Uniform terrain unit division method suitable for high and deep canyon region
Through technical means such as multi-feature terrain preprocessing and superpixel initial segmentation, the problem of unsatisfactory effect in high-deep canyon areas is solved, high-quality acquisition of surface undulating units and terrain digital segmentation is realized, and the terrain segmentation is improved. Accuracy and rationality.
Patent Information
- Application Number
- CN202411003821.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-15
AI Technical Summary
The existing method of uniform terrain unit division is not ideal in high-deep canyon areas, and it is difficult to consider the intermittent terrain structure in a refined manner. It is easy to produce mechanical cutting and semi-closed spatial units, and it is impossible to effectively match the terrain boundaries.
Using multi-feature terrain preprocessing, superpixel initial segmentation, region merging criteria and determining the optimal homogeneity threshold, the image texture structure is obtained through the grayscale symbiosis matrix, and the slope unit is optimized as a statistical window to perform multi-feature terrain normalization and principal component analysis, homogeneous clustering and merging were performed in combination with geometric morphology and texture features to determine the global optimal homogeneity threshold.
It realizes high-quality automatic acquisition of surface undulating units with similar internal and external differentiation in complex wide-area terrain areas, improves the accuracy of digital segmentation and classification of mountain slope terrain, and provides more reasonable homogeneity and heterogeneity expression. .
Smart Images

Figure CN118968294A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of topographic analysis, and in particular to a method for dividing uniform topographic units suitable for high and deep canyon areas. Background Art
[0002] Landslide susceptibility assessment can evaluate the extent of landslides that have occurred in a region and predict the spatial tendency of landslides to occur in the future. Its results can provide important decision-making basis for land use planning, the deployment of geological disaster monitoring and early warning networks, and risk transfer and prevention policies. It is a generally accepted and effective disaster prevention and mitigation method by scholars at home and abroad. The mapping unit is the smallest geographic spatial entity for data extraction and allocation in landslide risk assessment, and is the core link between landslides and disaster-prone environmental factors. Obtaining a suitable spatial expression carrier to accurately reflect the characteristics of landslide-prone environmental factors and clarify the landslide-prone mechanism is crucial for landslide susceptibility prediction. Common mapping unit types mainly include grid units, slope units, uniform terrain units, watershed units, and administrative division units.
[0003] As one of the most important mapping unit types besides grid units and slope units, the basic concept of uniform terrain unit is each relatively independent and homogeneous surface undulation unit in a continuous surface, such as Figure 1 As shown in the figure, a three-dimensional surface is usually formed by one or more predefined homogeneous morphological attributes, and the elements describing the geometric morphology usually include elevation, slope, curvature, slope aspect and slope position. Compared with the traditional slope unit, the uniform terrain unit can finely consider multiple terrain discontinuity structures represented by slope or curvature inside the slope unit, which can avoid neglecting terrain details in small-scale or high-precision application scenarios, and promote further exploration of the intrinsic relationship between disaster-prone factors and landslides and the improvement of the accuracy of landslide susceptibility prediction.
[0004] At present, the development and application of uniform terrain units are not ideal, and the theoretical concepts and construction methods are still in an extremely immature embryonic stage. The traditional method of obtaining uniform terrain units is mainly based on predefined rule knowledge. For example, when the terrain curvature value is higher than a certain threshold, the surface morphology is divided into a concave slope, otherwise it is a straight or convex slope. This method often cannot well consider the spatial topological adjacency relationship of terrain units and the matching of the boundaries of different terrain units with the actual terrain discontinuity by limiting the parameter threshold interval and its combination. Moreover, this strategy of first constructing terrain classification rules and then dividing boundaries is very likely to produce mechanically cut and semi-closed spatial units, which cannot directly serve as mapping units for depicting homogeneous functional spaces, and is prone to produce incomprehensible results in practical applications. Summary of the invention
[0005] In order to overcome the deficiencies of the prior art, the present invention aims to provide a method for dividing uniform terrain units suitable for high and deep canyon areas. Through multi-feature terrain preprocessing, superpixel initial segmentation, regional merging criteria and determination of the optimal homogeneity threshold, it is possible to automatically obtain surface undulation units with similar internal structure and differentiated external structure in complex wide-area terrain areas with high quality, especially providing a breakthrough solution for the digital segmentation and classification of fine-level terrain on mountain slopes.
[0006] The present invention is achieved through the following technical solutions:
[0007] A method for dividing uniform terrain units suitable for high and deep canyon areas comprises the following steps:
[0008] S1: Get the original image;
[0009] S2: The morphological characteristics of different high and low undulating units in the original image are described by slope length, slope and slope shape, the image texture structure is obtained by gray-level co-occurrence matrix, and the ridge line and valley line are used as the terrain skeleton edge information to fuse into the multi-feature terrain attribute field;
[0010] S3: using the optimized slope unit as a statistical window to perform normalization of multi-feature terrain, forming an independent circulation and normalization reorganization of terrain attributes with the optimized slope unit polygon as the spatial range; the slope unit is formed by the spatial intersection of the ridge line and the valley line;
[0011] S4: Convert the original image to CIELAB color space, and then implement homogeneity clustering according to color and distance similarity;
[0012] S5: According to the hierarchical and step-by-step optimization criteria, the adjacent regions with the minimum merging cost are searched one by one. If the increase of the global expression error is the smallest, it is the optimization target and the region merging is implemented;
[0013] S6: After merging, based on the terrain unit scale set formed by over-segmentation, moderate segmentation and under-segmentation, a scale set result with coarse and fine scale boundaries nested and fixed inheritance is formed through a step-by-step scale parameter strategy;
[0014] S7: According to the global homogeneity and global heterogeneity characteristics of terrain units at different threshold scales, a comprehensive heterogeneity index is obtained, and the maximum value of the comprehensive heterogeneity index is taken as the global optimal homogeneity threshold. The optimal selection is made based on the global optimal homogeneity threshold to obtain the division result.
