Landform stage division and DEM construction method based on space-time index

Abstract

The invention discloses a geomorphic stage division and DEM construction method based on an air substitution time index, and relates to the technical field of surveying and mapping remote sensing and geographic information processing.The method comprises the steps that firstly, space cutting is conducted on a high-resolution DEM, an independent watershed sample is extracted, terrain elements such as a valley line, a ridge line, a channel line and a watershed boundary line are extracted from the independent watershed sample, the terrain elements are spliced with the DEM, and an independent watershed model is obtained; constructing an image pair sample; secondly, for each sample, calculating four types of air generation time indexes including area elevation integral, main ditch-branch ditch ratio, gully density and surface roughness, giving weights to the indexes in combination with expert knowledge, and dividing the samples in the juvenile period, the adulty period and the old period according to a weighting result to obtain an air generation time data set; and finally, training the CGAN network in stages, and generating high-resolution DEMs in different development stages. Through scientific landform development stage division, the expression ability of the model to landform details is improved, and the adaptability and interpretability to different landform evolution stages are also enhanced.

Description

Technical Field

[0001] The present invention relates to the field of surveying, mapping, remote sensing and geographic information processing technology, and in particular to a landform stage division and DEM construction method based on space-time indicators. Background Art

[0002] A Digital Elevation Model (DEM) is a core geographic information product that represents the spatial structure of the Earth's surface. It is widely used in a variety of fields, including watershed geomorphological analysis, soil erosion simulation, terrain mapping, and disaster risk assessment. While high-resolution DEMs can accurately depict microtopographic features, they often face data gaps or inaccuracies in practical applications due to high data acquisition costs and limited coverage of complex terrain areas.

[0003] In recent years, the application of deep learning techniques in remote sensing image processing has rapidly advanced. In particular, the CGAN (Conditional Generative Adversarial Network), exemplified by Pix2Pix, has demonstrated excellent nonlinear modeling capabilities in image-to-image translation tasks, providing an effective path for generating corresponding DEMs from terrain feature maps. These methods improve the spatial resolution and structural reconstruction capabilities of DEMs by learning the mapping relationship between terrain features and elevation. However, existing methods often train models using a single dataset, failing to account for stage-by-stage differences in landform development. This results in inadequate representation of terrain structure at different stages of evolution.

[0004] On the other hand, the theory of space-for-time substitution, a key framework in landform evolution research, emphasizes the use of spatially distinct watersheds to replace time series observations in order to infer the evolutionary trends of landform systems. By constructing a system of quantifiable landform indicators, such as the elevation integral index, gully density, and surface roughness, it is possible to characterize and classify the evolutionary stages of watersheds, providing a theoretical basis for cognitive modeling of landform structural differences. Although these indicators have been widely used in landform classification and evolution research, their systematic integration and methodological support in deep learning-driven DEM construction remain lacking. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention provides a landform stage division and DEM construction method based on space-time indicators, comprising the following steps: S1. Obtain digital elevation model (DEM) data with a resolution of 5m or less in the study area, preprocess and manually crop it, and extract batches of independent small watershed units; embed each small watershed unit into a standardized image of 256×256 pixels, and batch construct DEM samples with a unified format; S2, based on the hydrological analysis method and the automatic segmentation method of loess positive and negative terrain, four terrain elements are extracted from each DEM sample, namely gully line, ridge line, valley line and watershed boundary line; S3. Encode the ditch line, ridge line, and valley line into the red, green, and blue channels of the RGB image respectively, and the watershed boundary line is represented by grayscale overlay, completing the fusion encoding of the four types of terrain elements and obtaining a terrain element map; S4, horizontally splicing the terrain feature map of each sample with the corresponding DEM sample to generate an image pair sample with a size of 512×256 pixels, and obtaining a terrain sample dataset; S5. Based on the space-time theory of geomorphology, four space-time indices are calculated for each image pair in the terrain sample dataset, namely, area-elevation integral, main gully to branch gully ratio, gully density, and surface roughness. S6. Standardize the four space-time indicators, assign weights to each indicator based on expert knowledge, and perform a weighted comprehensive score on all samples. Based on the score results, divide the terrain sample dataset into three different landform development stage categories: juvenile, mature, and old. Finally, the space-time dataset is obtained. The space-time dataset includes a juvenile stage dataset, a mature stage dataset, and a old stage dataset. S7. Use the childhood dataset, the adult dataset, and the elderly dataset to train the conditional generative adversarial network Pix2Pix in stages, and obtain three terrain generation models with the ability to perceive the characteristics of childhood, adult, and elderly stages respectively; S8. Input each validation set into the terrain generation model trained in the corresponding stage to generate a DEM with a resolution of 5m or less. Then perform accuracy verification based on quantitative and qualitative indicators to evaluate the reconstruction performance of the three terrain generation models in the space-time dataset.

