Method and system for terrain inversion under wetland vegetation based on fusion of unmanned aerial vehicle data and machine learning

Abstract

The invention provides a wetland vegetation terrain inversion method and system based on unmanned aerial vehicle data and machine learning fusion, and the method comprises the steps: obtaining an unmanned aerial vehicle multispectral image and laser radar point cloud data of a target wetland area, and generating a vegetation coverage area multidimensional data set through combining with RTK actual measurement ground elevation sampling points; the feature parameters and the terrain correlation parameters are stored in a vegetation feature set and a terrain correlation parameter set; forming a training sample set by the vegetation characteristic parameters, the terrain correlation parameters and RTK actual measurement ground elevation, constructing a terrain inversion model, and calculating a mapping relation between the vegetation characteristic parameters and the terrain correlation parameters and the ground elevation; calculating the feature correlation degree between the vegetation feature parameters and the terrain correlation parameters in the training sample set and the ground elevation, and the cross-modal coupling degree between the parameters; calculating a terrain inversion confidence coefficient according to the feature correlation degree, the cross-modal coupling degree and the parameter weight; and generating a ground elevation inversion result. According to the invention, the precision and reliability of terrain inversion of the vegetation-covered area can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing mapping and geographic information, and more particularly, relates to a wetland vegetation-underlying terrain inversion method and system based on unmanned aerial vehicle data and machine learning fusion. BACKGROUND

[0002] In the field of geographic information science and wetland ecology research today, the accurate inversion of wetland vegetation-underlying terrain has always been a highly challenging topic. Wetland ecosystems, due to their unique ecological functions and biodiversity, play a crucial role in global ecological balance. Accurate acquisition of terrain information under wetland vegetation coverage is of great significance for in-depth understanding of wetland ecological processes, assessment of wetland ecosystem health, and development of scientific and reasonable wetland protection and restoration strategies. However, traditional terrain mapping methods often face many limitations when dealing with vegetation-covered wetland areas. These methods mostly rely on field measurements or a single remote sensing data source. Field measurements are time-consuming and labor-intensive in wetland environments and are difficult to cover comprehensively. A single remote sensing data source, due to the shielding effect of vegetation, is difficult to directly obtain the true terrain information under vegetation, resulting in low precision of the acquired terrain data, which is difficult to meet the fine needs of wetland ecological research and protection work.

[0003] In recent years, with the rapid development of unmanned aerial vehicle technology and machine learning technology, new opportunities have been brought to solve this problem. Unmanned aerial vehicle technology, due to its high flexibility, low cost, and ability to quickly acquire high-resolution images and three-dimensional point cloud data, has been widely applied in the field of geographic information data acquisition. It can penetrate into the interior of the wetland and acquire multispectral images and laser radar point cloud data of the vegetation-covered area, providing a more abundant data basis for terrain inversion. Machine learning technology, with its powerful data processing and pattern recognition capabilities, has shown great potential in data mining and analysis. By building appropriate machine learning models, complex data acquired by unmanned aerial vehicles can be deeply analyzed and mined, establishing a mapping relationship between vegetation features and terrain correlation parameters, and providing the possibility for high-precision wetland vegetation-underlying terrain inversion.

[0004] There are at least the following problems or defects in the prior art: On the one hand, the precision of traditional terrain inversion methods in vegetation-covered areas is limited, making it difficult to meet the high-precision requirements of wetland ecological research on terrain data; on the other hand, although unmanned aerial vehicle technology and machine learning technology have applications in related fields, the systematic method of deeply integrating the two for wetland vegetation-underlying terrain inversion is not yet perfect, lacking effective model construction and parameter optimization means, resulting in the reliability and accuracy of the inversion results needing to be improved. SUMMARY

[0005] The application provides a wetland vegetation-underlying terrain inversion method and system based on unmanned aerial vehicle data and machine learning fusion.

[0006] In a first aspect of the application, a wetland vegetation-underlying terrain inversion method based on unmanned aerial vehicle data and machine learning fusion is provided, comprising: Obtaining unmanned aerial vehicle multispectral images and laser radar point cloud data of a target wetland, combining RTK measured ground elevation sampling points, and generating a vegetation-covered area multidimensional data set after preprocessing; Extracting vegetation feature parameters and terrain correlation parameters from the multidimensional data set and storing them in a vegetation feature set and a terrain correlation parameter set; Based on the vegetation feature set and the terrain correlation parameter set, each vegetation feature parameter and terrain correlation parameter and the corresponding RTK measured ground elevation form a training sample set, a terrain inversion model is constructed using a convolutional neural network, and the mapping relationship between the vegetation feature parameters and terrain correlation parameters and the ground elevation is calculated through the terrain inversion model; Calculating the feature correlation degree between the vegetation feature parameters and terrain correlation parameters and the ground elevation in each training sample set, and the cross-modal coupling degree between the parameters; calculating the terrain inversion confidence according to the feature correlation degree, the cross-modal coupling degree and the parameter weight; Based on the training samples with confidence exceeding the threshold, a ground elevation inversion result under the wetland vegetation coverage is generated.

[0007] Further, the calculation of the feature correlation degree between the vegetation feature parameters and terrain correlation parameters and the ground elevation in each training sample set comprises: Obtaining the vegetation spectrum set of each grid cell in the multidimensional data set, storing the vegetation spectrum and terrain correlation parameters containing each training sample in the multidimensional data set in a first feature set, storing the ground elevation corresponding to each training sample in a second feature set, and calculating the similarity between the first feature set and the second feature set as the feature correlation degree.

[0008] Further, the similarity between the first feature set and the second feature set is obtained by using a random forest model to obtain a first feature vector of the first feature set and a second feature vector of the second feature set, and calculating the Pearson correlation coefficient between the first feature vector and the second feature vector.

[0009] Further, the similarity between the first feature set and the second feature set is obtained by using principal component analysis and the variance contribution rate of the first feature set and the variance contribution rate of the second feature set to respectively obtain a first dimension reduction matrix of the first feature set and a second dimension reduction matrix of the second feature set, and calculating the maximum covariance value between the first dimension reduction matrix and the second dimension reduction matrix.