[0015] Furthermore, in step S2, the slope position index LTP is used to describe the slope length, and the calculation formula is:
[0016]
[0017] Where: ZO is the altitude of the point, Z max , Z min Indicates the maximum and minimum values of the altitude in the analysis window respectively. The analysis window is the slope unit vector where the point is located;
[0018] The curvature is used to quantitatively describe the curvature of the slope surface, which can be expressed as profile curvature and plane curvature in vertical and horizontal directions.
[0019] The terrain wetness index TWI is used to reflect the combined changes of different positions, convex and concave shapes, and slope steepness in the slope. The calculation formula is:
[0020]
[0021] Where: TWI is the terrain wetness index, SCA indicates the confluence area per unit contour length flowing through the point (m 2 m -1 ), β is the slope of the point.
[0022] Furthermore, in step S3, the calculation formula for the independent cycle and normalized recombination of the terrain attributes is:
[0023]
[0024] Where: y represents the normalized value, and its range is (-1,1); i Represents the original value; μ, y min and max They respectively represent the average, minimum and maximum values within the statistical area of the slope unit.
[0025] Furthermore, the step S3 further comprises the following steps:
[0026] After normalization and reorganization, data fusion and dimensionality reduction are also performed based on principal component analysis. Through orthogonal transformation, multiple groups of feature data are mapped to a limited number of unrelated principal components, and principal component fusion is implemented separately for geometric morphological attributes and texture structure features. The feature groups with a cumulative contribution rate of more than 90% are taken as the principal component set.
[0027] Furthermore, in step S4, the original image is converted into the CIELAB color space, and then homogeneity clustering is achieved according to color and distance similarity, including the following specific steps:
[0028] Assume that M*N is the size of the image, i is the number of pixels, the initial clustering parameter is set to K; the initial clustering seed point is C i =[l i ,a i ,b i ,x i ,y i] T , where the color space is [l,a,b] T , the pixel position is [x,y] T ; Then sample and generate K superpixel blocks of equal size, with a size of (M*N) / K and a sampling step of
[0029] Taking the cluster seed point as the center, set the search range, calculate the distance D between other clusters and the center, and assign pixel i to the nearest cluster center;
[0030] Calculate the average vector value of all pixels in K superpixels and update the cluster center until the residual is expected to converge to a certain threshold, then stop the iteration.
[0031] Furthermore, the characteristic distance calculation formula of the pixel point LAB color space is:
[0032]
[0033] Where:
[0034]
[0035] Where: d(i,C k ) indicates the characteristic distance of the pixel point in LAB color space, d c (i,C k ), d s (i,C k ) respectively indicate the color and spatial distance represented by the Euclidean distance, and m is the weight ratio of adjusting the color distance and spatial distance in the similarity measurement.
[0036] Furthermore, in step S5, the merging cost is defined as the increased fitting error of the image caused by merging the two regions:
[0037]
[0038] Where: S m , S n Respectively represent the areas of two adjacent regions m and n, D m,n represents the feature distance between two regions, C m,n Indicates the merging cost of two adjacent regions. The smaller the value, the higher the merging priority of the unit compared with other units.
[0039] Furthermore, in step S5, the characteristic distance D m,n It is defined as a combined distance of morphological and textural features weighted based on spatial constraints:
[0040]
[0041] in
[0042] w T =1-w G ;
[0043] Where: G C , G T Respectively represent the geometric shape and texture feature distance; w C 、w T Indicates the weights of geometry and texture distance in the combined distance metric respectively; max(f λ ) represents the maximum frequency of the grayscale histogram of the λth principal component; K represents the weighted maximum frequency of the histogram of each principal component variable in the current region, that is, first quantify the heterogeneity between single terrain attributes, then combine the heterogeneity of each single terrain attribute through adaptive weighting, and finally form the comprehensive heterogeneity of the unit and the adjacent units; l m,n is the common side length of adjacent regions m and n, and α is the influence coefficient of the common side.
[0044] Furthermore, in step S5, the G statistic is used to measure the characteristic distance between two adjacent regions in terms of slope position, slope, curvature or texture. Assume that the terrain attributes of two adjacent regions m and n in the tth time are represented as m t 、n t , then the G statistic is as follows:
[0045]
[0046] Where: G(m,n) represents the G statistic of adjacent regions m and n at the tth attribute; m i t 、n i t They represent the frequency of occurrence of the gray value i of the two histograms respectively; L represents the gray level. Considering the calculation efficiency and expression accuracy, the gray levels of various terrain attributes are uniformly compressed into an 8-bit unsigned integer data format with gray levels of 0 to 255; the value range of G(m,n) is [0,4ln2]. The larger the G statistic value, the greater the histogram distance between two adjacent areas.
[0047] Furthermore, in step S7, the heterogeneity level between the global adjacent units is measured according to the Moran index:
[0048]
[0049] Where: m represents the total number of units; w ih Represents the spatial proximity index. If SU h with SU iIf there is a common edge, the value is 1, otherwise it is 0; i represents the average slope aspect of all internal grid cells. represents the average slope aspect of all cells in the study area;
[0050] The level of homogeneity within the global population is measured in terms of variance:
[0051]
[0052] Where: s i and c i Respectively represent SU i The area and circular variance of the slope units with larger scale can be given higher weights to avoid fluctuations caused by small units;
[0053] The overall level of internal homogeneity and external heterogeneity of multi-feature terrain of global units is measured according to the comprehensive heterogeneity index (GS):
[0054]
[0055] Where: σ indicates the mean of the standard deviation of each principal component, Indicates the average value of all principal components in the entire workspace, θ i Indicates the mean of each principal component in region i, MI max(min) and V max(min) are the maximum and minimum values calculated in the heterogeneity level and homogeneity level, respectively. The GS maximum value corresponds to the global optimal homogeneity threshold.
