Investigation method for Danxia landform

Abstract

The invention discloses a Danxia landform investigation method. The method comprises the following steps: acquiring and processing a satellite remote sensing image and elevation data of a Danxia landform in a preset area; based on the processed satellite remote sensing image, obtaining landform features of the Danxia landform; based on the processed elevation data, obtaining topographic relief features of the Danxia landform; based on the landform features and the topographic relief features, constructing a Danxia landform recognition model, and obtaining a recognition result; and based on the identification result, combining field investigation and analysis of a formation mechanism of the Danxia landform in the preset area, and completing investigation of the Danxia landform in the preset area. Application and popularization of the Danxia landform investigation method based on satellite remote sensing images and elevation data can promote theoretical research and technical innovation of related subjects and inject new vitality for development of geology and geomorphology.

Description

Technical Field

[0001] The invention belongs to the technical field of landform investigation, and in particular relates to a method for investigating Danxia landform. Background Art

[0002] Remote sensing technology is an indispensable tool in modern geological surveys. Through satellite or aerial remote sensing images, large-scale surface information can be quickly obtained, including landform morphology, vegetation coverage, water distribution, etc. In the investigation of Danxia landforms, remote sensing images are used to identify the distribution range, morphological characteristics and boundaries of Danxia landforms with other landform types. High-resolution remote sensing images can also reveal the detailed characteristics of Danxia landforms, such as the occurrence of rock layers and the development of joints.

[0003] Data analysis and modeling are key steps in the investigation of Danxia landforms. Through comprehensive analysis of remote sensing images, elevation data, ground survey data, etc., the formation mechanism, development process and relationship between Danxia landforms and other geological phenomena can be revealed.

[0004] The existing technology has low data processing accuracy and poor model robustness, and there is still room for improvement in the type identification and cause analysis of Danxia landforms. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method for investigating Danxia landforms, which improves the investigation efficiency and provides valuable reference and scientific basis for related disciplines such as geology and geography.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A method for investigating Danxia landforms comprises the following steps:

[0008] Collect and process satellite remote sensing images and elevation data of Danxia landforms in the preset area;

[0009] Based on the processed satellite remote sensing image, the geomorphic features of the Danxia landform are obtained;

[0010] Based on the processed elevation data, the topographical undulation characteristics of Danxia landform are obtained;

[0011] Based on the geomorphic features and the terrain undulation features, a Danxia landform recognition model is constructed to obtain a recognition result;

[0012] Based on the identification results, combined with field research, the formation mechanism of the Danxia landform in the preset area is analyzed to complete the investigation of the Danxia landform in the preset area.

[0013] Preferably, the method for processing the satellite remote sensing image includes radiation calibration, atmospheric correction and clipping, wherein the atmospheric correction method is:

[0014] Based on the MODIS surface reflectance product, the ratio factor map of Danxia landform is obtained;

[0015] Performing aerosol inversion based on the ratio factor map to obtain an aerosol map;

[0016] Obtain atmospheric parameters when collecting satellite remote sensing images, and invert water vapor information based on the water vapor detection band to obtain high-resolution water vapor data;

[0017] Based on the aerosol map and the high-resolution water vapor data, atmospheric correction is performed on the satellite remote sensing image.

[0018] Preferably, the method for obtaining the geomorphic features of Danxia landform is:

[0019] Using principal component analysis, the preprocessed satellite remote sensing image is subjected to band analysis to obtain a covariance matrix of each band, and the eigenvalue of the covariance matrix is ​​calculated;

[0020] Selecting principal components based on the eigenvalues ​​to obtain satellite remote sensing images after dimensionality reduction;

[0021] Decompose the satellite remote sensing image after dimension reduction to obtain image blocks with central pixels as labels;

[0022] A three-dimensional convolutional neural network is used to extract features from the image blocks to obtain geomorphic features of the Danxia landform; wherein the geomorphic features include spectral information, color information, texture information, geometric information and mineral composition information.

[0023] Preferably, the method for processing the elevation data is:

[0024] Processing the elevation data based on a local enhancement mechanism to obtain an improved elevation data sequence;

[0025] Based on the improved elevation data sequence and the Butterworth filter coefficients, smooth elevation data is obtained to complete the processing of the elevation data.