[0010] Further, the first dimension reduction matrix of the first feature set and the second dimension reduction matrix of the second feature set are obtained by the following steps: The first feature set and the second feature set are normalized, and a covariance matrix is calculated; The covariance matrix is subjected to eigenvalue decomposition, and the eigenvectors corresponding to the first k largest eigenvalues are selected as the projection matrix; The original feature set is multiplied by the corresponding projection matrix respectively, and the first dimension reduction matrix and the second dimension reduction matrix are obtained.

[0011] Further, the cross-modal coupling degree between the plurality of terrain correlation parameters is obtained by calculating the reciprocal of the variance inflation factor of the second dimension reduction matrix.

[0012] Further, the weight of the vegetation feature parameter and the terrain correlation parameter in each training sample is obtained by feeding back the ground elevation inversion result corresponding to each training sample to the multi-dimensional data set, calculating the random forest feature importance score of each vegetation feature parameter and terrain correlation parameter in the updated multi-dimensional data set, and then normalizing the weight of the vegetation feature parameter and the terrain correlation parameter in each training sample.

[0013] Further, the confidence of each training sample is obtained by normalizing the weight of the vegetation feature parameter and the terrain correlation parameter in each training sample, and then weighting and summing the corresponding feature correlation degree and cross-modal coupling degree.

[0014] Further, the vegetation feature parameters and the terrain correlation parameters are extracted from the multi-dimensional data set and stored in the vegetation feature set and the terrain correlation parameter set, including: Radiometric calibration is performed on the unmanned aerial vehicle multi-spectral image to generate a normalized vegetation spectral layer; Based on the laser radar point cloud data, the vegetation canopy height and the point cloud density parameters are extracted; The normalized vegetation spectral layer, the vegetation canopy height parameter and the point cloud density parameter are grid-matched with the RTK measured ground elevation through spatial overlay analysis, and the vegetation feature parameters and the terrain correlation parameters of each grid unit are obtained.

[0015] In the second aspect of the present application, a wetland vegetation under-ground terrain inversion system based on unmanned aerial vehicle data and machine learning fusion is provided, comprising: A data acquisition module is used to acquire unmanned aerial vehicle multi-spectral images, laser radar point cloud data and RTK measured ground elevation data of a target wetland, and to generate a vegetation coverage multi-dimensional data set after preprocessing; A feature extraction module is used to extract vegetation feature parameters and terrain correlation parameters from the multi-dimensional data set and store them in the vegetation feature set and the terrain correlation parameter set; a model training module configured to construct a terrain inversion model based on the set of vegetation feature parameters and the set of terrain-related parameters and the ground elevation, and calculate a mapping relationship between the vegetation feature parameters and the terrain-related parameters and the ground elevation; a confidence calculation module configured to calculate the feature correlation degree, the cross-modal coupling degree and the parameter weight of the training sample, and generate a terrain inversion confidence evaluation result; a terrain inversion module configured to generate a ground elevation inversion result under wetland vegetation coverage according to the training sample with confidence up to standard.

[0016] The above-mentioned embodiments of the present application have at least the following beneficial effects: the present application can improve the accuracy and efficiency of terrain inversion under wetland vegetation. By fusing unmanned aerial vehicle multi-spectral image and laser radar point cloud data, more comprehensive and high-resolution multi-dimensional data set of vegetation coverage area can be obtained, providing a solid data foundation for terrain inversion. At the same time, the convolutional neural network is used to construct the terrain inversion model, which can automatically learn the complex mapping relationship between the vegetation feature parameters and the terrain-related parameters and the ground elevation, avoid the loss of ground elevation information caused by vegetation shielding in traditional methods, and only obtain the vegetation canopy height, which can improve the accuracy of the inversion result. In addition, by calculating the correlation degree, coupling degree and weight of the training sample, and generating the confidence evaluation result, the performance of the inversion model can be quantitatively evaluated to ensure the reliability of the inversion result, and high-quality terrain data support is provided for wetland ecological research.

[0017] The present application can also realize the rapid inversion of terrain under wetland vegetation, shorten the time of data processing and analysis. The efficiency of unmanned aerial vehicle data acquisition enables the acquisition of data of large area wetland in a short time, and the automatic processing capability of machine learning model can further accelerate the speed of terrain inversion, so as to meet the demand of rapid update of terrain information in wetland protection and repair work. At the same time, the systematic design of the present application makes the whole inversion process more standardized and standardized, which is convenient for popularization and application to different wetland ecosystems, can provide an efficient and reliable terrain inversion tool for wetland ecological research and protection work, and is helpful to promote the development of wetland science and sustainable management of wetland ecosystem. BRIEF DESCRIPTION OF DRAWINGS

[0018] The above and other objects, features and advantages of the exemplary embodiments of the present application will be more apparent from the following detailed description taken in conjunction with the accompanying drawings, in which: Figure 1 A flowchart of a wetland vegetation under terrain inversion method based on unmanned aerial vehicle data and machine learning fusion provided by an embodiment of the present application is shown in the figure; Figure 2A structural schematic diagram of a wetland vegetation under-ground topography inversion system based on unmanned aerial vehicle data and machine learning fusion is provided for an embodiment of the present application. Figure 3 A structural schematic diagram of an electronic device of an embodiment of the present application is schematically shown. DETAILED DESCRIPTION

[0019] The principles and spirits of the present application will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are given only for better understanding of the present application by those skilled in the art, and do not limit the scope of the present application in any way. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0020] Those skilled in the art know that the embodiments of the present application can be implemented as a system, a device, an apparatus, a method or a computer program product. Therefore, the present application can be embodied as a complete hardware, a complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0021] It should be noted that the number of any elements in the drawings is used for example only and not limitation, and any naming is only for distinction and does not have any limiting meaning.