[0056] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0057] The present invention constructs a homogeneous terrain unit extraction method SSM-HTU (Superpixel Segmentation and Merging-Homogeneous Terrain Unit) that integrates superpixel initial segmentation and geometric morphology-texture approximation merging, which includes four steps: multi-feature terrain preprocessing, superpixel initial segmentation, regional merging criteria, and determining the optimal homogeneity threshold. It can achieve high-quality automatic acquisition of surface undulation units that are similar internally and differentiated externally in complex wide-area terrain areas, and especially provides a breakthrough solution for the digital segmentation and classification of fine-level terrain on mountain slopes. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:
[0059] Figure 1 Definition and schematic diagram of uniform terrain unit ((a) discontinuity structure of terrain unit; (b) geometric morphological attributes of terrain unit; (c) schematic diagram of slope unit; (d) schematic diagram of uniform terrain unit);
[0060] Figure 2 A flow chart of a method for obtaining uniform terrain units based on the SSM-HTU method provided by the present invention;
[0061] Figure 3 Schematic diagram of the principle of the initial segmentation technology based on the superpixel algorithm provided by the present invention ((a) original image; (b) color space transformation; (c) cluster center initialization; (d) iterative clustering; (e) enhanced communication to obtain superpixels);
[0062] Figure 4 Schematic diagram of the hierarchical scale set provided by the present invention ((a) schematic diagram of the upper and lower scale hierarchical connection network; (b) schematic diagram of the terrain unit scale set);
[0063] Figure 5 A spatial location diagram of an application example area provided by the present invention;
[0064] Figure 6 This is an example of the initial segmentation result of the terrain based on the superpixel algorithm provided by the present invention ((a) setting the initial clustering parameter K; (b) setting the color distance and spatial distance weight parameter m);
[0065] Figure 7 An example diagram of superpixel-level principal component features provided by the present invention (PCA1 indicates principal component 1; PCA2 indicates principal component 2; PCA3 indicates principal component 3; PCA indicates principal component 4);
[0066] Figure 8 The comparison diagram of superpixel grayscale (SLIC-PCA) and original pixel grayscale (O-PCA) attributes provided by the present invention ((a) observed profile position; (b) original terrain profile line; (c) PCA1; (d) PCA2; (e) PCA3; (f) PCA4);
[0067] Fig. 9 An example diagram of the segmentation results of uniform terrain units at each scale in the scale set provided by the present invention;
[0068] Fig.10 The regional merging process and global exponential evolution curve provided by the present invention ((a) the number of uniform terrain units in scale concentration; (b) global homogeneity curve; (c) global heterogeneity curve; (d) global comprehensive heterogeneity curve);
[0069] Fig.11 A local example diagram of the SSM-HTU uniform terrain unit in a plateau wide valley area provided by the present invention ((a) remote sensing image of the example area; (b) uniform terrain unit extracted by the SSM-HTU method; (c) on-site photos and terrain discontinuity lines; (d) local example). DETAILED DESCRIPTION
[0070] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.
[0071] Embodiment 1:
[0072] In view of the limitations of traditional uniform terrain unit automatic division methods in terrain adaptation, the present invention adopts an object-oriented processing strategy and data-driven image segmentation theory to propose a new type of uniform terrain unit automatic construction method suitable for high and deep canyon areas. Usually in remote sensing image segmentation scenes, different objects such as farmland, roads, forests, buildings and water bodies have distinct contour boundaries. However, terrain segmentation scenes are not completely equivalent to image segmentation. Because in the boundary identification and segmentation of terrain units, the spatial structure of terrain unit objects with different heights is smooth and fuzzy, and the contour boundaries show weak discontinuity characteristics. How to efficiently integrate multi-dimensional feature terrain and significantly enhance the spatial structural characteristics of homogeneous aggregation and heterogeneous discontinuity is the key technical difficulty that needs to be solved in the present invention to obtain terrain units based on image segmentation theory.
[0073] The present invention constructs a homogeneous terrain unit extraction method SSM-HTU (Superpixel Segmentation and Merging-Homogeneous Terrain Unit) that integrates superpixel initial segmentation and geometric morphology-texture approximation merging, which includes four steps: multi-feature terrain preprocessing, superpixel initial segmentation, region merging criteria, and determining the optimal homogeneity threshold. Figure 2 As shown, it is possible to automatically obtain high-quality surface undulation units that are similar internally and differentiated externally in complex wide-area terrain areas, especially providing a breakthrough solution for the digital segmentation and classification of fine-level terrain on mountain slopes.
[0074] Please refer to 1- Figure 4The detailed invention content of this embodiment 1 includes the following specific steps:
[0075] (1) Multi-feature terrain preprocessing:
[0076] 1) Terrain geometry;
[0077] This paper describes the morphological characteristics of each high and low undulating unit from the perspective of slope length, slope and slope shape. The slope position index (LTP) is used to describe the slope length, that is, the vertical height of the grid pixel in the slope surface, and the calculation method is shown in formula (1). The curvature is used to quantitatively describe the curvature of the slope surface, which is expressed as profile curvature and plane curvature in vertical and horizontal directions. Since the terrain wetness index (TWI) can comprehensively reflect the changes in different positions, convex and concave shapes, and slope steepness in the slope through differences in water collection capacity, it can also be used as an important geomorphological factor, and the calculation method is shown in formula (2).
[0078]
[0079] Where: Z O is the altitude of the point, Z max , Z min They respectively indicate the maximum and minimum values of the altitude within the analysis window, and the analysis window is the slope unit vector where the point is located.
[0080]
[0081] Where: TWI is the terrain wetness index, SCA indicates the confluence area per unit contour length flowing through the point (m 2 m -1 ), β is the slope of the point.
[0082] 2) Terrain texture structure;
[0083] At the same time, texture features can specifically characterize the spatial visual structure formed by the slow and periodic changes in the grayscale attributes in the image. Gray Level Co-occurrence Matrix (GLCM) is a classic and effective method for obtaining image texture structure. It obtains texture information by considering the combination rules of the gray levels of two adjacent pixels. This paper selects 5 groups of the most commonly used texture structure indicators, specifically angular second moment (ASM) and entropy, which indicate texture disorder and order; contrast, which indicates diversity and dominance; correlation and inverse difference moment (IDM), which indicate similarity. For the detailed calculation and construction methods of the above five groups of texture structure indicators, please refer to reference [1]. Due to space limitations, this paper will not elaborate on them; reference [1] is: Texture feature extraction based on gray level co-occurrence matrix [J]. Computer system application.