[0006] The technical solution further defined in the present invention is: Furthermore, in step S1, manual clipping is performed based on the DEM with a resolution of 5 m or less, and the river network and the initial small watershed boundary are extracted using the hydrological analysis method. The small watershed units with complete outlines and independent structures are screened by manual review, and the corresponding DEM area is clipped through mask operation.

[0007] As described above, the landform stage division and DEM construction method based on space-time indicators, in step S2, extracting four terrain elements from each DEM sample specifically includes the following sub-steps: S2.1. Based on the constructed batch DEM samples, hydrological analysis methods are used to fill depressions, calculate flow direction and flow, and extract ridge lines in combination with the confluence threshold. S2.2. Construct a negative terrain model by elevation inversion and extract valley lines using hydrological analysis methods; S2.3. Based on the automatic segmentation method of loess positive and negative terrain, a difference elevation map is constructed to extract the local positive and negative terrain boundaries and obtain the ditch line; S2.4. Based on the slope and water system distribution, delineate the closed watershed and extract the watershed boundary line.

[0008] As described above, in the landform stage division and DEM construction method based on space-time indicators, in step S5, the area-elevation integral reveals important indicators of valley landform morphology and development characteristics by statistically analyzing the elevation combination information of the watershed surface. The specific calculation method of the area-elevation integral is as follows: First, divide the normalized elevation value into n equal-width intervals: Then count the number of pixels in each elevation interval that are higher than the normalized elevation value of the current segment. i , and calculate the normalized area: Finally, the trapezoidal method is used to approximate the integral and calculate HI: Among them, H i Indicates the normalized elevation value, z i represents the original elevation value of the ith elevation segment, A(h i ) represents the normalized area above the current segment normalized elevation value, a j Indicates that the elevation is greater than h i The number of pixels, A represents the total effective area, n represents the number of segments, z min and z max They represent the minimum and maximum elevation values ​​in the DEM, HI represents the area elevation integral, and h i and h i+1 They represent the normalized elevation values ​​of the i-th segment and the i+1-th segment respectively.

[0009] As mentioned above, in the landform stage division and DEM construction method based on space-time indicators, in step S5, the main ditch to branch ditch ratio is defined as the ratio of the main ditch length to the total ditch length in the basin, which is used to reflect the development degree of branch ditch in the ditch system from the perspective of ditch length. Its mathematical expression is: Among them, R represents the ratio of main ditch to branch ditch, L0 represents the length of main ditch valley, and L represents the total length of ditch valley.

[0010] As mentioned above, a landform stage division and DEM construction method based on space-time indicators, in step S5, the gully density is defined as the total length of gullies per unit area, in km / km 2 , its mathematical expression is: Among them, D s represents the gully density; ∑L represents the total length of gullies in the study area, in km; S represents the area of ​​a specific sample area, in km 2 .

