A high-resolution digital elevation model generation processing method

Abstract

The application relates to the technical field of digital elevation model generation, and proposes a high-resolution digital elevation model generation processing method, which comprises the following steps: collecting a low-resolution digital elevation model of a to-be-generated region of a high-resolution digital elevation model, a high-resolution digital elevation model of a region with the same address type as the to-be-generated region and a low-resolution digital elevation model, and constructing a training data set; establishing a characteristic value loss function and taking the same as a loss function; training a convolutional neural network using the training data set, obtaining a feature map extraction neural network, using the feature map extraction neural network, obtaining a feature map of the to-be-generated region, and dividing the feature data block; determining a feature difference function, taking the feature difference function as an objective function of a least square method, obtaining reconstructed high-resolution digital elevation data of the to-be-generated region using the least square method, and obtaining a reconstructed digital elevation model of the to-be-generated region. The application aims to generate a more accurate high-resolution digital elevation model.

Description

Technical Field

[0001] This application relates to the field of digital elevation model generation technology, specifically to a method for generating and processing high-resolution digital elevation models. Background Technology

[0002] A high-resolution digital elevation model (DEM) is a spatial data model based on geographic coordinates that stores elevation information in a regular or irregular grid format to represent the topographic surface morphology of the Earth's surface or a specific region. Compared to low-resolution DEMs, high-resolution DEMs can capture more subtle geomorphic features, such as ridges, valleys, and riverbeds. However, high-resolution DEMs have higher equipment costs, time costs, and data acquisition costs.

[0003] To obtain high-resolution digital elevation models (DEMs) at a lower cost, neural networks can be used to reconstruct low-resolution DEMs to obtain the corresponding high-resolution DEMs. However, existing techniques do not consider the impact of terrain features in the DEM on the reconstructed high-resolution DEM during the neural network training process, often resulting in poor reconstruction results. Summary of the Invention

[0004] This application provides a high-resolution digital elevation model (DEM) generation and processing method to solve the problem of poor reconstruction results caused by not considering the influence of digital terrain features on the reconstruction process when reconstructing a high-resolution DEM from a low-resolution DEM. The specific technical solution adopted is as follows:

[0005] One embodiment of this application provides a method for generating and processing high-resolution digital elevation models, the method comprising the following steps:

[0006] Collect low-resolution digital elevation models of the area to be generated from the high-resolution digital elevation model, as well as high-resolution and low-resolution digital elevation models of a preset number of areas with the same geological type as the area to be generated. Based on the high-resolution and low-resolution digital elevation models of all areas corresponding to the same area, obtain gradient maps, curvature maps, and valley line elevation maps, divide the data into blocks, and construct a training dataset.

[0007] Based on the training dataset, an eigenvalue loss function is established. The eigenvalue loss function is used as the loss function. The convolutional neural network is trained using the training dataset to obtain a feature map extraction neural network. The feature map extraction neural network is used to obtain the gradient map, curvature map, and valley line elevation map of the low-resolution digital elevation model of the region to be generated. All feature maps of the region to be generated are divided into equally sized feature data blocks, where each feature data block corresponds to a data block.

[0008] Based on the feature data blocks contained in the feature map and the elevation values ​​of the pixels within the corresponding data blocks, the types of feature data blocks are determined, and the gradient weight, curvature weight, and elevation value weight of the feature data blocks are assigned values ​​respectively. Based on the difference between the low-resolution digital elevation model of the region to be generated and the high-resolution digital elevation model generated from the low-resolution digital elevation model, as well as the gradient weight, curvature weight, and elevation value weight of the feature data blocks, the feature difference function is determined. The feature difference function is used as the objective function of the least squares method, and the least squares method is used to obtain the reconstructed high-resolution digital elevation data of the region to be generated. The digital elevation model is generated based on the reconstructed high-resolution digital elevation data, and the reconstructed digital elevation model of the region to be generated is obtained.

[0009] Furthermore, the methods for obtaining the gradient map, curvature map, and valley line elevation map are as follows:

[0010] Gradient plots, curvature plots, and elevation plots of the high-resolution digital elevation model and the low-resolution digital elevation model corresponding to the same region are obtained respectively.

[0011] Valley lines are obtained from both the high-resolution digital elevation model (DEM) and the low-resolution digital elevation model (DEM) for each corresponding region. Pixels not included in the valley lines in the elevation map and those included in the valley lines in the high-resolution DEM are identified. The elevation values ​​of the pixels not included in the identified valley lines in the elevation map are set to 0, while the elevation values ​​of the pixels included in the valley lines in both the elevation map and the high-resolution DEM are retained, thus obtaining the valley line elevation map of the high-resolution DEM. Similarly, the elevation values ​​of the pixels not included in the identified valley lines in the elevation map are set to 0, while the elevation values ​​of the pixels included in the valley lines in both the low-resolution DEM and the low-resolution DEM are retained, thus obtaining the valley line elevation map of the low-resolution DEM.