[0084] 3) Multi-feature terrain normalization preprocessing;
[0085] At the macro level, ridge lines and valley lines are important edge information of terrain skeletons. Fusion of them as important spatial edge information into multi-feature terrain attribute fields is expected to significantly enhance the discontinuity structure between different terrain skeletons. The slope unit is essentially the area formed by the spatial intersection of the terrain feature ridge lines and valley lines. Therefore, this paper uses the optimized slope unit as a statistical window to perform normalization of multi-feature terrain, forming an independent circulation and normalized reorganization of terrain attributes with the optimized slope unit polygon as the spatial range. The calculation method is shown in formula (3):
[0086]
[0087] Where: y represents the normalized value, and its range is (-1,1); i Represents the original value; μ, y min and max They respectively represent the average, minimum and maximum values within the statistical area of the slope unit.
[0088] 4) Data fusion and dimensionality reduction;
[0089] The combination of multiple terrain features can significantly improve the accurate description of the uniform terrain attribute space, but in parallel computing scenarios, too many features will also increase the complexity of the data, especially when there is a strong correlation between the data, it may also cause errors. Therefore, this paper also needs to implement data fusion and dimensionality reduction based on principal component analysis (Principal Component Analysis, PCA), and map multiple sets of feature data to a limited number of independent principal components through orthogonal transformation. Due to the difference in data types, principal component fusion is implemented separately for geometric morphological attributes and texture structure features, and the feature group with a cumulative contribution rate of more than 90% is used as the principal component set. Among them, the dimensionality reduction method based on principal component analysis belongs to the conventional preprocessing technology. For its basic principle and implementation process, please refer to reference [2]; reference [2] is: Dimensionality reduction method based on principal component analysis and kernel independent component analysis [J]. System Engineering and Electronic Technology.
[0090] (2) Initial segmentation based on superpixel algorithm;
[0091] Implementing the initial segmentation preprocessing process based on superpixels can obtain the following significant advantages in terrain segmentation scenarios: 1) Each superpixel will not destroy the original discontinuous boundary information and has strong irregular boundary self-adaptation capabilities. 2) Through the simplified merging process of aggregating adjacent small areas, superpixel blocks can greatly reduce the redundancy of original data; 3) In the process of aggregating single grid pixels into superpixel blocks based on proximity and similarity features, merging and eliminating a large number of local abnormal noise points will also be eliminated; 4) The primary terrain superpixel block is used as the starting point for subsequent area merging, and the merging of adjacent units will also be carried out along the existing superpixel block boundaries.
[0092] The core process of the initial segmentation based on the superpixel algorithm is to convert the original image into the CIELAB color space, and then implement homogeneous clustering according to color and distance similarity: Assume that M*N is the size of the image, i is the number of pixels, and the initial clustering parameter is set to K; the initial clustering seed point is C i =[l i ,a i ,b i ,x i ,y i ] T , where the color space is [l,a,b] T , the pixel position is [x,y] T ; Then sample and generate K superpixel blocks of equal size, with a size of (M*N) / K and a sampling step of With the cluster seed point as the center, the search range is 2S×2S. The distance D between other clusters and the center is calculated according to formula (3), and pixel i is assigned to the nearest cluster center. The average vector value of all pixels in K superpixels is calculated, and the cluster center is updated until the residual is expected to converge to a certain threshold, then the iteration is stopped.
[0093]
[0094] Where:
[0095]
[0096] Where: d(i,C k ) indicates the characteristic distance of the pixel point in LAB color space, d c (i,C k ), d s (i,C k ) indicate the color and spatial distance represented by the Euclidean distance respectively, and m is the weight ratio of adjusting the color distance and spatial distance in the similarity measurement. The larger the m setting is, the more compact and regular superpixels will be formed.
[0097] (3) Merging criteria based on geometry-texture approximation;
[0098] When two adjacent regions are merged to produce a new larger region, the image expression accuracy is bound to be lost. This process can be analogized to the need for data simplification, where the conversion of fine spatial resolution to a coarser spatial resolution will result in image blur, i.e., loss of expression accuracy. Therefore, when two sets of very similar adjacent regions are merged, the cost of image accuracy loss is low. The ability to find the most similar region merge in a multi-feature terrain field is one of the decisive factors for the quality of terrain segmentation. The present invention proposes a regional similarity merging measurement method based on the application scenario of multi-feature remote sensing image surface object segmentation, namely, a geometric morphology-spatial texture similarity merging criterion based on spatial constraints.
[0099] First, according to the hierarchical and step-by-step optimization criteria, the adjacent regions with the smallest merging cost are searched one by one, and the region merging is performed with the smallest increase in global expression error as the optimization goal. The merging cost is defined as the increased fitting error of the image caused by the merging of two regions:
[0100]
[0101] Where: S m , S n Respectively represent the areas of two adjacent regions m and n, D m,n represents the feature distance between two regions (similarity measure), C m,n Indicates the merging cost of two adjacent regions. The smaller the value, the higher the merging priority of the unit compared with other units.
[0102] Feature distance D m,n It is defined as a combined distance of morphological-textural features weighted by spatial constraints:
[0103]
[0104] in
[0105] w T =1-w G (9);
[0106] Where: G C , G T Represent the geometric and texture feature distances, w C 、w T Indicates the weights of geometry and texture distance in the combined distance metric, max(f λ) represents the maximum frequency of the grayscale histogram of the λth principal component. K represents the weighted maximum frequency of the histogram of each principal component variable in the current region, that is, first quantify the heterogeneity between single terrain attributes, then combine the heterogeneity of each single terrain attribute through adaptive weighting, and finally form the comprehensive heterogeneity of the unit and the adjacent units. m,n is the common side length of adjacent regions m and n, α is the influence coefficient of the common side. Under the same feature distance, adjacent units with longer shared common boundaries are encouraged to be merged first, which is conducive to forming a more compact and regular segmentation object. Therefore, α is recommended to be 0.5.