[0011] As mentioned above, in the landform stage division and DEM construction method based on space-time indicators, in step S5, the surface roughness is defined as the ratio of the surface area of ​​the bin to its projected area on the horizontal plane, which is used to characterize the degree of surface erosion and relief. Its mathematical formula is expressed as: Among them, S 曲面 represents the surface area of ​​the cell, S 平面 Represents the projected area of ​​the surface element on the horizontal plane.

[0012] As mentioned above, in the landform stage division and DEM construction method based on space-time indicators, in step S6, the four space-time indicators are standardized using the Z-score standardization method. This method converts the eigenvalues ​​into a distribution with a mean of 0 and a standard deviation of 1. Its mathematical expression is: Among them, x′ represents the standardized value of each space-time index, x represents the original index value of each space-time index, μ represents the average value of each space-time index, and σ represents the average difference of each space-time index.

[0013] As mentioned above, a method for landform stage division and DEM construction based on space-time indicators, in step S6, expert knowledge is based on indicators used to reflect landform development, which include the size of the effective pixel area in each sample, the length of the main valley, the geomorphological space-time theory and the Strahler classification method.

[0014] As described above, a landform stage division and DEM construction method based on space-time indicators is proposed. In step S7, the conditional generative adversarial network Pix2Pix includes a generator and a discriminator. The generator adopts a U-Net structure, which consists of an encoder and a decoder. The encoder is used to extract image features, and the decoder is used to generate output images. The discriminator adopts a PatchGAN structure to perform local comparison and discrimination between the generated results and the real images.

[0015] The beneficial effects of the present invention are: (1) In this invention, by training the CGAN-based generative model in stages and optimizing it specifically for the characteristics of different landform development stages, the spatial detail expression capability of high-resolution DEM is significantly improved. Compared with the rough treatment of terrain complexity by traditional DEM construction methods, this invention makes full use of key terrain element information such as valley lines and ridge lines, effectively enhancing the true restoration of terrain details, making the generated DEM more delicate and accurate in terms of water system morphology and landform sections. High-resolution DEM not only provides a more reliable data basis for applications such as geological exploration and flood simulation, but also creates conditions for refined terrain analysis and environmental monitoring. (2) In this invention, based on multiple space-time indicators such as area-elevation integral and main-ditch-branch ditch ratio, combined with expert empowerment, a scientific division of landform stages is achieved; this stage division based on the landform evolution process breaks through the limitation of the lack of landform evolution background in previous DEM construction, making the model training and generation process more targeted and hierarchical; through staged modeling, the model not only has better adaptability to the terrain characteristics of different evolutionary stages, but also improves the geomorphological interpretability of the generated results, facilitates expert verification and subsequent analysis, and helps promote the in-depth application of digital terrain products in fields such as geomorphology and environmental science; (3) In the present invention, a complete automated process is designed, from high-resolution DEM spatial clipping, terrain feature extraction, to space-time indicator calculation and stage division, and then to stage-by-stage CGAN training. This process greatly reduces manual participation, improves the efficiency of data preprocessing and model training, and ensures the standardization and repeatability of the method. At the same time, the present invention has good adaptability and versatility, can be quickly applied to different terrain areas and various watershed types, and facilitates the construction and updating of large-scale digital terrain products. In addition, the present invention has important engineering promotion value for improving the automation level of remote sensing and terrain data processing and promoting the intelligent generation of terrain data. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the overall process of the present invention; Figure 2 Schematic diagram of the 5-meter DEM data of the study area and the DEM samples extracted therefrom in an embodiment of the present invention; Figure 3 Flowchart for producing a terrain element image pair sample according to an embodiment of the present invention; Figure 4 Schematic diagram of a space-time dataset in an embodiment of the present invention; each column from left to right in the figure is a schematic diagram of a portion of the dataset for childhood, adulthood, and old age, and from top to bottom are schematic diagrams of image pair samples for different independent small watersheds in each stage; Figure 5 This is a principal component analysis distribution diagram of landform development stage samples in an embodiment of the present invention; Figure 6 This is a schematic diagram of the network structure of the conditional generative adversarial network pix2pix in an embodiment of the present invention; Figure 7 This is a comparison chart of the DEM construction results of each development stage in the space-time data set in an embodiment of the present invention; from left to right and from top to bottom in the figure, there are the reference DEM schematic diagram of three independent small watersheds representing the childhood, middle-aged and old stages, the generated virtual DEM schematic diagram, the reference DEM hill shadow map and the generated virtual DEM hill shadow map. DETAILED DESCRIPTION