[0012] Furthermore, the method for obtaining the data block and the training dataset is as follows:

[0013] The gradient plots, curvature plots, and valley line elevation plots of all high-resolution and low-resolution digital elevation models corresponding to the same region are divided into data blocks of a preset size.

[0014] The dataset consisting of all data blocks is denoted as the training dataset.

[0015] Furthermore, the eigenvalue loss function is specifically as follows:

[0016]

[0017] In the formula, L represents the eigenvalue loss function; n represents the number of pixels contained in the gradient map of the high-resolution digital elevation model; G i This represents the gradient value of the i-th pixel in the gradient map of a high-resolution digital elevation model. This represents the gradient value of the i-th pixel in the gradient map of a low-resolution digital elevation model, corresponding to the pixel in the corresponding feature map; G max C represents the maximum gradient value of all pixels in the gradient map of a high-resolution digital elevation model. i The curvature value of the i-th pixel in the curvature map of a high-resolution digital elevation model; C represents the gradient value of the i-th pixel in the curvature map of a low-resolution digital elevation model, corresponding to the pixel in the feature map. max T represents the maximum curvature value of all pixels in the curvature map of a high-resolution digital elevation model. i This represents the elevation value of the i-th pixel in the valley line elevation map of a high-resolution digital elevation model. T represents the gradient value of the i-th pixel in the valley line elevation map of a low-resolution digital elevation model, corresponding to the pixel in the corresponding feature map; max This represents the maximum elevation value of all pixels in the valley line elevation map of a high-resolution digital elevation model.

[0018] Furthermore, the specific methods for training the convolutional neural network using the training dataset to obtain the feature map extraction neural network are as follows:

[0019] The training dataset was divided into a training set, a validation set, and a test set in a 7:2:1 ratio. A convolutional neural network was trained to obtain a feature map extraction neural network. Specifically, all data blocks in the gradient map, curvature map, and valley line elevation map corresponding to each low-resolution digital elevation model in the training dataset were used as inputs to the feature map extraction neural network. The feature map extraction neural network output the feature maps corresponding to the gradient map, curvature map, and valley line elevation map of the low-resolution digital elevation model, respectively.

[0020] Furthermore, the specific method for determining the type of feature data block based on the feature data blocks contained in the feature map and the elevation values ​​of the pixels within the corresponding data blocks includes:

[0021] Calculate the gradient of all pixels within the feature data block of the feature map, and record the ratio of the average gradient of all pixels within the feature data block to the standard deviation as the gradient variation coefficient of the feature data block.

[0022] Calculate the curvature of all pixels within the feature data block of the feature map, and record the skewness of the curvature of all pixels within the feature data block of the feature map as the curvature skewness of the feature data block.

[0023] Determine the data block corresponding to the feature data block. The number of pixels with non-zero elevation values ​​contained in the data block corresponding to the feature data block is denoted as the non-zero elevation number of the feature data block. The total number of all pixels contained in the data block corresponding to the feature data block is denoted as the total elevation number of the feature data block. The ratio of the non-zero elevation number of the feature data block to the total elevation number is denoted as the valley line density of the feature data block.

[0024] The type of feature data block is determined based on its gradient variation coefficient, curvature skewness, and valley line density.

[0025] Furthermore, the specific method for determining the type of feature data block based on its gradient variation coefficient, curvature skewness, and valley line density includes:

[0026] Feature data blocks with a gradient variation coefficient greater than 0.7 are labeled as steep slopes;

[0027] Feature data blocks with a gradient variation coefficient less than 0.7, curvature skewness less than -0.4, and valley line density greater than 0.35 are labeled as river confluence areas.

[0028] Feature data blocks with a gradient variation coefficient less than 0.4, curvature skewness less than -0.3, and valley line density greater than 0.1 and less than 0.35 are labeled as gentle valleys.

[0029] Feature data blocks with a gradient variation coefficient greater than 0.4 and less than 0.6, a curvature skewness greater than 0.5, and a valley line density less than 0.05 are labeled as ridges;

[0030] Mark all other feature data blocks as ordinary regions.

[0031] Furthermore, the specific methods for assigning values ​​to the gradient weights, curvature weights, and elevation weights of the feature data blocks are as follows:

[0032] When the feature data block is labeled as a steep slope, the gradient weight, curvature weight, and elevation weight corresponding to the gradient variation coefficient, curvature skewness, and valley linear density of the feature data block are assigned to 0.7, 0.2, and 0.1, respectively.

[0033] When the type of feature data block is a river confluence area, the gradient weight, curvature weight, and elevation weight corresponding to the gradient variation coefficient, curvature skewness, and valley linear density of the feature data block are assigned values ​​of 0.2, 0.2, and 0.6, respectively.

