: preprocessing the collected data; S3: Class-specific feature selection: The Pearson correlation coefficient is used to calculate the correlation coefficient between features in each class, and a class-specific correlation matrix is ​​constructed. By visualizing the class-specific correlation matrix, the distribution of features in different classes and the interaction between features can be understood. Based on the analysis results of the class-specific correlation matrix, features with significant correlations in different classes are selected as key features. These features can more accurately reflect the drought conditions, thereby improving the performance of the model. S4: Model construction and training: The selected features and label data are used to train the neural network. During the training process, the CBPSR algorithm is introduced to optimize the structure and performance of the neural network, and then a drought monitoring model is constructed; S5: Real-time monitoring: Real-time monitoring and prediction of drought can be achieved based on real-time data and drought monitoring models.

[0006] The above scheme further comprises: Furthermore, in S1, satellite remote sensing technology is used to acquire and preliminarily evaluate a wide range of surface information, including vegetation index, soil moisture, and surface temperature, and unmanned aerial vehicle remote sensing technology is used to acquire high-resolution images and multispectral / hyperspectral data.

[0007] Further, in S2, the preprocessing operations include data cleaning, data denoising and data normalization, wherein the data cleaning is intended to identify and correct errors, anomalies or missing values ​​in the data, the data denoising is intended to reduce noise in the data, and the data normalization is intended to scale the data to be within a specific numerical range.

[0008] Further, in S3, the specific steps to construct the class-specific correlation matrix are: Data preparation: The preprocessed data forms a data set, which contains multiple categories (such as category A, category B, etc.) and multiple features (such as feature X, feature Y, etc.). Each sample in the data set belongs to a specific category and has values ​​for all features. Calculate the Pearson correlation coefficient: For each category, calculate the Pearson correlation coefficient between its features. The calculation formula of the Pearson correlation coefficient is: ; in, represents the Pearson correlation coefficient, and Represent the values ​​of the two features corresponding to category i, and Represent the means of the two features respectively, n represents the number of categories, and the Pearson correlation coefficient calculation formula obtains a value between -1 and 1, which represents the degree of linear correlation between the two features, where 1 represents a complete positive correlation, -1 represents a complete negative correlation, and 0 represents no correlation; Construct a class-specific correlation matrix: Divide the data into categories, calculate the Pearson correlation coefficient for the features in each category, and for each category, organize the calculated Pearson correlation coefficients between all features into a matrix, namely the class-specific correlation matrix, in which each element represents the Pearson correlation coefficient between two features.

[0009] Furthermore, in S3, a heat map is used to visualize the class-specific correlation matrix. In the heat map, different colors represent different correlation coefficient values, and the depth of the color reflects the strength of the correlation coefficient.

[0010] Furthermore, in S3, the specific steps of selecting features with significant correlation in different categories as key features are: Set threshold: Set a correlation threshold, which means that only features with correlation coefficients greater than or equal to this correlation threshold are selected as key features; Select key features: After setting the correlation threshold, for each category, traverse the rows or columns of the class-specific correlation matrix and find those features whose correlation coefficients are greater than or equal to the correlation threshold. The features are the key features with significant correlation in the category. Verification and optimization: After selecting key features, the selected key features are verified and optimized. The purpose of verification is to ensure that the selected features can accurately reflect drought conditions and improve the performance of the model. Optimization methods may include adjusting thresholds, increasing or decreasing the number of features, etc. Through verification and optimization, we can obtain a more accurate and reliable set of key features.

[0011] Furthermore, in S4, the specific steps of introducing the CBPSR algorithm to optimize the structure and performance of the neural network are as follows: Evaluate contribution utility: CBPSR evaluates the contribution utility of each connection and weight to determine which units are inefficient, which is usually measured based on the magnitude of the product of the activation value of the hidden unit and the output weight. In the drought monitoring task, inefficient units may correspond to those hidden units that contribute less to the distinction of drought categories; Selective reinitialization: Selectively reinitialize these inefficient units. Reinitialization usually means setting the weights and biases of the units to initial or random values ​​to introduce new randomness and maintain the plasticity of the network, which helps the model jump out of the local optimal solution during training and explore a wider parameter space; Maturity threshold: Set a maturity threshold. To protect newly added hidden units from immediate reinitialization, a hidden unit is reinitialized only when its age (i.e., the number of times or epochs it has been trained) exceeds the maturity threshold. This helps ensure the stability and convergence of the model during training.