[0017] This embodiment provides a method for landform stage division and DEM construction based on space-time indicators. It is suitable for application scenarios such as terrain modeling, landform evolution research, and terrain data reconstruction in complex areas. The landform stages are divided using space-time indicators such as area-elevation integral and main-to-branch ditch ratio. The method is combined with the CGAN network for staged training to improve the model's ability to express terrain structures at different development stages and the reconstruction accuracy.

[0018] like Figure 1 As shown in the figure, a landform stage division and DEM construction method based on space-time indicators includes the following steps: S1, first obtain Figure 2 The 5-meter high-resolution DEM (Digital Elevation Model) data of the Jiuyuangou watershed in the Loess Plateau of northern Shaanxi Province was preprocessed by projection conversion and filtering and denoising. Then, the river network and initial small watershed boundaries were extracted based on the hydrological analysis method. The small watershed units with complete outlines and independent structures were selected by manual review, and the corresponding DEM areas were manually cropped through masking operations. The cropped batches of small watersheds were each centered and embedded in an image of 256×256 pixels, with the surrounding area filled with NoData values. The batches were constructed as shown in the figure. Figure 2 A sample DEM in a uniform format is shown.

[0019] S2. Based on the constructed batch DEM samples, a hydrological analysis method is used to fill depressions, calculate flow direction and flow, and set the confluence threshold to 200 to extract ridge lines. The negative terrain model is constructed by subtracting the original DEM value from the maximum elevation value of the region. The hydrological analysis method with the same threshold is applied to the negative terrain to extract valley lines. Based on the automatic segmentation method of loess positive and negative terrain, a difference elevation map is constructed, and the local positive and negative terrain boundaries are extracted to obtain ditch lines. Combined with the slope and water system distribution, closed watershed watersheds are delineated and the watershed boundary lines are extracted.

[0020] S3. Encode the processed terrain features (gully lines, ridge lines, valley lines, and watershed boundaries) into a three-channel RGB image, where valley lines are encoded into the blue channel, ridge lines are encoded into the green channel, gully lines are encoded into the red channel, and watershed boundaries are superimposed in gray (220, 220, 220) to obtain a terrain feature map.

[0021] S4, such as Figure 3 As shown in Figure 1, the terrain feature map of each sample is horizontally spliced ​​with the corresponding DEM image to generate an image pair sample with a size of 512×256 pixels, and a terrain sample dataset is obtained.

[0022] S5. Based on the space-time theory of geomorphology, four space-time indices are calculated for each image in the terrain sample dataset: area-elevation integral, main-gully-branch gully ratio, gully density, and surface roughness.

[0023] HI (Hypsometric Integral) is an important indicator that reveals the morphology and development characteristics of valleys by statistically analyzing the elevation combination information of the watershed surface. The specific calculation method is as follows: First, divide the normalized elevation value into n equal-width intervals: Then count the number of pixels in each elevation interval that are higher than the normalized elevation value of the current segment. i , and calculate the normalized area: Finally, the trapezoidal method is used to approximate the integral and calculate HI: Among them, H i Indicates the normalized elevation value, z i represents the original elevation value of the ith elevation segment, A(h i ) represents the normalized area above the current segment normalized elevation value, a j Indicates that the elevation is greater than h i The number of pixels, A represents the total effective area, n represents the number of segments, z min and zmax They represent the minimum and maximum elevation values ​​in the DEM, HI represents the area elevation integral, and h i and h i+1 They represent the normalized elevation values ​​of the i-th segment and the i+1-th segment respectively.