[0022] The principles and spirits of the present application will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are given only for better understanding of the present application by those skilled in the art, and do not limit the scope of the present application in any way. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art. Figure 1 Figure 1 A flowchart of a wetland vegetation under-ground topography inversion method based on unmanned aerial vehicle data and machine learning fusion is provided for an embodiment of the present application. As shown in Figure 1 The wetland vegetation under-ground topography inversion method based on unmanned aerial vehicle data and machine learning fusion includes the following steps: S1, acquiring unmanned aerial vehicle multi-spectral images and laser radar point cloud data of a target wetland, combining with RTK measured ground elevation sampling points, and generating a vegetation coverage area multi-dimensional data set after preprocessing; S2, extracting vegetation feature parameters and terrain correlation parameters from the multi-dimensional data set and storing them in a vegetation feature set and a terrain correlation parameter set; S3, based on the vegetation feature set and the terrain correlation parameter set, combining each vegetation feature parameter and terrain correlation parameter with the corresponding RTK measured ground elevation to form a training sample set, constructing a terrain inversion model by using a convolutional neural network, and calculating the mapping relationship between the vegetation feature parameters and terrain correlation parameters and the ground elevation by using the terrain inversion model; S4, calculating the feature correlation degree between the vegetation feature parameters and terrain correlation parameters and the ground elevation in each training sample set, and the cross-modal coupling degree between the parameters, and calculating the terrain inversion confidence according to the feature correlation degree, the cross-modal coupling degree and the parameter weight;​ S5, generating the ground elevation inversion result under the wetland vegetation cover based on the training samples with the confidence exceeding the threshold.

[0023] It should be noted that after obtaining the unmanned aerial vehicle multispectral image and laser radar point cloud data of the target wetland, combined with the RTK measured ground elevation sampling points, a multi-dimensional dataset of the vegetation coverage area is generated by preprocessing, and vegetation feature parameters and terrain correlation parameters are extracted from the dataset and stored in the corresponding set. The preprocessing here refers to a series of operations on the original data to make it more suitable for subsequent analysis. For example, radiation calibration is performed on the multispectral image to generate a standardized vegetation spectral layer; based on the laser radar point cloud data, vegetation canopy height and point cloud density parameters are extracted; through spatial overlay analysis, the standardized vegetation spectral layer, vegetation canopy height parameters and point cloud density parameters are grid-matched with the RTK measured ground elevation to obtain the vegetation feature parameters and terrain correlation parameters of each grid cell. The vegetation feature parameters refer to parameters that can reflect the growth status and structure of the vegetation, including the spectral characteristics of all bands; the terrain correlation parameters include vegetation canopy height and point cloud density, etc., which are used to describe the terrain undulation and morphology. The extraction of these parameters is based on the analysis and processing of the multi-dimensional dataset, and is realized through specific algorithms and models.

[0024] Specifically, the process of extracting vegetation feature parameters and terrain correlation parameters from the multi-dimensional dataset is as follows: the parameters in the vegetation feature set include the spectrum of all bands, which is obtained by radiation calibration of the multispectral image, and can effectively reflect the growth status, plant height and biomass of the vegetation. The parameters in the terrain correlation parameter set include vegetation canopy height and point cloud density, which are extracted by processing the laser radar point cloud data, and can accurately represent the undulating changes of the terrain. The ground elevation is measured based on RTK, which is the actual ground elevation. The extraction of these parameters is based on spatial overlay analysis, which grid-matches the multispectral layer with the ground elevation to obtain the vegetation feature parameters and terrain correlation parameters of each grid cell, providing data basis for the subsequent terrain inversion model construction.

[0025] Preferably, in constructing the terrain inversion model, a convolutional neural network (CNN) is used. CNN is a deep learning model with strong feature extraction and pattern recognition capabilities. In constructing the model, the extracted vegetation feature parameters and terrain correlation parameters are used as input data, and through multiple convolutional layers and pooling layers, the data is extracted and reduced in dimension, and then through a fully connected layer, classification or regression analysis is performed to obtain the mapping relationship between the vegetation feature parameters, terrain correlation parameters and ground elevation. During model training, a large number of training sample sets are used to optimize model parameters and improve model accuracy and generalization ability. Meanwhile, when calculating the feature correlation between the vegetation feature parameters and terrain correlation parameters and the ground elevation in each training sample set, one of the following two methods can be used: After obtaining the first feature vector of the first feature set and the second feature vector of the second feature set using the random forest model, the Pearson correlation coefficient between the first feature vector and the second feature vector is calculated to obtain.

[0026] The first dimension reduction matrix of the first feature set and the second dimension reduction matrix of the second feature set are obtained using principal component analysis and the variance contribution rate of the first feature set and the variance contribution rate of the second feature set, and the maximum covariance value between the first dimension reduction matrix and the second dimension reduction matrix is calculated. The first feature set and the second feature set are obtained by obtaining the vegetation spectrum set of each grid cell in the multi-dimensional data set, storing the vegetation spectrum and terrain correlation parameters of each training sample in the multi-dimensional data set into the first feature set, and storing the ground elevation corresponding to each training sample into the second feature set. The steps of obtaining the first dimension reduction matrix and the second dimension reduction matrix include: standardizing the first feature set and the second feature set, calculating the covariance matrix; performing eigenvalue decomposition on the covariance matrix, and selecting the eigenvectors corresponding to the first k largest eigenvalues as the projection matrix; multiplying the original feature set by the corresponding projection matrix to obtain the first dimension reduction matrix and the second dimension reduction matrix. In addition, when calculating the confidence, the weight of the vegetation feature parameters and the terrain correlation parameters in each training sample, as well as the corresponding feature correlation and cross-modal coupling degree, are obtained by weighted summation. The cross-modal coupling degree between multiple terrain correlation parameters is obtained by calculating the inverse of the variance inflation factor of the second dimension reduction matrix; the weight of the vegetation feature parameters and the terrain correlation parameters in each training sample is obtained by feeding back the ground elevation inversion result corresponding to each training sample to the multi-dimensional data set, calculating the random forest feature importance score of each vegetation feature parameter and terrain correlation parameter in the updated multi-dimensional data set, and obtaining the respective weights.

[0027] In some embodiments, the calculating the feature correlation degree between the vegetation feature parameters and the terrain correlation parameters and the ground elevation in each training sample set comprises: obtaining a vegetation spectrum set of each grid cell in the multi-dimensional data set, storing the vegetation spectrum and the terrain correlation parameters of each training sample in the multi-dimensional data set into a first feature set, storing the ground elevation corresponding to each training sample into a second feature set, and calculating the similarity between the first feature set and the second feature set as the feature correlation degree.