[0034] When the feature data block is labeled as a gentle valley, the gradient weight, curvature weight, and elevation weight corresponding to the gradient variation coefficient, curvature skewness, and valley line density of the feature data block are assigned to 0.1, 0.45, and 0.45, respectively.

[0035] When the feature data block is labeled as a ridge, the gradient weight, curvature weight, and elevation weight corresponding to the gradient variation coefficient, curvature skewness, and valley linear density of the feature data block are assigned to 0.45, 0.45, and 0.1, respectively.

[0036] When the feature data block is labeled as a normal region, the gradient weight, curvature weight, and elevation weight corresponding to the gradient variation coefficient, curvature skewness, and valley line density of the feature data block are assigned to 0.33, 0.33, and 0.34, respectively.

[0037] Furthermore, the feature difference function is specifically as follows:

[0038]

[0039] In the formula, The argmin{} represents the feature difference function; the argmin function represents the minimum value function. ω represents the L2 norm; X represents the low-resolution digital elevation model of the region to be generated; Y represents the reconstructed digital elevation model of the region to be generated; Y↓ represents the reconstructed digital elevation model of the region to be generated after downsampling; u,1 ω represents the gradient weight of the u-th feature data block; u,2 ω represents the curvature weight of the u-th feature data block; u,3 G represents the elevation value weight of the u-th feature data block; N represents the number of data blocks divided by the gradient map of the low-resolution digital elevation model of the region to be generated; u,1 G represents the u-th feature data block in the gradient plot of the low-resolution digital elevation model of the region to be generated; u,2 G represents the u-th feature data block in the curvature map of the low-resolution digital elevation model of the region to be generated; u,3 The u-th feature data block in the valley line elevation map of the low-resolution digital elevation model of the region to be generated; The u-th feature data block in the gradient plot of the reconstructed digital elevation model of the region to be generated; The u-th feature data block in the curvature plot of the reconstructed digital elevation model of the region to be generated; This represents the u-th feature data block in the valley line elevation map of the reconstructed digital elevation model of the region to be generated.

[0040] Furthermore, the specific method for using the feature difference function as the objective function of the least squares method to obtain the reconstructed high-resolution digital elevation data of the region to be generated includes:

[0041] The feature difference function is used as the objective function of the least squares method. The least squares method is used to find the optimal solution so that the difference between the downsampling result of the reconstructed digital elevation model of the area to be generated and the low-resolution digital elevation model of the area to be generated is minimized, and the reconstructed high-resolution digital elevation data of the area to be generated is obtained.

[0042] The beneficial effects of this application are:

[0043] This application collects high-resolution and low-resolution digital elevation models (DEMs) of regions with the same geological type as the region to be generated. Based on the corresponding high- and low-resolution DEMs of regions with the same geological type, gradient maps, curvature maps, and valley line elevation maps are extracted. The gradient maps, curvature maps, and valley line elevation maps extract features from the corresponding high- and low-resolution DEMs of the same region, respectively, based on the spatial intensity changes, morphological distribution patterns, and topographic connectivity caused by the terrain slope of the geological type of the region to be generated. Then, loss functions are constructed based on gradient loss, curvature loss, and valley line elevation loss, respectively. A convolutional neural network is trained using the training dataset and the loss functions to obtain a feature map extraction neural network. The network can extract feature maps such as gradient map, curvature map, and valley line elevation map of the low-resolution digital elevation model of the area to be generated. Further, based on the features extracted from the feature maps, the geological features of each feature data block are analyzed to determine the type of feature data block. Different weights are assigned to the differences between each feature data block and its corresponding data block according to the type of feature data block. The least squares method is used to find the optimal value and obtain the reconstructed high-resolution digital elevation data. Finally, a digital elevation model is generated based on the reconstructed high-resolution digital elevation data to obtain the reconstructed digital elevation model of the area to be generated. This solves the problem of poor reconstruction results caused by not considering the influence of digital terrain features on the reconstruction process when reconstructing a high-resolution digital elevation model from a low-resolution digital elevation model. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1A schematic flowchart of a high-resolution digital elevation model generation and processing method provided in one embodiment of this application;

[0046] Figure 2 This is a flowchart illustrating the data block acquisition process provided in one embodiment of this application. Detailed Implementation

[0047] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0048] Please see Figure 1 The diagram illustrates a flowchart of a high-resolution digital elevation model generation and processing method according to an embodiment of this application. The method includes the following steps:

[0049] Step S001: Collect the low-resolution digital elevation model of the area to be generated from the high-resolution digital elevation model, and a preset number of high-resolution digital elevation models and low-resolution digital elevation models of areas with the same geological type as the area to be generated. Based on the high-resolution digital elevation models and low-resolution digital elevation models of all areas corresponding to the same area, obtain gradient maps, curvature maps, and valley line elevation maps, divide the data into blocks, and construct a training dataset.