[0026] Preferably, the method for obtaining the topographic relief characteristics of Danxia landform is:

[0027] Based on the gradient of the processed elevation data in two dimensions, the slope information of Danxia landform is obtained;

[0028] Based on the coefficient of variation of the processed elevation data, the terrain roughness of the Danxia landform is obtained;

[0029] Based on the slope information and the terrain roughness, the terrain undulation characteristics of the Danxia landform are obtained.

[0030] Preferably, based on the geomorphic features and the terrain undulation features, the method for constructing a Danxia landform recognition model is:

[0031] Performing element-by-element product fusion on the geomorphic features and the terrain relief features to obtain initial fusion features;

[0032] Based on the channel dimension of the initial fusion feature, channel extraction is performed;

[0033] Based on the fully connected layer convolution kernel of the neural network and the preset activation function, the channel relationship of the initial fusion feature is calculated;

[0034] Performing element-wise multiplication of the channel extraction result and the channel relationship to obtain a reconstructed feature;

[0035] Based on the reconstructed features and the initial fused features, obtaining final fused features;

[0036] Based on the final fusion features, the landform features and the terrain undulation features, decision fusion is performed to complete the construction of the Danxia landform recognition model.

[0037] Preferably, the recognition results include mountain peaks, cliffs, canyons, caves and landslides.

[0038] Preferably, the method for analyzing the formation mechanism of Danxia landform in a preset area is:

[0039] Based on the recognition results, the distribution range and area size of Danxia landforms in a preset area are counted to obtain the spatial distribution characteristics of Danxia landforms;

[0040] Based on the spatial distribution characteristics, the corresponding genetic mechanism parameters of Danxia landform are collected; wherein the genetic mechanism parameters include geological background, climate factors and hydrological conditions;

[0041] Calculate the correlation between each genetic mechanism parameter and Danxia landform type to obtain the correlation sequence;

[0042] Based on the correlation sequence, the significant causal parameters of Danxia landform are obtained, and the analysis of the causal mechanism of Danxia landform in the preset area is completed.

[0043] Compared with the prior art, the beneficial effects of the present invention are: collecting and processing satellite remote sensing images and elevation data of Danxia landforms in a preset area; the present invention comprehensively and accurately obtains information on Danxia landforms. Based on the processed satellite remote sensing images, the geomorphic features of Danxia landforms are obtained; based on the processed elevation data, the terrain undulation features of Danxia landforms are obtained; remote sensing technology and GIS software are used to automatically process images and elevation data, which reduces manual intervention and improves the speed and accuracy of data processing. Based on the geomorphic features and terrain undulation features, a Danxia landform recognition model is constructed to obtain recognition results; the Danxia landform recognition model constructed based on the processed data can more accurately identify the characteristics and distribution range of Danxia landforms, avoiding human errors and subjective judgments in traditional survey methods. Based on the recognition results, the causal mechanism of Danxia landforms in the preset area is analyzed, and the investigation of Danxia landforms in the preset area is completed. The analysis results of the causal mechanism can provide scientific basis and guidance for geological research, ecological protection, tourism development and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0045] Figure 1 The present invention is a flowchart of a method for investigating Danxia landforms according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] 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.

[0047] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0048] Embodiment 1

[0049] like Figure 1 As shown, a method for investigating Danxia landform comprises the following steps:

[0050] S1: Collect and process satellite remote sensing images and elevation data of Danxia landforms in the preset area;

[0051] A further implementation method is that the method for processing satellite remote sensing images includes radiometric calibration, atmospheric correction and clipping, wherein the atmospheric correction method is:

[0052] Based on the MODIS surface reflectance product, a ratio factor map of Danxia landform is obtained; in this embodiment, the red light band and the short-wave infrared band are selected for ratio calculation to obtain a ratio map; based on a preset threshold, the ratio map is divided into Danxia landform area and other areas to obtain a ratio factor map.

[0053] Perform aerosol inversion based on the ratio factor map to obtain an aerosol map;

[0054] In this embodiment, an aerosol model is established based on the aerosol type and concentration characteristics of a preset area. The information in the ratio factor map is used as input, and combined with the collected meteorological and surface parameters, the dark pixel and deep blue combination algorithm is used to invert the aerosol optical thickness to obtain an aerosol optical thickness map, that is, an aerosol map.

[0055] The atmospheric parameters when collecting satellite remote sensing images are obtained, and water vapor information is inverted based on the water vapor detection band to obtain high-resolution water vapor data; in this embodiment, the near-infrared water vapor absorption band is selected to use the radiation transmission model to perform water vapor inversion to obtain water vapor content information in the atmosphere; based on the water vapor content information, a high-resolution water vapor content distribution map is generated.