[0107] The G statistic is used to measure the characteristic distance between two adjacent regions in terms of slope position, slope, curvature or texture. Assume that the terrain attributes of two adjacent regions m and n in the tth time are represented as m t 、n t , then the G statistic is as follows:
[0108]
[0109] Where: G(m,n) represents the G statistic of adjacent regions m and n at the tth attribute; m i t 、n i t Respectively represent the frequency of occurrence of the gray value i of the two histograms. L represents the gray level. Considering the calculation efficiency and expression accuracy, the gray level of each terrain attribute is uniformly compressed into an 8-bit unsigned integer data format with gray levels of 0 to 255. The value range of G(m,n) is [0,4ln2]. The larger the G statistic value, the greater the distance between the histograms of two adjacent regions.
[0110] (4) Determine the optimal homogeneity threshold
[0111] This paper constructs the hierarchical scale set results of homogeneous terrain units that evolve step by step. The principle is as follows Figure 3-6 As shown in Figure 2, the detailed continuity of the result includes over-segmentation, moderate segmentation and under-segmentation. The initial segmentation superpixel block is taken as the absolute over-segmentation and is set to the starting scale S 0 , when the number of region merged is less than 1000 slope units, it is regarded as an absolute under-segmentation and is set as the upper limit scale S max , the scale interval is set to 1, and the predefined scale sequence is {S 1 , S 2 ,...,S K ,...,S max} Terrain unit scale set. At the same time, the present invention implements a step-by-step scale parameter strategy (SSPS) to form a scale set result with coarse and fine scale boundaries nested and fixed inheritance. Scale S KThe result is based on the initial segmentation scale S of the superpixel 0 As the starting area for regional merging, the scale is increased by 1 to meet S 1 , and then obtain S through K-scale iterations K The results at higher scales are combinations of fine objects at lower scales, forming a hierarchical network of boundary constraints and superior-subordinate connections.
[0112] In addition, the variance (V) and Moran's index (MI) are used to indicate the global homogeneity and global heterogeneity characteristics of terrain units at different threshold scales. Among them, the Moran's index (MI) can measure the heterogeneity level between adjacent units in the global whole:
[0113]
[0114] Where: m represents the total number of units; w ih Represents the spatial proximity index. If SU h with SU i If there is a common edge, the value is 1, otherwise it is 0; i Represents the average slope of all internal grid cells. Represents the average slope aspect of all cells in the study area
[0115] Variance (V) can measure the level of homogeneity within the global unit:
[0116]
[0117] Where: s i and c i Respectively represent SU i The area and circular variance of the slope units can be assigned higher weights to avoid fluctuations caused by small units.
[0118] The comprehensive heterogeneity index (GS) measures the overall realization level of internal homogeneity and external heterogeneity of multi-feature terrain of the global unit:
[0119]
[0120] Where: σ indicates the mean of the standard deviation of each principal component, Indicates the average value of all principal components in the entire workspace, θ i Indicates the mean of each principal component in region i, MI max(min) and V max(min) are the maximum and minimum values calculated in equation (11) and equation (12), respectively. The maximum value of GS corresponds to the global optimal homogeneity threshold.
[0121] Embodiment 2:
[0122] Example 2 is further limited on the basis of Example 1 and provides a specific implementation case.
[0123] See also Figure 5-Figure 11 ; The sample area is the Yuqu River Basin, a first-level tributary on the left bank of the middle reaches of the Nujiang River. The regional location is as follows Figure 5 As shown, the geographical coordinates are 97°30′~98°30′E, 28°~30°30′N, and the basin area is 5136km 2 , the river is 335km long and 10-45km wide from east to west. The rich diversity of topography is very representative in the deep river valley area in southeastern Tibet. According to the altitude and landform, its topography is divided into plateau wide valley, plateau mountain, canyon transition area and high mountain canyon area. Therefore, the Yuqu River Basin was selected as a demonstration application area for the division and mapping of uniform terrain units. These successful experiences can also have a high reference value in other deep river valley areas in southeastern Tibet.
[0124] (1) DEM and image data;
[0125] The elevation data comes from the ALOS-12m terrain data released by the Advanced Land Observing Satellite-1 project of the Japan Aerospace Exploration Agency, with a resolution of 12.5m. The remote sensing images come from the Gaofen-2 satellite data, acquired on January 20, 2020, with a spatial resolution of 1m. The DEM elevation and images are uniformly processed into a projection coordinate system of WGS_1984_UTM_Zone_47N and imported into ArcScene 10.6.1 to obtain a real three-dimensional terrain perspective. At the same time, a drone (Phantom 4pro V2.0, flying at an altitude of 500m) was used for field surveys to obtain real photos of typical locations.
[0126] (2) Initial segmentation based on superpixels;
[0127] First, the principal component analysis results are shown in Table 1. Three principal components (PCA1, PCA2, PCA3) are selected for terrain geometry features, with a cumulative contribution rate of 91.72%. One principal component PCA4 is selected for texture structure features, with a contribution rate of 90.83%. The geometric morphological principal components PCA1, PCA2, and PCA3 are selected as the three-channel variables of the image R, G, and B. The initial terrain segmentation based on the SLIC algorithm is implemented, and the superpixel blocks with good boundary adaptability and compact and regular morphology are obtained by calibrating the parameters K and m. The example segmentation area is as follows Figure 6 As shown in the figure, the pixel size is 700×700 and the pixel resolution is 12.5m. Based on the existing research, it is suggested that the number of original pixels that each superpixel block should contain is about 100 to 200, and the parameter K is set to (2200 to 3500). Figure 6As shown in a, as the parameter k gradually decreases, the superpixel blocks gradually transition from dense to sparse, especially when the parameter k < 3200, superpixel block objects that cross the boundaries of different terrains begin to appear one after another. Therefore, setting the parameter k = 3200 means that each superpixel contains 153 original pixels. At the same time, it is also necessary to optimize the boundary shape through the shape parameter m, as shown in Figure 6 When m=22, the superpixel block can form a regular and uniform spatial shape, but as the parameter m continues to increase, the boundary shape will be too flat.