[0012] A drought monitoring system used in a drought monitoring method integrating multi-source remote sensing data includes: Data collection and preprocessing module: responsible for collecting multi-source data such as vegetation index, soil moisture, surface temperature, etc. from various remote sensing platforms (such as satellites, drones, etc.), and preprocessing the collected multi-source data to ensure the quality and consistency of the data, providing a reliable basis for subsequent analysis and modeling; Class-specific feature selection module: responsible for extracting features that are significantly relevant for drought monitoring from the preprocessed data; Model building and training module: Use the selected features and label data to train the neural network model and build a drought monitoring model; Real-time monitoring module: responsible for the real-time monitoring and prediction of drought based on real-time data and drought monitoring models.

[0013] The present invention has the following beneficial effects: In the present invention, by constructing a class-specific correlation matrix, the distribution and interaction of features in different categories are intuitively displayed, which can more accurately capture the feature differences between different drought categories, thereby improving the classification performance and generalization ability of the model. It can flexibly respond to drought monitoring needs in different regions and time scales, and has strong adaptability and scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A method step diagram of a drought monitoring method integrating multi-source remote sensing data proposed by the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] See also Figure 1 As shown, the present invention is a drought monitoring method integrating multi-source remote sensing data, comprising the following steps: S1: Data collection: Collect multi-source data such as vegetation index and soil moisture from various remote sensing platforms such as satellites and drones; S2: Data preprocessing: preprocessing the collected data; S3: Class-specific feature selection: The Pearson correlation coefficient is used to calculate the correlation coefficient between features in each class, and a class-specific correlation matrix is ​​constructed. By visualizing the class-specific correlation matrix, the distribution of features in different classes and the interaction between features can be understood. Based on the analysis results of the class-specific correlation matrix, features with significant correlations in different classes are selected as key features. These features can more accurately reflect the drought conditions, thereby improving the performance of the model. S4: Model construction and training: The selected features and label data are used to train the neural network. During the training process, the CBPSR algorithm is introduced to optimize the structure and performance of the neural network, and then a drought monitoring model is constructed; S5: Real-time monitoring: Real-time monitoring and prediction of drought can be achieved based on real-time data and drought monitoring models.

[0017] In one embodiment, for the above S1, in S1, satellite remote sensing technology is used to acquire and preliminarily evaluate a large range of surface information, wherein the surface information includes vegetation index, soil moisture and surface temperature, and unmanned aerial vehicle remote sensing technology is used to acquire high-resolution images and multispectral / hyperspectral data.

[0018] In one embodiment, for the above S2, in S2, the preprocessing operation includes data cleaning, data denoising and data normalization, wherein the data cleaning is intended to identify and correct errors, anomalies or missing values ​​in the data, the data denoising is intended to reduce the noise in the data, and the data normalization is intended to scale the data to be within a specific numerical range.

[0019] In one embodiment, for the above S3, in S3, the specific steps of constructing the class-specific correlation matrix are: Data preparation: The preprocessed data forms a data set, which contains multiple categories (such as category A, category B, etc.) and multiple features (such as feature X, feature Y, etc.). Each sample in the data set belongs to a specific category and has values ​​for all features. Calculate the Pearson correlation coefficient: For each category, calculate the Pearson correlation coefficient between its features. The calculation formula of the Pearson correlation coefficient is: ; in, represents the Pearson correlation coefficient, and Represent the values ​​of the two features corresponding to category i, and Represent the means of the two features respectively, n represents the number of categories, and the Pearson correlation coefficient calculation formula obtains a value between -1 and 1, which represents the degree of linear correlation between the two features, where 1 represents a complete positive correlation, -1 represents a complete negative correlation, and 0 represents no correlation; Construct a class-specific correlation matrix: Divide the data into categories, calculate the Pearson correlation coefficient for the features in each category, and for each category, organize the calculated Pearson correlation coefficients between all features into a matrix, namely the class-specific correlation matrix, in which each element represents the Pearson correlation coefficient between two features.

[0020] Assume that we have a dataset with two classes (different regions or different time periods) and four features (NDVI, AVI, VCI, TADI). Follow these steps to build a class-specific correlation matrix and visualize it: Preprocess the dataset to ensure data quality and consistency.

[0021] The Pearson correlation coefficient was calculated for feature pairs (such as NDVI and AVI, NDVI and VCI, etc.) in each category (different regions or different time periods).