[0028] It should be noted that the calculating the correlation degree between the vegetation feature parameters and the terrain correlation parameters and the ground elevation in each training sample set is realized by obtaining the data set of each grid cell in the multi-dimensional data set. Here, the correlation degree refers to the correlation or similarity degree between the vegetation feature parameters and the terrain correlation parameters and the ground elevation, which is used to measure the mutual relationship between them. The vegetation spectrum set refers to a set containing multiple vegetation spectra, which can reflect the growth status, plant height, biomass and other information of vegetation. By storing the vegetation spectrum and the terrain correlation parameters of each training sample into the first feature set, and storing the ground elevation corresponding to each training sample into the second feature set, a data basis can be provided for calculating the correlation degree. The similarity between the first feature set and the second feature set is calculated to obtain the correlation degree, which is realized by a specific algorithm based on the analysis and processing of the data set.

[0029] Specifically, the first feature set refers to the vegetation spectrum set and the terrain correlation parameters of each training sample, and the vegetation spectra are obtained by processing and analyzing the unmanned aerial vehicle multi-spectral image; the terrain correlation parameters are the elevation values, slopes, and slope directions in the digital elevation model (DSM) obtained by processing the laser radar point cloud data. The second feature set refers to the ground elevation corresponding to each training sample. When calculating the similarity between the first feature set and the second feature set, various methods can be used, such as Pearson correlation coefficient, cosine similarity, etc. These methods compare the numerical relationship between the two feature sets to obtain a similarity value between -1 and 1, where 1 represents complete positive correlation, -1 represents complete negative correlation, and 0 represents no correlation. The calculation result of the similarity can be used as a quantitative indicator of the correlation degree, which is used to evaluate the mutual relationship between the vegetation feature parameters and the terrain correlation parameters and the ground elevation.

[0030] Preferably, when calculating the similarity between the first feature set and the second feature set, a random forest model can be used to obtain the feature vectors. Random forest is an ensemble learning algorithm that builds multiple decision trees to classify or regress data. In this embodiment, the random forest model can be used to extract important features from the first feature set and the second feature set and convert them into feature vectors. Then, the similarity is obtained by calculating the Pearson correlation coefficient between the two feature vectors. The Pearson correlation coefficient is calculated by comparing the numerical trend of the two feature vectors, specifically by calculating the ratio of the product of the covariance of the two vectors and the standard deviation of each. This method can effectively handle high-dimensional data and take into account the relationship between features, resulting in more accurate similarity values and improving the accuracy of the correlation calculation.

[0031] In some embodiments, the calculation of the similarity between the first feature set and the second feature set is obtained by calculating the Pearson correlation coefficient between the first feature vector and the second feature vector after obtaining the first feature vector of the first feature set and the second feature vector of the second feature set using the random forest model.

[0032] It should be noted that the calculation of the feature correlation between the vegetation feature parameters and the terrain correlation parameters and the ground elevation in each training sample set is achieved by obtaining the data set of each grid cell in the multi-dimensional data set. The feature correlation here refers to the correlation or similarity between the vegetation feature parameters and the terrain correlation parameters and the ground elevation, which is used to measure the mutual relationship between them. The vegetation spectrum set refers to a set of multiple vegetation spectra that can reflect the growth status, height, biomass, and other information of the vegetation. By storing the vegetation spectrum and terrain correlation parameters of each training sample in the first feature set and the ground elevation corresponding to each training sample in the second feature set, data basis can be provided for calculating the feature correlation. The similarity between the first feature set and the second feature set is calculated to obtain the feature correlation, which is based on the analysis and processing of the data set and achieved through a specific algorithm.

[0033] Specifically, the first feature set refers to a set of vegetation spectra and terrain-related parameters for each training sample. The vegetation spectra are obtained by processing and analyzing the multi-spectral images of the unmanned aerial vehicle, such as radiation calibration. The terrain-related parameters are obtained by processing the laser radar point cloud data, such as vegetation canopy height and point cloud density. The second feature set refers to the corresponding ground elevation of each training sample, which is obtained based on RTK measurement. When calculating the similarity between the first and second feature sets, various methods can be used, such as Pearson correlation coefficient and cosine similarity. These methods compare the numerical relationship between the two feature sets to obtain a similarity value between -1 and 1, where 1 represents complete positive correlation, -1 represents complete negative correlation, and 0 represents no correlation. The calculation result of the similarity can be used as a quantitative indicator of the feature correlation degree to evaluate the mutual relationship between the vegetation feature parameters and the terrain-related parameters and the ground elevation.

[0034] Preferably, when calculating the similarity between the first and second feature sets, a random forest model can be used to obtain the feature vector. Random forest is an ensemble learning algorithm that builds multiple decision trees to classify or regress data. In this embodiment, the random forest model can be used to extract important features from the first and second feature sets and convert them into a feature vector. Then, the similarity is obtained by calculating the Pearson correlation coefficient between the two feature vectors. The Pearson correlation coefficient is calculated by comparing the numerical trend of the two feature vectors, specifically by calculating the ratio of the product of the covariance and the standard deviation of each vector. This method can effectively handle high-dimensional data and consider the mutual relationship between features, resulting in more accurate similarity values and improving the accuracy of feature correlation degree calculation.

[0035] In some embodiments, the similarity between the first and second feature sets is calculated using principal component analysis, and the variance contribution rate of the first feature set and the variance contribution rate of the second feature set are used to obtain a first dimension reduction matrix of the first feature set and a second dimension reduction matrix of the second feature set, respectively. The maximum covariance value between the first and second dimension reduction matrices is calculated.

[0036] It should be noted that the similarity between the first feature set and the second feature set is calculated by using principal component analysis (PCA), and the variance contribution rate of the first feature set and the variance contribution rate of the second feature set, respectively, to obtain the first dimension reduction matrix of the first feature set and the second dimension reduction matrix of the second feature set, and then calculate the maximum covariance value between the two dimension reduction matrices. Principal component analysis is a commonly used dimension reduction technique, which can convert high-dimensional data into low-dimensional data while preserving the main information in the data. The variance contribution rate refers to the proportion of the variance explained by each principal component to the total variance, which reflects the importance of the principal component in the data. Through principal component analysis, the main features of the data can be extracted, thereby simplifying the calculation process and improving the efficiency of similarity calculation.