[0128]
[0129] Table 1 Contribution rate of main components of terrain geometry and spatial texture
[0130] (3) Multi-dimensional terrain feature preprocessing;
[0131] According to the standard that each superpixel contains 153 pixels, the parameter K is set, and the value of m is 22. The superpixel segmentation based on the SLIC algorithm is performed in the Yuqu River Basin with an area size of 10075×15537, and 9.78×10 6 The grayscale attribute is further expanded from the pixel level to the superpixel level by taking the pixel average, as shown in the figure. Figure 7 As shown in the principal component PCA2, this principal component is mainly related to the slope, and the cultivated land gentle slope platform shown in the red area shows an excellent low slope attribute aggregation state, and at the same time shows a distinct difference discontinuity feature with the steep slope at the edge of the platform. The principal components PCA1 and PCA3 are easily inferred to be related to the slope index and profile curvature, respectively. Especially for PCA3, the superpixel block shows a red minimum value that is approximately a linear closed form at the edge of the ridgeline contour, and there is a distinct boundary discontinuity with the adjacent area.
[0132] In order to further objectively quantify and analyze the effect of the preprocessing of the fusion of coarse and fine edge information on the enhancement of the discontinuous characteristics of the terrain structure, the observation sections AB and CD are intercepted as follows: Figure 8 a. First, there are multiple groups of discontinuous surface undulation units between the ridge line and the valley line according to the elevation change, including steep slopes, gentle slopes, platforms and concave slopes, convex slopes and straight slopes. The attribute changes of the original grid pixels (O-PCA) and superpixel fusion preprocessed pixels (SLIC-PCA) of each principal component at the profile lines AB and CD are shown in the figure. Figure 8 cf. This is particularly evident in the benign transformation of data quality in PCA3, which is related to the profile curvature feature. In the original grid grayscale value, it is extremely chaotic and irregular. However, after superpixel fusion and feature expansion, the grayscale 125 is roughly the dividing line. Above this value corresponds to a convex slope and below corresponds to a concave slope. The interval near grayscale 125 shows a good match with the linear slope. At the same time, Figure 8 As shown in Figure 5, the superpixel block PCA4 that reflects the texture structure also presents better homogeneity aggregation and heterogeneity discontinuity by suppressing abnormal local noise. These multi-source heterogeneous information fusion processes can indeed improve the weak discontinuity restrictions of boundaries and structures in terrain segmentation application scenarios, and lay a good foundation for subsequent accurate regional merging.
[0133] (4) regional proximity merging scale set;
[0134] The superpixel block result is used as the starting scale S 0 It is agreed that the increment interval of homogeneity merging tolerance is 1. When the scale increases to 30, the number of uniform terrain units after regional merging is close to the number of slope units 6500, which is taken as the upper limit of the absolute under-segmentation scale. 1 ~S 30 A set of progressively evolving scales for homogeneity thresholds covering over-segmentation, moderate segmentation, and under-segmentation, for example Fig. 9 As shown in the figure. When the homogeneity merge tolerance threshold is set to only 5, a large number of dense over-segmented surface relief cells are generated. Compared with the total number of merges required to merge from the primary superpixel block to only 6500 terrain cells, the merging process is about 60% complete. As the homogeneity threshold standard is gradually increased, the size of the uniform terrain cell also increases accordingly. When the tolerance threshold is increased to 15, the regional merging process is close to 95% complete, and under-segmented terrain cells that span different valleys and adjacent slopes have begun to appear.
[0135] The impact of scale threshold on the region merging process is further quantified by counting the number of scales, such as Fig.10 As shown in a. The degree of terrain segmentation shows extremely uneven evolutionary changes with increasing scale: the number of terrain units in the homogeneity threshold scale interval of 1 to 9 decreases sharply with increasing scale, corresponding to the rapid merging of a large number of homogeneous and low-level objects in the scale concentration, that is, the over-segmentation stage; while in the scale interval of 9 to 15, the number of terrain units continues to decrease with increasing scale, but the rate of decrease is significantly reduced, indicating that the basic structure of terrain segmentation has been fixed, and the merging process tends to be stable, which is consistent with the characteristics of transition from over-segmentation to moderate segmentation; while in the scale interval of 15 to 30, the number of regions changes very slowly or stagnates with the increase of scale, and the regional merging phenomenon is only observed in local areas, and is usually manifested as the reluctant merging of large heterogeneous objects, which is consistent with the evolutionary characteristics of the under-segmentation stage.
[0136] (5) Best uniform terrain unit result;
[0137] Determining the optimal homogeneity threshold from the hierarchical progressive scale set should also follow the optimal balance between the maximum homogeneity within the unit and the maximum heterogeneity between units. Therefore, the evolution curves of the scale set global variance (V) and global Moran index (MI) with the homogeneity threshold scale are constructed respectively, as shown in Fig.10 As shown in Figure 2bd. The global variance V and Moran's index MI show a monotonic and synchronous change trend as the scale increases. As the homogeneous small areas gradually merge, the monotonic increase of the V index indicates that the homogeneity level gradually decreases, while the monotonic decrease of the MI index indicates that the external heterogeneity gradually increases. The convergence robustness change characteristics prove the effective usability of the V and MI values jointly characterized by multiple terrain features in quantifying the evolution characteristics of homogeneity and heterogeneity of the scale set. By obtaining the maximum value coordinates (12, 1.26) through the global comprehensive heterogeneity index GS curve, it can be determined that the threshold value of the maximum balance between internal homogeneity and external heterogeneity achieved by the regional uniform terrain unit as a whole is 12.