[0024] The main-to-branch ratio is the ratio of the main valley length to the total valley length in the basin. It reflects the development degree of the branch valleys in the valley system from the perspective of valley length. Its mathematical expression is: Among them, R represents the ratio of main ditch to branch ditch, L0 represents the length of main ditch valley, and L represents the total length of ditch valley.

[0025] Gully density, also known as valley density or channel density, refers to the total length of gullies per unit area, in km / km 2 , its mathematical expression is: Among them, D s represents the gully density; ∑L represents the total length of gullies in the study area, in km; S represents the area of ​​a specific sample area, in km 2 .

[0026] Surface roughness is an indicator of the degree of surface erosion and relief. It is usually defined as the ratio of the surface area of ​​a cell to its projected area on the horizontal plane. Its mathematical formula is: Among them, S 曲面 represents the surface area of ​​the cell, S 平面 Represents the projected area of ​​the surface element on the horizontal plane.

[0027] The calculation results of the terrain indicators of the samples of some images of the study area in this embodiment are shown in Table 1 below.

[0028] Table 1 Calculation results of space-time index for small watersheds in the study area

[0029] S6. First, the calculated four space-time index values ​​are subjected to Z-score standardization (standard deviation normalization) to convert them into a standard normal distribution with a mean of 0 and a standard deviation of 1. The mathematical expression is: Among them, x′ represents the standardized value of each space-time index, x represents the original index value of each space-time index, μ represents the average value of each space-time index, and σ represents the average difference of each space-time index.

[0030] Secondly, based on expert knowledge, a weight of 0.4 was assigned to the main-to-branch ratio and surface roughness, and a weight of 0.1 was assigned to the gully density and area-elevation integral, and a weighted comprehensive score of 0.1 was calculated for all samples. The expert knowledge was based on indicators that can reflect landform development, such as the size of the effective pixel area in each sample, the length of the main gully, the geomorphological space-time theory, and the Strahler classification method. Subsequently, based on the scoring results, the terrain sample dataset was divided into three landform development stages: infancy, maturity, and old age, and the following was obtained: Figure 4 Shown is a space-generation-time dataset with three developmental stages.

[0031] In order to intuitively display the distribution characteristics of each sample under multiple index dimensions, this embodiment performs PCA (Principal Component Analysis) dimensionality reduction processing on the standardized indexes, such as Figure 5 The figure shows the distribution of image pair samples in the principal component space (PC1 and PC2). Each dot represents a sample, and its position reflects its comprehensive characteristics in the two principal component dimensions.

[0032] S7. Divide the samples of the three developmental stages in the space-time dataset into training and validation sets at a ratio of 80% and 20% respectively, and then train the conditional generative adversarial network pix2pix separately to obtain three terrain generation models with the ability to perceive features at different stages.

[0033] The network structure of the conditional generation adversarial network pix2pix is ​​as follows Figure 6 As shown in the figure, it includes a generator and a discriminator; the generator adopts a U-Net structure, which consists of an encoder and a decoder. The encoder is used to extract image features, and the decoder is used to generate output images; the discriminator adopts a PatchGAN structure, which performs local comparison and discrimination between the generated results and the real image to improve the detail quality and structural consistency of the generated image.

[0034] S8. The validation sets of the three stages were input into the corresponding terrain generation models to generate high-resolution virtual DEMs. Quantitative evaluation was performed based on the root mean square error (RMSE) of elevation and slope, and qualitative evaluation was performed based on hillshade. The performance differences of the models at different landform development stages in the space-time generation dataset were compared. The space-time generation terrain modeling accuracy verification results are shown in Table 2 below. The DEM construction results of each development stage are compared. Figure 7 shown.