[0050] Using the USGS website, low-resolution digital elevation models of the area to be generated from the high-resolution digital elevation model are extracted, and multiple high-resolution and low-resolution digital elevation models corresponding to the same area are collected.

[0051] It should be noted that the geological type of the region corresponding to the high-resolution digital elevation model and the low-resolution digital elevation model should be the same as that of the region to be generated in the high-resolution digital elevation model.

[0052] Preferably, in one embodiment of this application, a total of 500 high-resolution digital elevation models and low-resolution digital elevation models of regions with the same geological type as the region to be generated are collected; in this embodiment, the geological type of all regions collected and the region to be generated is valley.

[0053] Compared to other geological types, valley topography has more distinct landform features. The overall shape of a valley is a long, narrow depression, with ridges on both sides and complex water flow paths. The valley slopes are steep with abrupt changes in gradient. The valley floor is narrow and relatively flat, and usually contains natural landforms such as streams, presenting a concave shape. The ridges on both sides of the valley generally gradually widen and thicken from top to bottom, exhibiting convex curved surfaces due to deposition. The valley contains a main river valley and a direct current converging to form a continuous dendritic network.

[0054] For these morphological features of valleys, gradient features are used to describe their slope changes, which can accurately quantify the spatial intensity changes caused by the terrain slope and effectively characterize the abrupt changes brought about by the transition between valley ridges and valley floors. Curvature features can accurately describe the morphological features of naturally formed concave valley floors and convex ridges, distinguish and quantify their concave and convex shape distribution patterns. Based on the valley line features, the complete framework of the entire valley water system network is constrained, which can ensure the terrain connectivity during the reconstruction process and solve the problems of small DC breaks and destruction of the continuous topological structure of valleys that are prone to occur in low-resolution digital elevation models.

[0055] The Sobel operator was used to obtain gradient maps of the high-resolution digital elevation model (DEM) and low-resolution digital elevation model (DEM) for each corresponding region. ArcGIS tools were used to obtain curvature maps of the high-resolution and low-resolution DEMs for each corresponding region. Valley lines in the high-resolution and low-resolution DEMs for each corresponding region were obtained using hydrological analysis. Elevation maps corresponding to the high-resolution and low-resolution DEMs were obtained using GIS software. Pixels not included in the valley lines of the high-resolution DEM and those included in the elevation maps were identified. The elevation values ​​of the pixels not included in the elevation maps were set to 0, while the elevation values ​​of the pixels included in the valley lines of the high-resolution DEMs were retained. This yielded the valley line elevation map for the high-resolution DEM. Similarly, the elevation values ​​of the pixels not included in the elevation maps were set to 0, while the elevation values ​​of the pixels included in the valley lines of the low-resolution DEMs were retained. This yielded the valley line elevation map for the low-resolution DEM.

[0056] At this point, gradient maps, curvature maps, and valley line elevation maps of the high-resolution and low-resolution digital elevation models corresponding to the same region are obtained.

[0057] Gradient maps, curvature maps, and valley line elevation maps of all high-resolution and low-resolution digital elevation models corresponding to the same region are divided into equally sized data blocks, and the dataset composed of all data blocks is denoted as the training dataset.

[0058] Preferably, in one embodiment of this application, each data block is a 128*128 data block, and the data block acquisition flowchart is as follows. Figure 2 As shown.

[0059] At this point, the training dataset and the low-resolution digital elevation model of the region to be generated are obtained.

[0060] Step S002: Establish an eigenvalue loss function based on the training dataset, use the eigenvalue loss function as the loss function, train the convolutional neural network using the training dataset, obtain the feature map extraction neural network, use the feature map extraction neural network to obtain the gradient map, curvature map and valley line elevation map of the low-resolution digital elevation model of the region to be generated, and divide all feature maps of the region to be generated into equally sized feature data blocks, where each feature data block corresponds to a data block.

[0061] Based on the training dataset, an eigenvalue loss function is established. Specifically, the eigenvalue loss function L is:

[0062]

[0063] In the formula, L represents the eigenvalue loss function; n represents the number of pixels contained in the gradient map of the high-resolution digital elevation model; G i This represents the gradient value of the i-th pixel in the gradient map of a high-resolution digital elevation model. This represents the gradient value of the i-th pixel in the gradient map of a low-resolution digital elevation model, corresponding to the pixel in the corresponding feature map; G max C represents the maximum gradient value of all pixels in the gradient map of a high-resolution digital elevation model. i The curvature value of the i-th pixel in the curvature map of a high-resolution digital elevation model; C represents the gradient value of the i-th pixel in the curvature map of a low-resolution digital elevation model, corresponding to the pixel in the feature map. max T represents the maximum curvature value of all pixels in the curvature map of a high-resolution digital elevation model. i This represents the elevation value of the i-th pixel in the valley line elevation map of a high-resolution digital elevation model. T represents the gradient value of the i-th pixel in the valley line elevation map of a low-resolution digital elevation model, corresponding to the pixel in the corresponding feature map; max This represents the maximum elevation value of all pixels in the valley line elevation map of a high-resolution digital elevation model.