[0138] Using this as the global optimal homogeneity threshold, 37,207 SSM-HTU homogeneous terrain units were obtained in the Yuqu Basin, with an average area of 0.15 km 3 The plateau wide valley and alpine canyon landforms are used as verification examples to obtain the discontinuous structural lines of the real local terrain to verify the rationality and effectiveness of the boundaries of the automatic segmentation method results. First, the plateau wide valley area example Fig.11 As shown in the figure, it was first exposed to the surface during the multi-stage tectonic uplift process and experienced a long process of external weathering and erosion. The overall terrain is gently undulating. Among them, plateau planation surfaces, rounded hills, broad ridges, and foothill accumulation fans are typical geomorphic elements. Fig.11 As shown in Fig. c, there are clear discontinuous structural lines in local areas such as the edge of the planation surface, the junction of the wide ridge and the steep slope, and the transition from the slope surface to the foothill platform. The uniform terrain unit constructed by the new method can form a closed and independent polygon along the visually marked terrain discontinuity line. Fig.11 As shown in Figure d, for example, there is a clear matching relationship between the boundaries of the terrain units and the gentle terraces on the top of the hills, the slopes of the mountains, the gentle slopes at the foot of the mountains, the wide valleys of the river floodplains and other geomorphic elements.
[0139] In the alpine canyon area, typical alpine canyon landforms are formed during the process of river erosion. For example, "V"-shaped valleys, steep slopes, local erosion platforms, and slope-fold valleys are all typical topographic elements. The boundaries of uniform terrain units constructed by the new method can cut the complex canyon terrain into discrete units along these discontinuous structures. The segmentation results can present reasonable adaptive fit characteristics with gradual terrains such as local erosion platforms, gentle slopes adjacent to platforms, and steep slopes between gentle slopes and river valleys.
[0140] (6) Conclusion
[0141] Uniform conditional unit is the most important type of mapping unit besides grid unit and slope unit. Based on simplifying the connotation of uniform condition into the uniform homogeneity of multi-feature terrain, this paper proposes a new uniform terrain unit construction method SSM-HTU method which integrates superpixel segmentation and geometric-texture feature similarity merging. The data preprocessing method that fuses the rough skeleton edge of the terrain represented by the slope unit with the fine edge represented by the primary superpixel block of the terrain can significantly enhance the spatial structural characteristics of the similarity aggregation and difference discontinuity of the multi-dimensional terrain field attributes. This refined edge fusion method is an effective way to solve the fuzzy spatial structure of terrain units and the difficulty of smooth boundary transition in specific scenes of terrain segmentation. The innovative contribution of this new method is mainly reflected in the efficient integration of spatial structural information of multi-source heterogeneous terrain attribute fields, which significantly improves the expression of spatial aggregation and discontinuity of the original terrain and the measurement accuracy of terrain similarity.
[0142] In the example application of automatic extraction of uniform terrain units in Yuqu River Basin, the optimal homogeneity threshold of 12 was determined to achieve the maximum balance between internal homogeneity and external heterogeneity of uniform terrain units based on the geometric-texture similarity region merging process of primary terrain superpixel blocks. Therefore, 37,207 uniform terrain units were further obtained based on 6,573 optimized slope units. The local field verification results show that the newly constructed SSM-HTU method in this paper has significant advantages in obtaining complete and reasonable uniform terrain units that match the discontinuity of the real terrain, so as to provide a new high-quality mapping unit type for landslide susceptibility evaluation in complex disaster-prone environments.
[0143] Embodiment 3:
[0144] In some exemplary embodiments, the present embodiment also provides a terminal device for uniform terrain unit division suitable for high and deep canyon areas, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a minimum technical solution for a uniform terrain unit division method suitable for high and deep canyon areas, as described in Example 1, for achieving the purpose of "automatically acquiring surface undulation units with similar internal structure and differentiated external structure in complex and wide-area terrain areas with high quality through multi-feature terrain preprocessing, superpixel initial segmentation, regional merging criteria, and determining the optimal homogeneity threshold value."
[0145] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.
[0146] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0147] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0148] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0149] A person of ordinary skill in the art can understand that all or part of the steps in realizing the above-mentioned facts and methods can be completed by instructing the relevant hardware through a program, and the program involved or the program described can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: At this time, the corresponding method steps are derived, and the storage medium can be ROM / RAM, a disk, an optical disk, etc.
[0150] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A method for dividing uniform terrain units suitable for high and deep canyon areas, characterized in that: The following steps are involved: S1: Get the original image; S2: The morphological characteristics of different high and low undulating units in the original image are described by slope length, slope and slope shape, the image texture structure is obtained by gray-level co-occurrence matrix, and the ridge line and valley line are used as the terrain skeleton edge information to fuse into the multi-feature terrain attribute field; S3: using the optimized slope unit as a statistical window to perform normalization of multi-feature terrain, forming an independent circulation and normalization reorganization of terrain attributes with the optimized slope unit polygon as the spatial range; the slope unit is formed by the spatial intersection of the ridge line and the valley line; S4: Convert the original image to CIELAB color space, and then implement homogeneity clustering according to color and distance similarity; S5: According to the hierarchical and step-by-step optimization criteria, the adjacent regions with the minimum merging cost are searched one by one. If the increase of the global expression error is the smallest, it is the optimization target and the region merging is implemented; S6: After merging, based on the terrain unit scale set formed by over-segmentation, moderate segmentation and under-segmentation, a scale set result with coarse and fine scale boundaries nested and fixed inheritance is formed through a step-by-step scale parameter strategy; S7: According to the global homogeneity and global heterogeneity characteristics of terrain units at different threshold scales, a comprehensive heterogeneity index is obtained, and the maximum value of the comprehensive heterogeneity index is taken as the global optimal homogeneity threshold. The optimal selection is made based on the global optimal homogeneity threshold to obtain the division result.
2. A method for dividing uniform terrain units suitable for high and deep canyon areas according to claim 1, characterized in that: In step S2, the slope position index LTP is used to describe the slope length, and the calculation formula is: Where: Z O is the altitude of the point, Z max , Z min Indicates the maximum and minimum values of the altitude in the analysis window respectively. The analysis window is the slope unit vector where the point is located; The curvature is used to quantitatively describe the curvature of the slope surface, which can be expressed as profile curvature and plane curvature in vertical and horizontal directions. The terrain wetness index TWI is used to reflect the combined changes of different positions, convex and concave shapes, and slope steepness in the slope. The calculation formula is: Where: TWI is the terrain wetness index, SCA indicates the confluence area per unit contour length flowing through the point (m 2 m -1 ), β is the slope of the point.