[0056] Based on the aerosol map and high-resolution water vapor data, the satellite remote sensing image is atmospherically corrected. In this embodiment, the aerosol map and the water vapor content distribution map are used to construct an atmospheric correction model; based on the atmospheric correction model, the original satellite remote sensing image is corrected to remove the influence of the atmosphere on the surface reflectivity. The atmospherically corrected satellite remote sensing images are output, which are closer to the real surface features in color and texture, and can be used for subsequent geological, ecological, and environmental monitoring research.

[0057] A further implementation method is that the method for processing the elevation data is:

[0058] The elevation data is processed based on the local enhancement mechanism to obtain an improved elevation data sequence;

[0059] In this embodiment, the acquired elevation data is segmented according to the characteristics of the preset area, and the local extreme gradient value and the corresponding elevation data of each segment are determined based on the local enhancement mechanism;

[0060] Based on the local extreme value gradient, the absolute value of the local extreme value gradient and the corresponding elevation data, an improved elevation data sequence is obtained; in this embodiment, the improved elevation data sequence fluctuates less than the original elevation data.

[0061] Based on the improved elevation data sequence and Butterworth filter coefficients, smooth elevation data is obtained to complete the processing of elevation data. The spatial resolution of the smoothed elevation data is better than 5 meters.

[0062] S2: Based on the processed satellite remote sensing images, the geomorphic features of Danxia landform are obtained;

[0063] A further implementation method is to obtain the geomorphic features of Danxia landform by:

[0064] The principal component analysis method is used to perform band analysis on the preprocessed satellite remote sensing images to obtain the covariance matrix of each band and calculate the eigenvalues ​​of the covariance matrix. Specifically, the spectral data of the satellite remote sensing images are decentralized and stored row by row in a data matrix with a preset number of rows and columns to obtain the covariance matrix of the data matrix and the eigenvalues ​​of the covariance matrix.

[0065] Principal components are selected based on eigenvalues ​​to obtain satellite remote sensing images after dimensionality reduction;

[0066] Decompose the satellite remote sensing image after dimension reduction to obtain image blocks with central pixels as labels;

[0067] A three-dimensional convolutional neural network is used to extract features from image blocks to obtain the geomorphic features of Danxia landforms, which include spectral information, color information, texture information, geometric information, and mineral composition information. Among them, the mineral composition information is reflected by the spectrum. The purpose of extracting geomorphic features is to accurately segment the three geomorphic units of the top platform, steep cliff, and foothill accumulation of Danxia landforms.

[0068] The three-dimensional convolutional neural network used in this embodiment extracts features from different spatial scales, uses the maximum pooling layer to divide the input image block into several matrices, and takes the maximum value in each matrix as the output of the area; in the network structure, batch normalization is used to alleviate the gradient vanishing problem and speed up the network convergence. The discard method is used in the fully connected layer to set the components of the input signal to 0 with a probability of 0.5 to alleviate overfitting. The output is generated in the Softmax layer, and the cross entropy loss function is combined to accelerate the convergence of the network. The Adam optimizer is used for optimization calculations.

[0069] In this embodiment, when a three-dimensional convolutional neural network is used to extract features from image blocks, the spectral differences of the three landform units are utilized. For example, the top platform and the foothill accumulation are usually covered with vegetation, showing red and green, while the steep cliff is exposed, showing significantly different spectral characteristics, and the spectral mean, variance and other features are extracted; in terms of texture, the steep cliff has a large slope and obvious shadows, while the top platform and the foothill accumulation are relatively flat, and the GLCM texture features are extracted; in terms of geometric shape, the top platform has good continuity and a large area, while the steep cliff is narrow and broken, and the foothill accumulation is discretely distributed, and the shape index, compactness and other features are extracted to construct a segmentation rule set to achieve accurate segmentation.

[0070] In this embodiment, it also includes obtaining high-precision point cloud data of the cliff layer, extracting geometric features of the cliff using three-dimensional laser scanning technology, and classifying the cliff using a machine learning algorithm;

[0071] First, a three-dimensional scan of the target cliff area was performed using a laser radar with a scanning range of 500 meters × 500 meters and a scanning resolution of 5 centimeters to obtain up to 100 million three-dimensional point cloud data. The point cloud density was set to 100 points per square meter. The statistical filtering method was used to remove noise points and outliers. Points in the point cloud data that deviated from the surrounding points by more than 5 times the standard deviation were considered outliers and removed. Then, the point cloud was segmented using a normal vector-based regional growing algorithm. The normal vector angle threshold was set to 15 degrees and the curvature threshold was set to 1, and a point cloud subset of the cliff area was extracted.