[0064] Among them, the predicted values ​​of pixel gradient, curvature, and elevation are the values ​​of the corresponding pixels in the feature maps of the gradient map, curvature map, and valley line elevation map of the low-resolution digital elevation model, respectively.

[0065] Understandably, the gradient map, curvature map, and valley elevation map of a high-resolution digital elevation model contain the same number of pixels; the eigenvalue loss function consists of three parts: gradient loss, curvature loss, and valley elevation loss. Specifically, Corresponding to gradient loss, Corresponding curvature loss, The elevation loss corresponding to the valley line.

[0066] The eigenvalue loss function is used as the loss function. The convolutional neural network (CNN) is trained using the training dataset to obtain feature maps for feature extraction. Specifically, the training dataset is divided into training, validation, and test sets in a 7:2:1 ratio for training. All data blocks in the gradient map, curvature map, and valley line elevation map corresponding to each low-resolution digital elevation model in the training dataset are used as inputs to the CNN. The CNN outputs the feature maps corresponding to the gradient map, curvature map, and valley line elevation map of the low-resolution digital elevation model. Adjacent convolutional layers in the CNN are connected by activation layers using the ReLU function. The initial learning rate is set to 0.001, and a learning rate decay strategy is adopted, reducing the learning rate to one-fifth of its original value every 1000 training epochs.

[0067] The process of training a convolutional neural network is a well-known technique and will not be described in detail here.

[0068] Understandably, by inputting all data blocks corresponding to the gradient map of a low-resolution digital elevation model (DEM) into a feature map extraction neural network, the feature map corresponding to the gradient map of the DEM can be obtained; similarly, by inputting all data blocks corresponding to the curvature map of a low-resolution DEM into a feature map extraction neural network, the feature map corresponding to the curvature map can be obtained; and similarly, by inputting all data blocks corresponding to the valley line elevation map of a low-resolution DEM into a feature map extraction neural network, the feature map corresponding to the valley line elevation map can be obtained. The pixel value of each pixel in the feature maps corresponding to the gradient map, curvature map, and valley line elevation map of the low-resolution DEM is used to measure the changes in gradient, curvature, and elevation value of the pixel, respectively.

[0069] Based on the low-resolution digital elevation model of the region to be generated, the gradient map, curvature map, and valley line elevation map of the low-resolution digital elevation model of the region to be generated are obtained. The gradient map, curvature map, and valley line elevation map of the low-resolution digital elevation model of the region to be generated are input into the feature map extraction neural network to obtain the feature maps of the gradient map, curvature map, and valley line elevation map of the region to be generated, respectively.

[0070] The feature maps of the gradient map, curvature map, and valley line elevation map of the region to be generated are divided into equally sized feature data blocks, and each feature data block corresponds to a data block of the low-resolution digital elevation model of the region to be generated.

[0071] At this point, the feature maps of the gradient map, curvature map, and valley line elevation map of the region to be generated, as well as the feature data blocks of all feature maps, are obtained.

[0072] Step S003: Based on the feature data blocks contained in the feature map and the elevation values ​​of the pixels within the corresponding data blocks, determine the type of feature data blocks, and assign values ​​to the gradient weight, curvature weight, and elevation value weight of the feature data blocks respectively. Based on the difference between the low-resolution digital elevation model of the region to be generated and the high-resolution digital elevation model generated from the low-resolution digital elevation model, as well as the gradient weight, curvature weight, and elevation value weight of the feature data blocks, determine the feature difference function. Use the feature difference function as the objective function of the least squares method, use the least squares method to obtain the reconstructed high-resolution digital elevation data of the region to be generated, generate the digital elevation model based on the reconstructed high-resolution digital elevation data, and obtain the reconstructed digital elevation model of the region to be generated.

[0073] Based on the feature data blocks contained in the feature map and the elevation values ​​of the pixels within the corresponding data blocks, the gradient variation coefficient, curvature skewness, and valley line density of the feature data blocks are determined respectively.

[0074] Calculate the gradient of all pixels within the feature data block of the feature map, and record the ratio of the average gradient to the standard deviation of all pixels within the feature data block as the gradient variation coefficient of the feature data block. Calculate the curvature of all pixels within the feature data block of the feature map, and record the skewness of the curvature of all pixels within the feature data block as the curvature skewness of the feature data block. Since each feature data block corresponds to a data block in the low-resolution digital elevation model of the region to be generated, determine the data block corresponding to the feature data block. Record the number of pixels with non-zero elevation values ​​within the data block corresponding to the feature data block as the non-zero elevation count of the feature data block, record the total number of pixels within the data block corresponding to the feature data block as the total elevation count of the feature data block, and record the ratio of the non-zero elevation count of the feature data block to the total elevation count as the valley line density of the feature data block.