[0037] Specifically, the first feature set refers to a set of vegetation spectra and terrain-related parameters. The vegetation spectra include reflectance of all bands, which can reflect the growth status, height and biomass of vegetation. The terrain-related parameters include elevation, slope and aspect, which can describe the relief and morphology of the terrain. The second feature set refers to the ground elevation measured by RTK. Principal component analysis calculates the covariance matrix of the data and performs eigenvalue decomposition to obtain the principal components and the corresponding variance contribution rate. The first dimension reduction matrix and the second dimension reduction matrix are obtained by projecting the original feature set into the principal component space, which preserves the main information of the original data. The maximum covariance value refers to the maximum covariance between the two dimension reduction matrices, which reflects the correlation between the two matrices. By calculating the maximum covariance value, the similarity between the first feature set and the second feature set can be quantified, thereby providing an important reference for the terrain inversion model.

[0038] Preferably, when performing principal component analysis, the first feature set and the second feature set need to be standardized first to eliminate the dimensional difference between different features. Then, the covariance matrix of each feature set is calculated and eigenvalue decomposition is performed to obtain the principal components and the corresponding variance contribution rate. The first k principal components with higher variance contribution rate are selected as the projection matrix, and the original feature set is projected onto these principal components to obtain the first dimension reduction matrix and the second dimension reduction matrix. When calculating the maximum covariance value, the covariance matrix between the two dimension reduction matrices can be calculated, and then the maximum value is selected. This method not only effectively handles high-dimensional data, but also preserves the main information in the data, thereby improving the accuracy and efficiency of similarity calculation.

[0039] In some embodiments, the first dimension reduction matrix of the first feature set and the second dimension reduction matrix of the second feature set are obtained by the following steps: Standardizing the first feature set and the second feature set to calculate the covariance matrix; Eigenvalue decomposition is performed on the covariance matrix, and the eigenvectors corresponding to the first k largest eigenvalues are selected as the projection matrix; The original feature set is multiplied by the corresponding projection matrix to obtain the first and second reduced dimension matrices.

[0040] It should be noted that the process of obtaining the first reduced dimension matrix of the first feature set and the second reduced dimension matrix of the second feature set includes standardizing the feature set, calculating the covariance matrix, performing eigen decomposition, and projecting into the principal component space. This process is to reduce the dimension of high-dimensional data while retaining the main information in the data, so as to more efficiently perform subsequent similarity calculation. Standardization refers to converting the data to a distribution with a mean of zero and a standard deviation of one, so as to eliminate the dimensional differences between different features. The covariance matrix reflects the correlation between different features in the data, and the eigen decomposition is to extract the main components of the data, i.e. those components that can explain the most variance of the data.

[0041] Specifically, the first feature set refers to a set containing vegetation spectra and terrain-related parameters. The vegetation spectra include reflectance of all bands, which can reflect the growth status, height and biomass of vegetation. The terrain-related parameters include elevation, slope and aspect, which can describe the relief and morphology of the terrain. The second feature set refers to the ground elevation measured by RTK. The standardization is achieved by subtracting the mean of each feature value and dividing by its standard deviation, so as to ensure that the mean of each feature is zero and the standard deviation is one. The calculation of the covariance matrix is obtained by multiplying the standardized data matrix with its transpose matrix, which reflects the linear relationship between different features. The eigen decomposition is achieved by solving the eigenvalues and eigenvectors of the covariance matrix, the eigenvalues represent the variance contribution rate of each principal component, and the eigenvectors represent the direction of data in the principal component space. The eigenvectors corresponding to the first k largest eigenvalues are selected as the projection matrix, and the original feature set is projected onto these principal components to obtain the reduced dimension matrix.

[0042] Preferably, when performing eigen decomposition, methods such as singular value decomposition (SVD) or eigenvalue decomposition (EVD) can be used to extract principal components. Singular value decomposition is a more stable numerical method suitable for processing large data sets. When selecting the projection matrix, the value of k can be determined according to the cumulative variance contribution rate, and usually the principal components with a cumulative variance contribution rate of 85% to 95% are selected to ensure that the reduced dimension data can retain most of the original information. In the projection process, each data point of the original feature set is multiplied by the projection matrix to obtain the reduced dimension data point, forming the first and second reduced dimension matrices. This process not only effectively reduces the dimension of the data, but also retains the main information in the data, providing a more efficient data structure for subsequent similarity calculation.

[0043] In some embodiments, the coupling degree between the plurality of terrain-related parameters is obtained by calculating the inverse of the variance inflation factor of the second reduced dimension matrix.

[0044] It should be noted that the coupling degree between multiple terrain-related parameters is obtained by calculating the reciprocal of the variance inflation factor VIF of the second reduced dimension matrix. The coupling degree refers to the degree of mutual correlation and mutual influence between multiple terrain-related parameters, and is used to evaluate the complex relationship between terrain-related parameters. The variance inflation factor is a statistical index used to measure the influence of multicollinearity on regression analysis. By calculating the reciprocal of the variance inflation factor, the coupling degree between terrain-related parameters can be quantified, thereby providing more accurate parameter relationship evaluation for the terrain inversion model.

[0045] Specifically, the second reduced dimension matrix refers to the terrain-related parameter matrix after principal component analysis dimension reduction. It contains the main information of the terrain-related parameters, and after dimension reduction, the coupling degree can be calculated more effectively. The variance inflation factor VIF is obtained by calculating the variance inflation degree of each terrain-related parameter in the regression model. The calculation of VIF is based on regression analysis, by taking each terrain-related parameter as the dependent variable and other terrain-related parameters as the independent variables for regression analysis, the VIF value of each terrain-related parameter is obtained. The larger the VIF value, the stronger the multicollinearity between the terrain-related parameter and other terrain-related parameters, and the higher the coupling degree. By calculating the reciprocal of VIF, the coupling degree can be obtained, and the smaller the value, the higher the coupling degree.

[0046] Preferably, when calculating the variance inflation factor VIF, the stepwise regression analysis method can be used. First, select one terrain-related parameter as the dependent variable and other terrain-related parameters as the independent variables to construct a regression model, and calculate the VIF value of the dependent variable. Then, the same operation is performed on each terrain-related parameter to obtain the VIF value of all terrain-related parameters. During the calculation process, a threshold value can be set, for example, when the VIF value is greater than 10, it is considered that there is serious multicollinearity in the terrain-related parameter. By calculating the VIF value of each terrain-related parameter and taking its reciprocal, the coupling degree is obtained. This method can effectively evaluate the mutual relationship between terrain-related parameters and provide an important reference for the construction of terrain inversion model.