[0072] For the extracted subset of cliff point cloud, the normal vector and curvature of the point cloud are calculated by principal component analysis to obtain characteristic parameters reflecting the geometric morphology of the cliff. Specifically, the normal vector and curvature of each point are calculated by principal component analysis, and the points with curvature greater than 0.5 are taken as the characteristic points of the cliff. Then the RANSAC algorithm is used to fit the cliff point cloud to a plane, and the distance threshold from the point to the plane is set to 2 meters, and the number of iterations is 1000 times to obtain the optimal plane fitting model. The slope of the cliff is calculated by the angle between the fitting plane and the horizontal plane, and it is judged as a steep cliff when the angle is greater than 75 degrees. In the cliff point cloud, the angle between the normal vector of each point and the normal vector of the fitting plane is calculated. When the angle is greater than 45 degrees, the curvature value of the point is large, reflecting that the local morphology of the cliff changes dramatically. Finally, the parameters such as the slope, aspect, and curvature of the cliff are combined into a 12-dimensional feature vector as the terrain data for the stability analysis of the cliff.

[0073] According to the acquired terrain data, the geometric features of the terrain are extracted, including slope, aspect, curvature and other features. The least squares support vector machine algorithm is used to classify and train the steep cliff and non-steep cliff areas using the extracted geometric features. The kernel function type and kernel function parameters of the support vector machine are optimized by the cross-validation method to improve the classification accuracy. The trained support vector machine classifier is used as a steep cliff classification model for subsequent steep cliff identification tasks. For the newly input terrain data, its geometric features are extracted, and the constructed steep cliff classification model is used for classification prediction. The classification prediction results are post-processed to eliminate isolated misclassified areas and obtain the final steep cliff distribution results.

[0074] Geometric features are extracted from terrain data, and the Horn algorithm is used to calculate the slope value of each grid, with a slope range of 0° to 90°; the elevation values ​​of the adjacent 8 grids are used to calculate the slope aspect, with a slope aspect range of 0° to 360°; the curvature can be obtained by fitting a quadratic polynomial surface, and the larger the curvature value, the steeper the terrain. Then, appropriate training samples are selected, and 1000 sample points are selected for the steep cliff and non-steep cliff areas, and their corresponding feature vectors such as slope, slope aspect, and curvature are extracted. The least squares support vector machine algorithm is then used to train the sample data, and the optimal Gaussian kernel function parameters are determined through 5-fold cross validation, such as the penalty factor C=10 and the kernel function width σ=1, so as to construct a steep cliff classification model. For new terrain data, its geometric features are extracted and input into the classification model for prediction, and the preliminary steep cliff distribution results are obtained. Finally, the morphological closing operation is used to post-process the classification results, and the 3×3 structural element is used for expansion and corrosion operations to eliminate isolated areas with an area of ​​less than 100 square meters to obtain the final steep cliff distribution map. By overlaying the distribution of steep cliffs with the terrain shading base map, a fused three-dimensional visualization scene is generated, which intuitively displays the spatial distribution characteristics of steep cliffs in the terrain and provides a reference for applications such as disaster prevention, risk avoidance, and engineering planning.

[0075] S3: Based on the processed elevation data, the topographical characteristics of Danxia landform are obtained;

[0076] A further implementation method is to obtain the topographic relief characteristics of Danxia landform by:

[0077] Based on the gradient of the processed elevation data in two dimensions, the slope information of Danxia landform is obtained;

[0078] Specifically, the numerical differentiation method is used to calculate the gradient of each point in the x-direction and y-direction in the elevation data matrix. Assuming that the elevation data matrix is ​​H(x,y), the gradient in the x-direction is and the gradient in the y direction It can be approximated by dividing the elevation difference of adjacent points by the distance.

[0079] The slope α can be calculated by the vector sum of the gradients:

[0080] where α is the slope angle in radians.

[0081] Based on the coefficient of variation of the processed elevation data, the terrain roughness of the Danxia landform is obtained;

[0082] Specifically, first calculate the average value of the elevation data

[0083] Where N is the total number of data points, H i is the elevation value of each point;

[0084] Then calculate the standard deviation σ of the elevation data H :

[0085]

[0086] The coefficient of variation, C, is the ratio of the standard deviation to the mean:

[0087] The larger the coefficient of variation, the rougher the terrain.