[0075] The type of feature data block is determined based on its gradient variation coefficient, curvature skewness, and valley line density.

[0076] Feature data blocks with a gradient coefficient of variation greater than 0.7 are labeled as steep slopes; feature data blocks with a gradient coefficient of variation less than 0.7, curvature skewness less than -0.4, and valley line density greater than 0.35 are labeled as river confluences; feature data blocks with a gradient coefficient of variation less than 0.4, curvature skewness less than -0.3, and valley line density greater than 0.1 and less than 0.35 are labeled as gentle valley bottoms; feature data blocks with a gradient coefficient of variation greater than 0.4 and less than 0.6, curvature skewness greater than 0.5, and valley line density less than 0.05 are labeled as ridges; all other feature data blocks are labeled as ordinary regions.

[0077] Based on the type of feature data block, as well as the gradient variation coefficient, curvature skewness, and valley line density of the feature data block, the gradient weight, curvature weight, and elevation weight of the feature data block are assigned values ​​respectively.

[0078] When the feature data block is labeled as a steep slope, the gradient weights, curvature weights, and elevation weights corresponding to the gradient variation coefficient, curvature skewness, and valley line density are assigned to 0.7, 0.2, and 0.1, respectively; when the feature data block is labeled as a river confluence, the gradient weights, curvature weights, and elevation weights corresponding to the gradient variation coefficient, curvature skewness, and valley line density are assigned to 0.2, 0.2, and 0.6, respectively; when the feature data block is labeled as a gentle valley floor, the gradient weights, curvature skewness, and valley line density are assigned to 0.7, 0.2, and 0.1, respectively. The corresponding gradient weights, curvature weights, and elevation weights are assigned to 0.1, 0.45, and 0.45, respectively. When the feature data block is labeled as a ridge, the gradient weights, curvature weights, and elevation weights corresponding to the gradient variation coefficient, curvature skewness, and valley line density of the feature data block are assigned to 0.45, 0.45, and 0.1, respectively. When the feature data block is labeled as a normal region, the gradient weights, curvature weights, and elevation weights corresponding to the gradient variation coefficient, curvature skewness, and valley line density of the feature data block are assigned to 0.33, 0.33, and 0.34, respectively.

[0079] Based on the differences between the low-resolution digital elevation model (DEM) of the region to be generated and the high-resolution DEM generated from the low-resolution DEM, as well as the gradient weights, curvature weights, and elevation value weights of the feature data blocks, a feature difference function is determined. Specifically, the feature difference function is:

[0080]

[0081] In the formula, The argmin function represents the characteristic difference function; argmin{} represents the minimum value function argmin. ω represents the L2 norm; X represents the low-resolution digital elevation model of the region to be generated; Y represents the reconstructed digital elevation model of the region to be generated; Y↓ represents the downsampled reconstructed digital elevation model of the region to be generated, which has the same resolution as the low-resolution digital elevation model of the region to be generated; u,1 ω represents the gradient weight of the u-th feature data block; u,2 ω represents the curvature weight of the u-th feature data block; u,3 G represents the elevation value weight of the u-th feature data block; N represents the number of data blocks divided by the gradient map of the low-resolution digital elevation model of the region to be generated; u,1 G represents the u-th feature data block in the gradient plot of the low-resolution digital elevation model of the region to be generated; u,2 G represents the u-th feature data block in the curvature map of the low-resolution digital elevation model of the region to be generated; u,3 The u-th feature data block in the valley line elevation map of the low-resolution digital elevation model of the region to be generated; The u-th feature data block in the gradient plot of the reconstructed digital elevation model of the region to be generated; The u-th feature data block in the curvature plot of the reconstructed digital elevation model of the region to be generated; This represents the u-th feature data block in the valley line elevation map of the reconstructed digital elevation model of the region to be generated.

[0082] The feature difference function is used as the objective function of the least squares method. The least squares method is used to find the optimal solution. The feature map of the region to be generated is inverted and optimized to minimize the difference between the downsampling result of the reconstructed digital elevation model of the region to be generated and the low-resolution digital elevation model of the region to be generated, so as to obtain the reconstructed high-resolution digital elevation data of the region to be generated.

[0083] ArcGIS is used to process the reconstructed high-resolution digital elevation data of the area to be generated, and the reconstructed digital elevation model of the area to be generated is obtained. The reconstructed digital elevation model of the area to be generated is the high-resolution digital elevation model of the area to be generated based on the low-resolution digital elevation model of the area to be generated.

[0084] ArcGIS is a geographic information system software platform that can be used to generate high-resolution digital elevation models from digital elevation data. The use of ArcGIS to generate high-resolution digital elevation models and the use of the least squares method to find the optimal solution are well-known technologies and will not be elaborated further.