[0047] In some embodiments, the weight of each vegetation feature parameter and terrain-related parameter in each training sample is obtained by feeding back the ground elevation inversion result corresponding to each training sample to the multidimensional dataset, and calculating the random forest feature importance score of each vegetation feature parameter and terrain-related parameter in the updated multidimensional dataset.

[0048] It should be noted that the weight of each vegetation feature parameter and terrain correlation parameter in each training sample is obtained by feeding back the terrain correlation parameter inversion result corresponding to each training sample to the multidimensional dataset, and calculating the random forest feature importance score of each vegetation feature parameter and terrain correlation parameter in the updated multidimensional dataset. The weight here refers to the importance degree of each vegetation feature parameter and terrain correlation parameter in the model, which is used to measure their contribution to the terrain inversion result. The random forest feature importance score is a method for measuring the importance of features in the model, which determines the weight by evaluating the impact of features on the performance of the model. This process is to optimize the terrain inversion model to ensure that the model can more accurately reflect the relationship between vegetation features and terrain correlation parameters.

[0049] Specifically, the training sample refers to the data pair extracted from the vegetation feature set and the terrain correlation parameter set, which is used to train the terrain inversion model. The terrain correlation parameter inversion result refers to the terrain correlation parameter value predicted by the model. Feeding back these inversion results to the multidimensional dataset can update the terrain correlation parameter values in the dataset, thereby more accurately reflecting the actual terrain conditions. The random forest feature importance score is calculated by evaluating the contribution of each feature in the random forest model. In the random forest, different feature subsets are used in the construction process of each decision tree, and the feature importance score is obtained by calculating the average impurity reduction of each feature in all decision trees. These scores reflect the importance of each feature in the model, i.e., their contribution to the model prediction result.

[0050] Preferably, when calculating the random forest feature importance score, a permutation-based method can be used. First, a random forest model is trained, and then each feature is permuted to calculate the change in model performance after permutation. If the model performance significantly decreases after permuting a certain feature, it means that the feature is important to the model, and its importance score is high. The specific steps include: training a random forest model, calculating the initial performance indicators of the model such as mean square error or classification accuracy; permuting each feature and recalculating the model performance indicators; comparing the difference between the performance indicators before and after permutation to obtain the importance score of each feature. In this way, a weight can be assigned to each vegetation feature parameter and terrain correlation parameter, which will be used for subsequent confidence calculation, thereby improving the accuracy and reliability of the terrain inversion result.

[0051] In some embodiments, the confidence of each training sample is obtained by weighting and summing the weights of the vegetation feature parameters and terrain correlation parameters in each training sample after normalization of the corresponding correlation degree and coupling degree, respectively.

[0052] It should be noted that the confidence of each training sample is obtained by weighting and summing the weights of the vegetation feature parameters and the terrain correlation parameters in each training sample, and the corresponding correlation degree and coupling degree after normalization. The confidence here refers to the reliability evaluation of the model's prediction results for the training samples, which is used to measure the accuracy and credibility of the model's prediction. Normalization refers to converting data to a unified range, usually 0 to 1, for comparison and calculation. Weighted summation refers to summing up the weights according to the importance of each factor to evaluate the confidence of the training sample. This process is to ensure that the model can more accurately evaluate the reliability of each training sample, thereby improving the overall quality of the terrain inversion results.

[0053] Specifically, the weights of the vegetation feature parameters and the terrain correlation parameters refer to the importance of each parameter in the model obtained by random forest feature importance score. The correlation degree refers to the similarity or correlation between the vegetation feature parameters and the terrain correlation parameters and the ground elevation, reflecting the mutual relationship between them. The coupling degree refers to the degree of correlation between multiple terrain correlation parameters, which is used to evaluate the complex relationship between terrain correlation parameters. Normalization is achieved by subtracting the minimum value of each parameter from its value and dividing it by the difference between the maximum and minimum values. This way, all parameter values can be converted to the range of 0 to 1. Weighted summation is to add the normalized weights, correlation degrees and coupling degrees according to the weights of each parameter to obtain the confidence of each training sample. The closer this confidence value is to 1, the more reliable the model's prediction of the training sample is.

[0054] Preferably, when calculating the confidence, different weight coefficients can be assigned to the weights, correlation degrees and coupling degrees, which can be adjusted according to actual problems and experience. For example, if the vegetation feature parameters have a greater impact on terrain inversion, a higher weight coefficient can be assigned to the weights. The specific steps include: first, normalizing the weights, correlation degrees and coupling degrees of the vegetation feature parameters and the terrain correlation parameters; then, assigning weight coefficients according to the importance of each factor; finally, multiplying the normalized values by the corresponding weight coefficients and adding them up to obtain the confidence of each training sample. In this way, the influence of each factor on the confidence can be adjusted more flexibly, thereby more accurately evaluating the reliability of each training sample and providing an important basis for optimizing the terrain inversion model.

[0055] In some embodiments, the vegetation feature parameters and the terrain correlation parameters are extracted from the multi-dimensional data set and stored in the vegetation feature set and the terrain correlation parameter set, including: Radiometric calibration of the unmanned aerial vehicle multi-spectral image to generate a standardized vegetation spectral layer; Based on the laser radar point cloud data, the vegetation canopy height and point cloud density parameters are extracted; The normalized vegetation spectral layer, the vegetation canopy height parameter, and the point cloud density parameter are grid-matched with the RTK measured ground elevation through spatial overlay analysis to obtain the vegetation feature parameters and the terrain correlation parameters of each grid unit.

[0056] It should be noted that the process of extracting vegetation feature parameters and terrain correlation parameters from multi-dimensional data sets and storing them in corresponding sets includes radiometric calibration of unmanned aerial vehicle multi-spectral images to generate a normalized vegetation spectral layer; based on laser radar point cloud data, extracting vegetation canopy height and point cloud density parameters; through spatial overlay analysis, the normalized vegetation spectral layer, the vegetation canopy height parameter, and the point cloud density parameter are grid-matched with the RTK measured ground elevation to obtain the vegetation feature parameters and the terrain correlation parameters of each grid unit. This process is to integrate data from different sources to provide comprehensive and accurate data support for subsequent terrain inversion models. The vegetation spectral layer can reflect the growth status of the vegetation, the terrain correlation parameters can reflect the vegetation canopy structure and elevation, and the ground elevation provides detailed information of the terrain, and the combination of the three can more accurately describe the terrain characteristics under the wetland vegetation cover.