[0088] Based on the slope information and terrain roughness, the terrain undulation characteristics of Danxia landform are obtained. Specifically, the slope information is combined with the terrain roughness to analyze which areas have large slopes and high roughness. These areas often have significant terrain undulation characteristics. According to different combinations of slope and roughness, the terrain can be divided into different categories, such as flat areas, gentle slope areas, steep slope areas, rugged areas, etc.

[0089] S4: Based on the geomorphic features and terrain undulation characteristics, a Danxia landform recognition model is constructed to obtain recognition results;

[0090] A further implementation method is to construct a Danxia landform recognition model based on landform features and terrain undulation features:

[0091] The geomorphic features and terrain undulation features are fused element by element to obtain the initial fusion features. Specifically, first, the geomorphic features F geo and terrain relief features F topo Perform element-wise product operation to obtain the initial fusion feature F init This fusion method can emphasize the common salient parts of the two features while suppressing the insignificant areas.

[0092] F init =F geo ⊙F topo ,

[0093] where ⊙ represents the element-wise product.

[0094] Based on the channel dimension of the initial fusion feature, channel extraction is performed; specifically, the SE attention mechanism is used to dynamically adjust the importance of each channel to achieve channel extraction:

[0095] F selected =ChannelAttention(F init ).

[0096] Based on the convolution kernel of the fully connected layer of the neural network and the preset activation function (ReLU), the channel relationship of the initial fusion feature is calculated; specifically, a 1x1 convolution kernel (also called point convolution) is used to simulate this process because it allows information exchange across channels. Then, an activation function ReLU is applied to increase nonlinearity:

[0097] F relation =σ(Conv1x1(F init )),

[0098] Among them, σ is the activation function and Conv1x1 represents a 1x1 convolution operation.

[0099] The result of channel extraction F selected Relationship with channel F relation Perform element-wise multiplication to obtain the reconstructed features;

[0100] F reconstructed =F selected ⊙F relation .

[0101] Based on the reconstructed features and the initial fused features, the final fused features are obtained;

[0102] Based on the final fusion features, geomorphic features and terrain undulation features, decision fusion is performed to complete the construction of the Danxia landform recognition model.

[0103] S5: Based on the identification results and combined with field research, analyze the formation mechanism of Danxia landform in the preset area and complete the investigation of Danxia landform in the preset area.

[0104] The implementation method of the first step is that the recognition results include mountain peak class, steep cliff class, canyon class, cave class and landslide class.

[0105] A further implementation method is to analyze the formation mechanism of Danxia landform in a preset area by:

[0106] Based on the recognition results, the distribution range and area size of Danxia landforms in the preset area are counted to obtain the spatial distribution characteristics of Danxia landforms;

[0107] Based on the spatial distribution characteristics, the corresponding genetic mechanism parameters of Danxia landform are collected; the genetic mechanism parameters include geological background, climate factors and hydrological conditions;

[0108] Calculate the correlation between each genetic mechanism parameter and Danxia landform type to obtain the correlation sequence;

[0109] In this embodiment, the specific method of obtaining the correlation sequence is:

[0110] The genetic mechanism parameters are preprocessed to remove outliers and noise data to obtain a normalized data set. Based on the data set, a regression analysis model is constructed, and the regression relationship between the landform type and the genetic mechanism parameters is determined through model training. For the trained regression model, the cross-validation method is used to evaluate the goodness of fit and prediction performance of the model. If the model performance meets the preset threshold, it is determined to be the optimal regression model. Using the optimal regression model, the new landform type is predicted and analyzed to obtain the corresponding genetic mechanism parameter prediction results. The correlation coefficient matrix between each genetic mechanism parameter and the Danxia landform type is calculated to determine the correlation between each landform type and the genetic mechanism parameter. According to the correlation coefficient matrix, the correlation significance of the genetic mechanism parameters is judged, and the significantly correlated genetic mechanism parameter characteristics (correlation sequence, which is the genetic mechanism parameter sequence related to the formation of Danxia landform type) are obtained as the key features affecting the formation of landforms.

[0111] Based on the correlation sequence, the significant causal parameters of Danxia landform are obtained, and the analysis of the causal mechanism of Danxia landform in the preset area is completed.