[0085] This completes the generation of a high-resolution digital elevation model.

[0086] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A high-resolution digital elevation model generation processing method characterized by, The method includes the following steps: Collect low-resolution digital elevation models of the area to be generated from the high-resolution digital elevation model, as well as high-resolution and low-resolution digital elevation models of a preset number of areas with the same geological type as the area to be generated. Based on the high-resolution and low-resolution digital elevation models of all areas corresponding to the same area, obtain gradient maps, curvature maps, and valley line elevation maps, divide the data into blocks, and construct a training dataset. Based on the training dataset, an eigenvalue loss function is established. The eigenvalue loss function is used as the loss function. The convolutional neural network is trained using the training dataset to obtain a feature map extraction neural network. The feature map extraction neural network is used to obtain the gradient map, curvature map, and valley line elevation map of the low-resolution digital elevation model of the region to be generated. All feature maps of the region to be generated are divided into equally sized feature data blocks, where each feature data block corresponds to a data block. Based on the feature data blocks contained in the feature map and the elevation values ​​of the pixels within the corresponding data blocks, the types of feature data blocks are determined, and the gradient weight, curvature weight, and elevation value weight of the feature data blocks are assigned values ​​respectively. Based on the difference between the low-resolution digital elevation model of the region to be generated and the high-resolution digital elevation model generated from the low-resolution digital elevation model, as well as the gradient weight, curvature weight, and elevation value weight of the feature data blocks, the feature difference function is determined. The feature difference function is used as the objective function of the least squares method, and the least squares method is used to obtain the reconstructed high-resolution digital elevation data of the region to be generated. The digital elevation model is generated based on the reconstructed high-resolution digital elevation data, and the reconstructed digital elevation model of the region to be generated is obtained.

2. The method of claim 1, wherein The methods for obtaining the gradient map, curvature map, and valley line elevation map are as follows: Gradient plots, curvature plots, and elevation plots of the high-resolution digital elevation model and the low-resolution digital elevation model corresponding to the same region are obtained respectively. Valley lines were obtained from both the high-resolution and low-resolution digital elevation models corresponding to the same region. Identify the pixels not included in the valley line in the elevation map and the pixels included in the valley line in the high-resolution digital elevation model. Set the elevation value of the pixels not included in the elevation map to 0, and retain the elevation values ​​of the pixels included in the valley line in the elevation map and the high-resolution digital elevation model to obtain the valley line elevation map of the high-resolution digital elevation model. Set the elevation value of the pixels not included in the elevation map to 0, and retain the elevation values ​​of the pixels included in the valley line in the low-resolution digital elevation model to obtain the valley line elevation map of the low-resolution digital elevation model.

3. The method of claim 2, wherein The method for obtaining the data block and the training dataset is as follows: The gradient plots, curvature plots, and valley line elevation plots of all high-resolution and low-resolution digital elevation models corresponding to the same region are divided into data blocks of a preset size. The dataset consisting of all data blocks is denoted as the training dataset.

4. The method of claim 1, wherein The eigenvalue loss function is specifically as follows: In the formula, L represents the eigenvalue loss function; n represents the number of pixels contained in the gradient map of the high-resolution digital elevation model; G i This represents the gradient value of the i-th pixel in the gradient map of a high-resolution digital elevation model. This represents the gradient value of the i-th pixel in the gradient map of a low-resolution digital elevation model, corresponding to the pixel in the corresponding feature map; G max C represents the maximum gradient value of all pixels in the gradient map of a high-resolution digital elevation model. i The curvature value of the i-th pixel in the curvature map of a high-resolution digital elevation model; C represents the gradient value of the i-th pixel in the curvature map of a low-resolution digital elevation model, corresponding to the pixel in the feature map. max T represents the maximum curvature value of all pixels in the curvature map of a high-resolution digital elevation model. i This represents the elevation value of the i-th pixel in the valley line elevation map of a high-resolution digital elevation model. T represents the gradient value of the i-th pixel in the valley line elevation map of a low-resolution digital elevation model, corresponding to the pixel in the corresponding feature map; max This represents the maximum elevation value of all pixels in the valley line elevation map of a high-resolution digital elevation model.

5. The method of claim 1, wherein, The specific methods for training a convolutional neural network using a training dataset to obtain a feature map extraction neural network are as follows: The training dataset was divided into a training set, a validation set, and a test set in a 7:2:1 ratio. A convolutional neural network was trained to obtain a feature map extraction neural network. Specifically, all data blocks in the gradient map, curvature map, and valley line elevation map corresponding to each low-resolution digital elevation model in the training dataset were used as inputs to the feature map extraction neural network. The feature map extraction neural network output the feature maps corresponding to the gradient map, curvature map, and valley line elevation map of the low-resolution digital elevation model, respectively.