[0057] Specifically, the vegetation spectral information is calculated by radiometric calibration of unmanned aerial vehicle multi-spectral images, and the spectral characteristics of each band can effectively reflect the biomass, height, and health status of the vegetation. Laser radar point cloud data can extract point cloud data of the vegetation canopy through filtering and noise reduction processing to obtain parameters such as vegetation canopy structure, canopy elevation, and point cloud density. Spatial overlay analysis refers to aligning and merging different spatial data layers for analysis in the same spatial framework. In this embodiment, by grid-matching the normalized vegetation spectral layer, the vegetation canopy height parameter, and the point cloud density parameter with the RTK measured ground elevation, the vegetation feature parameters and the terrain correlation parameters of each grid unit can be obtained to provide a data basis for subsequent terrain inversion.

[0058] Preferably, when calculating the vegetation spectrum, the given radiometric calibration coefficient generated by the unmanned aerial vehicle is used for calculation. In the ground filtering processing of the laser radar point cloud data, a filtering algorithm based on terrain slope is used, which calculates the local slope of the point cloud data to obtain the three-dimensional structure and elevation of the non-ground points, thereby improving the accuracy of the ground elevation. In the spatial overlay analysis, geographic information system (GIS) software is used for operation, and by setting a grid resolution of 1 meter x 1 meter, it is ensured that the vegetation feature parameters and the terrain correlation parameters of each grid unit can accurately reflect the actual wetland conditions. Through these detailed operation steps, the extracted vegetation feature parameters and terrain correlation parameters can be more accurate and reliable, providing high-quality data support for terrain inversion under wetland vegetation.

[0059] The above various embodiments of the present application have the following beneficial effects: the present application can improve the precision and efficiency of terrain inversion in wetland vegetation coverage area. By fusing unmanned aerial vehicle multi-spectral image and laser radar point cloud data, a high-precision multi-dimensional data set can be constructed, and a convolutional neural network is used to establish the mapping relationship between vegetation features and terrain correlation parameters, effectively solving the terrain inversion problem caused by vegetation shielding. By calculating the correlation degree of vegetation features and terrain correlation parameters, the coupling degree between terrain correlation parameters, and the sample confidence, high-quality training samples can be intelligently selected, thereby improving the reliability of the inversion result.

[0060] In addition, the method can flexibly use machine learning algorithms such as random forest and principal component analysis to optimize the feature extraction and dimension reduction process, reduce data redundancy, and enhance the model generalization ability. By dynamically adjusting the weights of vegetation features and terrain correlation parameters, different wetland environments can be adapted, and the applicability of the inversion model can be improved. The finally generated high-precision terrain inversion result can provide reliable data support for wetland ecological monitoring, hydrological simulation and environmental protection.

[0061] As shown in Figure 2 Fig. 1, one embodiment of a wetland vegetation under terrain inversion system based on unmanned aerial vehicle data and machine learning fusion, the system comprises: A data acquisition module 201 is configured to acquire unmanned aerial vehicle multi-spectral image, laser radar point cloud data and RTK measured ground elevation data of a target wetland, and generate a vegetation coverage multi-dimensional data set after preprocessing; A feature extraction module 202 is configured to extract vegetation feature parameters and terrain correlation parameters from the multi-dimensional data set and store them in a vegetation feature set and a terrain correlation parameter set; A model training module 203 is configured to construct a terrain inversion model based on the vegetation feature set and the terrain correlation parameter set, and calculate the mapping relationship between the vegetation feature parameters, the terrain correlation parameters and the ground elevation; A confidence calculation module 204 is configured to calculate the feature correlation degree, the cross-modal coupling degree and the parameter weight of the training sample, and generate a terrain inversion confidence evaluation result; A terrain inversion module 205 is configured to generate a ground elevation inversion result under wetland vegetation coverage according to the training sample with confidence up to standard.

[0062] It can be understood that the modules described in the wetland vegetation under terrain inversion system based on unmanned aerial vehicle data and machine learning fusion are described in the Figure 1The steps in the described method of wetland vegetation under topography inversion based on unmanned aerial vehicle data and machine learning fusion correspond to each other. Therefore, the operations, features and beneficial effects described above for the method of wetland vegetation under topography inversion based on unmanned aerial vehicle data and machine learning fusion also apply to the system of wetland vegetation under topography inversion based on unmanned aerial vehicle data and machine learning fusion and the modules contained therein, which will not be repeated here.

[0063] Reference is made below Figure 3 to a block diagram showing the structure of an electronic device 300 suitable for use in implementing some embodiments of the present application. The electronic device in some embodiments of the present application can include, but is not limited to, a mobile terminal such as a mobile phone, a notebook computer, a digital broadcast receiver, a PDA (Personal Digital Assistant), a PAD (Tablet PC), a PMP (Portable Multimedia Player), a car terminal (e.g., a car navigation terminal), and the like, as well as a stationary terminal such as a digital TV, a desktop computer, and the like. Figure 3 The terminal device shown is merely an example and should not impose any limitation on the functions and scope of use of embodiments of the present application.

[0064] As shown in Figure 3 , the electronic device 300 can include a processing means (e.g., a central processor, a graphic processor, etc.) 301 that can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 302 or loaded into a random access memory (RAM) 303 from a storage means 308. Various programs and data required for the operation of the electronic device 300 are also stored in the RAM 303. The processing means 301, the ROM 302, and the RAM 303 are connected to each other through a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0065] In general, the following means can be connected to the I / O interface 305: an input means 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, and the like; an output means 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, and the like; a storage means 308 including, for example, a magnetic tape, a hard disk, and the like; and a communication means 309. The communication means 309 can allow the electronic device 300 to communicate wirelessly or by wire with other devices to exchange data. Although Figure 3 The electronic device 300 is shown with various means, but it is understood that all of the means shown are not required to be implemented or present. More or fewer means can alternatively be implemented or present. Figure 3 Each block shown in the diagram can represent one means or, as needed, multiple means.

[0066] Further, the storage medium of the embodiments of the present application stores program instructions capable of realizing all the methods described above, wherein the program instructions can be stored in the storage medium in the form of a software product, and include a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor execute all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media capable of storing program codes, or a computer, a server, a mobile phone, a tablet, and other terminal devices.