[0035] Table 2 Verification of space-time terrain modeling accuracy

[0036] The experimental results show that the landform stage division based on space-time indicators significantly improves the ability of DEM modeling to express stage characteristics; among them, the old stage model has the best accuracy, and its DEM mountain shadow fully presents the typical characteristics of wide and gentle gullies and smooth watersheds, with the best texture continuity; the mature stage model accurately depicts the balanced development of the gully system, and the transition of the main and branch gullies is natural; although the juvenile stage model is limited by the complexity of the terrain, it still effectively retains the initial development characteristics of steep gullies and sharp watersheds; the above results confirm the scientificity and practicality of the space-time indicator system, and provide a new technical paradigm for DEM modeling for landform evolution stages.

[0037] The method in this embodiment introduces a landform development stage division mechanism to construct a DEM construction system with stage perception capabilities. It not only focuses on the performance of the model in terms of overall elevation prediction accuracy, but also pays attention to its structural expression ability, morphological adaptability and generalization stability in different landform stages, thereby providing more scientific technical support for high-quality DEM construction and landform process simulation in complex terrain areas.

[0038] In addition to the above embodiments, the present invention may also have other implementations. Any technical solution formed by equivalent replacement or equivalent transformation falls within the scope of protection required by the present invention.

Claims

1. A method for landform stage division and DEM construction based on space-time indicators, characterized by: The following steps are involved: S1. Obtain digital elevation model (DEM) data with a resolution of 5m or less in the study area, preprocess and manually crop it, and extract batches of independent small watershed units; embed each small watershed unit into a standardized image of 256×256 pixels, and batch construct DEM samples with a unified format; S2, based on the hydrological analysis method and the automatic segmentation method of loess positive and negative terrain, four terrain elements are extracted from each DEM sample, namely gully line, ridge line, valley line and watershed boundary line; S3. Encode the ditch line, ridge line, and valley line into the red, green, and blue channels of the RGB image respectively, and the watershed boundary line is represented by grayscale overlay, completing the fusion encoding of the four types of terrain elements and obtaining a terrain element map; S4, horizontally splicing the terrain feature map of each sample with the corresponding DEM sample to generate an image pair sample with a size of 512×256 pixels, and obtaining a terrain sample dataset; S5. Based on the space-time theory of geomorphology, four space-time indices are calculated for each image pair in the terrain sample dataset, namely, area-elevation integral, main gully to branch gully ratio, gully density, and surface roughness. S6. Standardize the four space-time indicators, assign weights to each indicator based on expert knowledge, and perform a weighted comprehensive score on all samples. Based on the score results, divide the terrain sample dataset into three different landform development stage categories: juvenile, mature, and old. Finally, the space-time dataset is obtained. The space-time dataset includes a juvenile stage dataset, a mature stage dataset, and a old stage dataset. S7. Use the childhood dataset, the adult dataset, and the elderly dataset to train the conditional generative adversarial network Pix2Pix in stages, and obtain three terrain generation models with the ability to perceive the characteristics of childhood, adult, and elderly stages respectively; S8. Input each validation set into the terrain generation model trained in the corresponding stage to generate a DEM with a resolution of 5m or less. Then perform accuracy verification based on quantitative and qualitative indicators to evaluate the reconstruction performance of the three terrain generation models in the space-time dataset.

2. The method for landform stage division and DEM construction based on space-time indicators according to claim 1 is characterized by: In step S1, manual clipping is based on the DEM with a resolution of 5m or less, and a hydrological analysis method is used to extract the river network and the initial small watershed boundary. The small watershed units with complete outlines and independent structures are screened by manual review, and the corresponding DEM area is clipped through mask operation.

3. The method for landform stage division and DEM construction based on space-time indicators according to claim 1 is characterized by: In step S2, extracting four terrain elements from each DEM sample specifically includes the following sub-steps: S2.