[0022] Construct a correlation matrix for each category. For example, the correlation matrix for category A might look like this:

[0023] in, (autocorrelation coefficient is 1), (The correlation coefficient is symmetrical).

[0024] Calculate the Pearson correlation coefficient for each drought index pair in each region or time period. Construct a correlation matrix for each region or time period.

[0025] In one embodiment, for the above S3, in S3, a heat map is used to visualize the class-specific correlation matrix. In the heat map, different colors represent different correlation coefficient values, and the depth of the color reflects the strength of the correlation coefficient.

[0026] In one embodiment, for the above S3, in S3, the specific steps of selecting features with significant correlation in different categories as key features are: Set threshold: Set a correlation threshold, which means that only features with correlation coefficients greater than or equal to this correlation threshold are selected as key features; Select key features: After setting the correlation threshold, for each category, traverse the rows or columns of the class-specific correlation matrix and find those features whose correlation coefficients are greater than or equal to the correlation threshold. The features are the key features with significant correlation in the category. Verification and optimization: After selecting key features, the selected key features are verified and optimized. The purpose of verification is to ensure that the selected features can accurately reflect drought conditions and improve the performance of the model. Optimization methods may include adjusting thresholds, increasing or decreasing the number of features, etc. Through verification and optimization, we can obtain a more accurate and reliable set of key features.

[0027] Suppose we have two categories of drought monitoring data, namely Category A (spring drought data in a certain area) and Category B (summer drought data in the area). Each category contains three drought indices: NDVI (Normalized Difference Vegetation Index), TCI (Temperature Condition Index), and SMI (Soil Moisture Index). We have constructed a class-specific correlation matrix for these two categories and set the correlation threshold to 0.7.

[0028] For category A, we traverse the rows or columns of the class-specific correlation matrix and find that the correlation coefficient between NDVI and SMI is greater than 0.7, while the correlation coefficients between TCI and the other two indices are all less than 0.7. Therefore, we select NDVI and SMI as the key features of category A.

[0029] For category B, we also traverse the rows or columns of the class-specific correlation matrix and find that the correlation coefficient between NDVI and SMI is still greater than 0.7, while the correlation coefficient between TCI and NDVI is close to 0.7 but less than 0.7 (assuming 0.68), and the correlation coefficient with SMI is even smaller. In this case, we still choose NDVI and SMI as the key features of category B because they have strong correlations in both categories.

[0030] In one embodiment, for the above S4, in S4, the specific steps of introducing the CBPSR algorithm to optimize the structure and performance of the neural network are as follows: Evaluate contribution utility: CBPSR evaluates the contribution utility of each connection and weight to determine which units are inefficient, which is usually measured based on the magnitude of the product of the activation value of the hidden unit and the output weight. In the drought monitoring task, inefficient units may correspond to those hidden units that contribute less to the distinction of drought categories; Selective reinitialization: Selectively reinitialize these inefficient units. Reinitialization usually means setting the weights and biases of the units to initial or random values ​​to introduce new randomness and maintain the plasticity of the network, which helps the model jump out of the local optimal solution during training and explore a wider parameter space; Maturity threshold: Set a maturity threshold. To protect newly added hidden units from immediate reinitialization, a hidden unit is reinitialized only when its age (i.e., the number of times or epochs it has been trained) exceeds the maturity threshold. This helps ensure the stability and convergence of the model during training.

[0031] A drought monitoring system for integrated multi-source remote sensing data used in a drought monitoring method for integrated multi-source remote sensing data includes: Data collection and preprocessing module: responsible for collecting multi-source data such as vegetation index, soil moisture, surface temperature, etc. from various remote sensing platforms (such as satellites, drones, etc.), and preprocessing the collected multi-source data to ensure the quality and consistency of the data, providing a reliable basis for subsequent analysis and modeling; Class-specific feature selection module: responsible for extracting features that are significantly relevant for drought monitoring from the preprocessed data; Model building and training module: Use the selected features and label data to train the neural network model and build a drought monitoring model; Real-time monitoring module: responsible for the real-time monitoring and prediction of drought based on real-time data and drought monitoring models.