[0067] The above description is merely some preferred embodiments of the present application and a description of the principles of the technology used. Those skilled in the art should understand that the scope of the application involved in the embodiments of the present application is not limited to the technical solutions formed by the specific combinations of the technical features described above, and should also cover other technical solutions formed by any combinations of the technical features described above or their equivalent features without departing from the inventive concept described above. For example, the technical solutions formed by mutually replacing the above-described features and the technical features disclosed in the embodiments of the present application (but not limited to) having similar functions.

Claims

1. A method for wetland vegetation topography inversion based on the fusion of UAV data and machine learning, characterized in that, Includes the following steps: Acquire UAV multispectral imagery and lidar point cloud data of the target wetland area, combine them with RTK measured ground elevation sampling points, and generate a multidimensional dataset of vegetation cover area after preprocessing. Extract vegetation feature parameters and terrain association parameters from the multidimensional dataset and store them in the vegetation feature set and terrain association parameter set; Based on the vegetation feature set and the terrain association parameter set, each vegetation feature parameter and terrain association parameter is combined with the corresponding RTK measured ground elevation to form a training sample set, and a terrain inversion model is constructed using a convolutional neural network. The mapping relationship between vegetation characteristic parameters, terrain-related parameters, and elevation is calculated using the terrain inversion model. Calculate the feature correlation degree between vegetation feature parameters and terrain correlation parameters and ground elevation in each training sample set, as well as the cross-modal coupling degree between each parameter; The confidence level of terrain inversion is calculated based on feature correlation, cross-modal coupling, and parameter weights. Ground elevation inversion results under wetland vegetation cover are generated based on training samples with confidence levels exceeding a threshold.

2. The wetland vegetation under-vegetation topography inversion method based on the fusion of UAV data and machine learning according to claim 1, characterized in that, The calculation of the feature correlation degree between vegetation feature parameters and terrain correlation parameters and ground elevation in each training sample set includes: Obtain the vegetation spectrum set of each grid cell in the multidimensional dataset, store the vegetation spectrum and terrain association parameters of each training sample in the multidimensional dataset into the first feature set, and store the ground elevation corresponding to each training sample into the second feature set. Calculate the similarity between the first feature set and the second feature set as the feature association degree.

3. The wetland vegetation under-vegetation topography inversion method based on the fusion of UAV data and machine learning according to claim 2, characterized in that, The similarity between the first feature set and the second feature set is calculated by using a random forest model to obtain the first feature vector of the first feature set and the second feature vector of the second feature set, and then calculating the Pearson correlation coefficient between the first feature vector and the second feature vector.

4. The wetland vegetation under-vegetation topography inversion method based on the fusion of UAV data and machine learning according to claim 2, characterized in that, The similarity between the first feature set and the second feature set is calculated by using principal component analysis, the variance contribution rate of the first feature set and the variance contribution rate of the second feature set, to obtain the first dimensionality reduction matrix of the first feature set and the second dimensionality reduction matrix of the second feature set, and then calculating the maximum covariance value between the first dimensionality reduction matrix and the second dimensionality reduction matrix.

5. The wetland vegetation under-vegetation topography inversion method based on the fusion of UAV data and machine learning according to claim 4, characterized in that, The first dimensionality reduction matrix of the first feature set and the second dimensionality reduction matrix of the second feature set are obtained through the following steps: Standardize the first and second feature sets and calculate the covariance matrix; Perform eigenvalue decomposition on the covariance matrix and select the eigenvectors corresponding to the k largest eigenvalues ​​as the projection matrix; Multiply the original feature set by the corresponding projection matrix to obtain the first and second dimension reduction matrices.

6. The wetland vegetation under-vegetation topography inversion method based on the fusion of UAV data and machine learning according to claim 4, characterized in that, The cross-modal coupling degree between the multiple terrain-related parameters is obtained by calculating the reciprocal of the variance inflation factor of the second dimensionality reduction matrix.

7. The wetland vegetation under-vegetation topography inversion method based on the fusion of UAV data and machine learning according to claim 4, characterized in that, The weights of vegetation feature parameters and terrain association parameters in each training sample are obtained by feeding back the ground elevation inversion results corresponding to each training sample to the multidimensional dataset, and calculating the random forest feature importance score of each vegetation feature parameter and terrain association parameter in the updated multidimensional dataset.

8. The wetland vegetation under-vegetation topography inversion method based on the fusion of UAV data and machine learning according to claim 1, characterized in that, The confidence level of each training sample is obtained by normalizing the weights of the vegetation feature parameters and terrain association parameters in each training sample, as well as the corresponding feature association degree and cross-modal coupling degree, and then summing them by weight.

9. The wetland vegetation under-vegetation topography inversion method based on the fusion of UAV data and machine learning according to claim 1, characterized in that, Vegetation feature parameters and terrain association parameters are extracted from the multidimensional dataset and stored in a vegetation feature set and a terrain association parameter set, including: Radiometric calibration of UAV multispectral imagery was performed to generate a standardized vegetation spectral layer. Based on lidar point cloud data, vegetation canopy height and point cloud density parameters are extracted; By using spatial overlay analysis, the standardized vegetation spectral layer, vegetation canopy height parameters, and point cloud density parameters are matched with the RTK measured ground elevation in a grid to obtain the vegetation characteristic parameters and terrain association parameters of each grid cell.

10. A wetland vegetation under-vegetation topography inversion system based on the fusion of UAV data and machine learning, characterized in that, include: The data acquisition module is used to acquire UAV multispectral images, lidar point cloud data and RTK measured ground elevation data of the target wetland area, and generate a multidimensional dataset of vegetation cover area after preprocessing. The feature extraction module is used to extract vegetation feature parameters and terrain association parameters from the multidimensional dataset and store them in the vegetation feature set and the terrain association parameter set. The model training module is used to construct a terrain inversion model based on the vegetation feature set and the terrain association parameter set, and to calculate the mapping relationship between the vegetation feature parameters, the terrain association parameters and the ground elevation. The confidence calculation module is used to calculate the feature correlation, cross-modal coupling, and parameter weights of training samples, and generate the terrain inversion confidence assessment results. The terrain inversion module is used to generate ground elevation inversion results under wetland vegetation cover based on training samples that meet the confidence level.

Images