1. Based on the constructed batch DEM samples, hydrological analysis methods are used to fill depressions, calculate flow direction and flow, and extract ridge lines in combination with the confluence threshold. S2.

2. Construct a negative terrain model by elevation inversion and extract valley lines using hydrological analysis methods; S2.

3. Based on the automatic segmentation method of loess positive and negative terrain, a difference elevation map is constructed to extract the local positive and negative terrain boundaries and obtain the ditch line; S2.

4. Based on the slope and water system distribution, delineate the closed watershed and extract the watershed boundary line.

4. The method for landform stage division and DEM construction based on space-time indicators according to claim 1 is characterized by: In step S5, the area-elevation integral reveals important indicators of valley landform morphology and development characteristics by statistically analyzing the elevation combination information of the watershed surface. The specific calculation method of the area-elevation integral is as follows: First, divide the normalized elevation value into n equal-width intervals: Then count the number of pixels in each elevation interval that are higher than the normalized elevation value of the current segment. i , and calculate the normalized area: Finally, the trapezoidal method is used to approximate the integral and calculate HI: Among them, H i Indicates the normalized elevation value, z i represents the original elevation value of the ith elevation segment, A(h i ) represents the normalized area above the current segment normalized elevation value, a j Indicates that the elevation is greater than h i The number of pixels, A represents the total effective area, n represents the number of segments, z min and z max They represent the minimum and maximum elevation values ​​in the DEM, HI represents the area elevation integral, and h i and h i+1 They represent the normalized elevation values ​​of the i-th segment and the i+1-th segment respectively.

5. The method for landform stage division and DEM construction based on space-time indicators according to claim 1 is characterized by: In step S5, the main ditch to branch ditch ratio is defined as the ratio of the main ditch length to the total ditch length in the watershed, which is used to reflect the development degree of the branch ditch in the ditch system from the perspective of ditch length. Its mathematical expression is: Among them, R represents the ratio of main ditch to branch ditch, L0 represents the length of main ditch valley, and L represents the total length of ditch valley.

6. The method for landform stage division and DEM construction based on space-time indicators according to claim 1 is characterized by: In step S5, the gully density is defined as the total length of gullies per unit area, in km / km. 2 , its mathematical expression is: Among them, D s represents the gully density; ∑L represents the total length of gullies in the study area, in km; S represents the area of ​​a specific sample area, in km 2 .

7. The method for landform stage division and DEM construction based on space-time indicators according to claim 1 is characterized by: In step S5, surface roughness is defined as the ratio of the surface area of ​​a bin to its projected area on a horizontal plane, and is used to characterize the degree of surface erosion and undulation. Its mathematical formula is: Among them, S 曲面 represents the surface area of ​​the cell, S 平面 Represents the projected area of ​​the surface element on the horizontal plane.

8. The method for landform stage division and DEM construction based on space-time indicators according to claim 1 is characterized by: In step S6, the four space-time indices are standardized using the Z-score standardization method, which converts the eigenvalues ​​into a distribution with a mean of 0 and a standard deviation of 1. The mathematical expression is: Among them, x′ represents the standardized value of each space-time index, x represents the original index value of each space-time index, μ represents the average value of each space-time index, and σ represents the average difference of each space-time index.

9. The method for landform stage division and DEM construction based on space-time indicators according to claim 1 is characterized by: In step S6, the expert knowledge is based on indicators used to reflect landform development, including the size of the effective pixel area in each sample, the length of the main valley, the geomorphological space-time theory and the Strahler classification method.

10. The method for landform stage division and DEM construction based on space-time indicators according to claim 1 is characterized by: In step S7, the conditional generative adversarial network Pix2Pix includes a generator and a discriminator; the generator adopts a U-Net structure, which is composed of an encoder and a decoder, the encoder is used to extract image features, and the decoder is used to generate an output image; The discriminator adopts the PatchGAN structure to perform local comparison and discrimination between the generated results and the real images.

Images