[0112] In this embodiment, it also includes constructing a geological model of micro-geomorphological development based on the correlation sequence, clarifying the control effect of different geological conditions on the development of micro-geomorphology, and predicting the development and evolution trend of micro-geomorphology.

[0113] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for investigating Danxia landform, characterized in that: The following steps are involved: Collect and process satellite remote sensing images and elevation data of Danxia landforms in the preset area; Based on the processed satellite remote sensing image, the geomorphic features of the Danxia landform are obtained; Based on the processed elevation data, the terrain undulation characteristics of Danxia landform are obtained; Based on the geomorphic features and the terrain undulation features, a Danxia landform recognition model is constructed to obtain a recognition result; Based on the identification results, combined with field research, the formation mechanism of the Danxia landform in the preset area is analyzed to complete the investigation of the Danxia landform in the preset area.

2. The method for investigating Danxia landform according to claim 1, characterized in that: The method for processing the satellite remote sensing image includes radiation calibration, atmospheric correction and clipping, wherein the atmospheric correction method is: Based on the MODIS surface reflectance product, the ratio factor map of Danxia landform is obtained; Performing aerosol inversion based on the ratio factor map to obtain an aerosol map; Obtain atmospheric parameters when collecting satellite remote sensing images, and invert water vapor information based on the water vapor detection band to obtain high-resolution water vapor data; Based on the aerosol map and the high-resolution water vapor data, atmospheric correction is performed on the satellite remote sensing image.

3. The method for investigating Danxia landform according to claim 1, characterized in that: The method to obtain the geomorphic features of Danxia landform is: Using principal component analysis, the preprocessed satellite remote sensing image is subjected to band analysis to obtain a covariance matrix of each band, and the eigenvalue of the covariance matrix is ​​calculated; Selecting principal components based on the eigenvalues ​​to obtain satellite remote sensing images after dimensionality reduction; Decompose the satellite remote sensing image after dimension reduction to obtain image blocks with central pixels as labels; A three-dimensional convolutional neural network is used to extract features from the image blocks to obtain geomorphic features of the Danxia landform; wherein the geomorphic features include spectral information, color information, texture information, geometric information and mineral composition information.

4. The method for investigating Danxia landform according to claim 1, characterized in that: The method for processing the elevation data is: Processing the elevation data based on a local enhancement mechanism to obtain an improved elevation data sequence; Based on the improved elevation data sequence and the Butterworth filter coefficients, smooth elevation data is obtained to complete the processing of the elevation data.

5. The method for investigating Danxia landform according to claim 1, characterized in that: The method to obtain the topographic relief characteristics of Danxia landform is: Based on the gradient of the processed elevation data in two dimensions, the slope information of Danxia landform is obtained; Based on the coefficient of variation of the processed elevation data, the terrain roughness of the Danxia landform is obtained; Based on the slope information and the terrain roughness, the terrain undulation characteristics of the Danxia landform are obtained.

6. The method for investigating Danxia landform according to claim 1, characterized in that: Based on the geomorphic features and the terrain undulation features, the method for constructing the Danxia landform recognition model is: Performing element-by-element product fusion on the geomorphic features and the terrain relief features to obtain initial fusion features; Based on the channel dimension of the initial fusion feature, channel extraction is performed; Based on the fully connected layer convolution kernel of the neural network and the preset activation function, the channel relationship of the initial fusion feature is calculated; Performing element-wise multiplication of the channel extraction result and the channel relationship to obtain a reconstructed feature; Based on the reconstructed features and the initial fused features, obtaining final fused features; Based on the final fusion features, the landform features and the terrain undulation features, decision fusion is performed to complete the construction of the Danxia landform recognition model.

7. The method for investigating Danxia landform according to claim 1, characterized in that: The identification results include mountain peaks, steep cliffs, canyons, caves and landslides.

8. The method for investigating Danxia landform according to claim 1, characterized in that: The method for analyzing the formation mechanism of Danxia landform in the preset area is: Based on the recognition results, the distribution range and area size of Danxia landforms in a preset area are counted to obtain the spatial distribution characteristics of Danxia landforms; Based on the spatial distribution characteristics, the corresponding genetic mechanism parameters of Danxia landform are collected; wherein the genetic mechanism parameters include geological background, climate factors and hydrological conditions; Calculate the correlation between each genetic mechanism parameter and Danxia landform type to obtain the correlation sequence; Based on the correlation sequence, the significant causal parameters of Danxia landform are obtained, and the analysis of the causal mechanism of Danxia landform in the preset area is completed.

Images