[0032] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A drought monitoring method integrating multi-source remote sensing data, characterized in that: The following steps are involved: S1: Data collection: Collect multi-source data from multiple remote sensing platforms; S2: Data preprocessing: preprocessing the collected data; S3: Class-specific feature selection: Pearson correlation coefficient is used to calculate the correlation coefficient between features in each class, and a class-specific correlation matrix is ​​constructed. By visualizing the class-specific correlation matrix, the distribution of features in different classes and the interaction between features can be understood. Based on the analysis results of the class-specific correlation matrix, features with significant correlation in different classes are selected as key features. S4: Model construction and training: The selected features and label data are used to train the neural network. During the training process, the CBPSR algorithm is introduced to optimize the structure and performance of the neural network, and then a drought monitoring model is constructed; S5: Real-time monitoring: Real-time monitoring and prediction of drought can be achieved based on real-time data and drought monitoring models.

2. The drought monitoring method based on integrated multi-source remote sensing data according to claim 1, characterized in that: In S1, satellite remote sensing technology is used to acquire and preliminarily evaluate a wide range of surface information, including vegetation index, soil moisture, and surface temperature, and unmanned aerial vehicle remote sensing technology is used to acquire images and multispectral / hyperspectral data.

3. The drought monitoring method based on integrated multi-source remote sensing data according to claim 1, characterized in that: In S2, the preprocessing operations include data cleaning, data denoising, and data normalization, wherein data cleaning is intended to identify and correct errors, anomalies, or missing values ​​in the data, data denoising is intended to reduce noise in the data, and data normalization is intended to scale the data to be within a specific numerical range.

4. The drought monitoring method based on integrated multi-source remote sensing data according to claim 1, characterized in that: In S3, the specific steps to construct a class-specific correlation matrix are: Data preparation: The preprocessed data forms a dataset, which contains multiple categories and multiple features. Each sample in the dataset belongs to a specific category and has values ​​for all features. Calculate the Pearson correlation coefficient: For each category, calculate the Pearson correlation coefficient between its features. The calculation formula of the Pearson correlation coefficient is: ; in, represents the Pearson correlation coefficient, and Represent the values ​​of the two features corresponding to category i, and Represent the means of the two features respectively, n represents the number of categories, and the Pearson correlation coefficient calculation formula obtains a value between -1 and 1, which represents the degree of linear correlation between the two features, where 1 represents a complete positive correlation, -1 represents a complete negative correlation, and 0 represents no correlation; Construct a class-specific correlation matrix: Divide the data into categories, calculate the Pearson correlation coefficient for the features in each category, and for each category, organize the calculated Pearson correlation coefficients between all features into a matrix, namely the class-specific correlation matrix, in which each element represents the Pearson correlation coefficient between two features.

5. The drought monitoring method based on integrated multi-source remote sensing data according to claim 1, characterized in that: In S3, a heat map is used to visualize the class-specific correlation matrix. In the heat map, different colors represent different correlation coefficient values, and the depth of the color reflects the strength of the correlation coefficient.

6. The drought monitoring method based on integrated multi-source remote sensing data according to claim 1, characterized in that: In S3, the specific steps for selecting features with significant correlation in different categories as key features are: Set threshold: Set a correlation threshold, which means that only features with correlation coefficients greater than or equal to this correlation threshold are selected as key features; Select key features: After setting the correlation threshold, for each category, traverse the rows or columns of the class-specific correlation matrix and find those features whose correlation coefficients are greater than or equal to the correlation threshold. The features are the key features with significant correlation in the category. Verification and optimization: After selecting the key features, verify and optimize the selected key features.

7. The drought monitoring method based on integrated multi-source remote sensing data according to claim 1, characterized in that: In S4, the specific steps of introducing the CBPSR algorithm to optimize the structure and performance of the neural network are as follows: Evaluate contribution utility: CBPSR evaluates the contribution utility of each connection and weight to determine which units are inefficient; Selective reinitialization: selectively reinitialize these inefficient units; Maturity threshold: Set the maturity threshold. Only when the age of a hidden unit exceeds the maturity threshold, the hidden unit will be reinitialized.

8. The drought monitoring system for integrating multi-source remote sensing data used in the drought monitoring method for integrating multi-source remote sensing data according to claim 1 is characterized in that: include: Data collection and preprocessing module: responsible for collecting multi-source data from various remote sensing platforms and preprocessing the collected multi-source data; Class-specific feature selection module: responsible for extracting features that are significantly relevant for drought monitoring from the preprocessed data; Model building and training module: Use the selected features and label data to train the neural network model and build a drought monitoring model; Real-time monitoring module: responsible for real-time monitoring and prediction of drought based on real-time data and drought monitoring models.

Images