6. The method of claim 1, wherein, The method for determining the type of feature data block based on the feature data blocks contained in the feature map and the elevation values ​​of the pixels within the corresponding data blocks includes the following specific methods: Calculate the gradient of all pixels within the feature data block of the feature map, and record the ratio of the average gradient of all pixels within the feature data block to the standard deviation as the gradient variation coefficient of the feature data block. Calculate the curvature of all pixels within the feature data block of the feature map, and record the skewness of the curvature of all pixels within the feature data block of the feature map as the curvature skewness of the feature data block. Determine the data block corresponding to the feature data block. The number of pixels with non-zero elevation values ​​contained in the data block corresponding to the feature data block is denoted as the non-zero elevation number of the feature data block. The total number of all pixels contained in the data block corresponding to the feature data block is denoted as the total elevation number of the feature data block. The ratio of the non-zero elevation number of the feature data block to the total elevation number is denoted as the valley line density of the feature data block. The type of feature data block is determined based on its gradient variation coefficient, curvature skewness, and valley line density.

7. The method of claim 6, wherein The method for determining the type of feature data block based on its gradient variation coefficient, curvature skewness, and valley line density includes the following specific methods: Feature data blocks with a gradient variation coefficient greater than 0.7 are labeled as steep slopes; Feature data blocks with a gradient variation coefficient less than 0.7, curvature skewness less than -0.4, and valley line density greater than 0.35 are labeled as river confluence areas. Feature data blocks with a gradient variation coefficient less than 0.4, curvature skewness less than -0.3, and valley line density greater than 0.1 and less than 0.35 are labeled as gentle valleys. Feature data blocks with a gradient variation coefficient greater than 0.4 and less than 0.6, a curvature skewness greater than 0.5, and a valley line density less than 0.05 are labeled as ridges; Mark all other feature data blocks as ordinary regions.

8. The method of claim 7, wherein, The specific methods for assigning values ​​to the gradient weights, curvature weights, and elevation weights of the feature data blocks are as follows: When the feature data block is labeled as a steep slope, the gradient weight, curvature weight, and elevation weight corresponding to the gradient variation coefficient, curvature skewness, and valley linear density of the feature data block are assigned to 0.7, 0.2, and 0.1, respectively. When the type of feature data block is a river confluence area, the gradient weight, curvature weight, and elevation weight corresponding to the gradient variation coefficient, curvature skewness, and valley linear density of the feature data block are assigned values ​​of 0.2, 0.2, and 0.6, respectively. When the feature data block is labeled as a gentle valley, the gradient weight, curvature weight, and elevation weight corresponding to the gradient variation coefficient, curvature skewness, and valley line density of the feature data block are assigned to 0.1, 0.45, and 0.45, respectively. When the feature data block is labeled as a ridge, the gradient weight, curvature weight, and elevation weight corresponding to the gradient variation coefficient, curvature skewness, and valley linear density of the feature data block are assigned to 0.45, 0.45, and 0.1, respectively. When the feature data block is labeled as a normal region, the gradient weight, curvature weight, and elevation weight corresponding to the gradient variation coefficient, curvature skewness, and valley line density of the feature data block are assigned values ​​of 0.33, 0.33, and 0.34, respectively.

9. The method of claim 8, wherein, The specific feature difference function is: In the formula, denotes the characteristic difference function; artmin{} represents the minimum value function argmin; ω represents the L2 norm; X represents the low-resolution digital elevation model of the region to be generated; Y represents the reconstructed digital elevation model of the region to be generated; Y↓ represents the reconstructed digital elevation model of the region to be generated after downsampling; u,1 ω represents the gradient weight of the u-th feature data block; u,2 ω represents the curvature weight of the u-th feature data block; u,3 G represents the elevation value weight of the u-th feature data block; N represents the number of data blocks divided by the gradient map of the low-resolution digital elevation model of the region to be generated; u,1 The u-th feature data block in the gradient plot of the low-resolution digital elevation model of the region to be generated; G u,2 u-th feature data block in a curvature map representing a low resolution digital elevation model of the region to be generated; G u,3 The u-th feature data block in the valley line elevation map of the low-resolution digital elevation model of the region to be generated; The u-th feature data block in the gradient plot of the reconstructed digital elevation model of the region to be generated; The u-th feature data block in the curvature plot of the reconstructed digital elevation model of the region to be generated; This represents the u-th feature data block in the valley line elevation map of the reconstructed digital elevation model of the region to be generated.

10. The method of claim 9, wherein, The specific method for using the feature difference function as the objective function of the least squares method to obtain the reconstructed high-resolution digital elevation data of the region to be generated using the least squares method includes: The feature difference function is used as the objective function of the least squares method. The least squares method is used to find the optimal solution so that the difference between the downsampling result of the reconstructed digital elevation model of the area to be generated and the low-resolution digital elevation model of the area to be generated is minimized, and the reconstructed high-resolution digital elevation data of the area to be generated is obtained.

Images