3. The method for dividing uniform terrain units suitable for high and deep canyon areas according to claim 1, characterized in that: In step S3, the calculation formula for the independent circulation and normalized recombination of the terrain attributes is: Where: y represents the normalized value, and its range is (-1,1); i Represents the original value; μ, y min and max They respectively represent the average, minimum and maximum values within the statistical area of the slope unit.
4. The method for dividing uniform terrain units suitable for high and deep canyon areas according to claim 1, characterized in that: The step S3 further comprises the following steps: After normalization and reorganization, data fusion and dimensionality reduction are also performed based on principal component analysis. Through orthogonal transformation, multiple groups of feature data are mapped to a limited number of unrelated principal components, and principal component fusion is implemented separately for geometric morphological attributes and texture structure features. The feature groups with a cumulative contribution rate of more than 90% are taken as the principal component set.
5. The method for dividing uniform terrain units suitable for high and deep canyon areas according to claim 1, characterized in that: In step S4, the original image is converted into the CIELAB color space, and then homogeneity clustering is achieved according to color and distance similarity, including the following specific steps: Assume that M*N is the size of the image, i is the number of pixels, the initial clustering parameter is set to K; the initial clustering seed point is C i =[l i ,a i ,b i ,x i ,y i ] T , where the color space is [l,a,b] T , the pixel position is [x,y] T ; Then sample and generate K superpixel blocks of equal size, with a size of (M*N) / K and a sampling step of Taking the cluster seed point as the center, set the search range, calculate the distance D between other clusters and the center, and assign pixel i to the nearest cluster center; Calculate the average vector value of all pixels in K superpixels and update the cluster center until the residual is expected to converge to a certain threshold, then stop the iteration.
6. A method for dividing uniform terrain units suitable for high and deep canyon areas according to claim 5, characterized in that: The characteristic distance calculation formula of the pixel LAB color space is: Where: Where: d(i,C k ) indicates the characteristic distance of the pixel point in LAB color space, d c (i,C k ), d s (i,C k ) respectively indicate the color and spatial distance represented by the Euclidean distance, and m is the weight ratio of adjusting the color distance and spatial distance in the similarity measurement.
7. The method for dividing uniform terrain units suitable for high and deep canyon areas according to claim 1, characterized in that: In step S5, the merging cost is defined as the increased fitting error of the image caused by merging two regions: Where: S m , S n Respectively represent the areas of two adjacent regions m and n, D m,n represents the feature distance between two regions, C m,n Indicates the merging cost of two adjacent regions. The smaller the value, the higher the merging priority of the unit compared with other units.
8. A method for dividing uniform terrain units suitable for high and deep canyon areas according to claim 7, characterized in that: In step S5, the characteristic distance D m,n It is defined as a combined distance of morphological and textural features weighted based on spatial constraints: in In T =1-in G ; Where: G C , G T Respectively represent the geometric shape and texture feature distance; w C 、w T Indicates the weights of geometry and texture distance in the combined distance metric respectively; max(f λ ) represents the maximum frequency of the grayscale histogram of the λth principal component; K represents the weighted maximum frequency of the histogram of each principal component variable in the current region, that is, first quantify the heterogeneity between single terrain attributes, then combine the heterogeneity of each single terrain attribute through adaptive weighting, and finally form the comprehensive heterogeneity of the unit and the adjacent units; l m,n is the common side length of adjacent regions m and n, and α is the influence coefficient of the common side.
9. A method for dividing uniform terrain units suitable for high and deep canyon areas according to claim 8, characterized in that: In step S5, the G statistic is used to measure the characteristic distance between two adjacent regions in terms of slope position, slope, curvature or texture. Assume that the terrain attributes of two adjacent regions m and n in the tth time are represented as m t 、n t , then the G statistic is as follows: Where: G(m,n) represents the G statistic of adjacent regions m and n at the tth attribute; m i t 、n i t They represent the frequency of occurrence of the gray value i of the two histograms respectively; L represents the gray level. Considering the calculation efficiency and expression accuracy, the gray levels of various terrain attributes are uniformly compressed into an 8-bit unsigned integer data format with gray levels of 0 to 255; the value range of G(m,n) is [0,4ln2]. The larger the G statistic value, the greater the histogram distance between two adjacent areas.
10. The method for dividing uniform terrain units applicable to high and deep canyon areas according to claim 1, characterized in that: In step S7, the heterogeneity level between adjacent cells is measured according to the Moran index: Where: m represents the total number of units; w ih Represents the spatial proximity index. If SU h with SU i If there is a common edge, the value is 1, otherwise it is 0; i Represents the average slope of all internal grid cells. represents the average slope aspect of all cells in the study area; The level of homogeneity within the global population is measured in terms of variance: Where: s i and c i Respectively represent SU i The area and circular variance of the slope units with larger scale can be given higher weights to avoid fluctuations caused by small units; The overall level of internal homogeneity and external heterogeneity of multi-feature terrain of global units is measured according to the comprehensive heterogeneity index (GS): Where: σ indicates the mean of the standard deviation of each principal component, Indicates the average value of all principal components in the entire workspace, θ i Indicates the mean of each principal component in region i, MI max(min) and V max(min) are the maximum and minimum values calculated in the heterogeneity level and homogeneity level, respectively. The GS maximum value corresponds to the global optimal homogeneity threshold.
Citation Information
Patent Citations
Method of extracting slope unit in DEM
CN108257142A
Slope unit division method based on multi-scale image segmentation
CN110532969A
Slope unit automatic division method based on confluence segmentation
CN113850822A
Slope unit division model based on variable resolution extraction technology
CN114170392A
Method for setting optimal parameters in slope unit automatic division process and related products
CN118038250A
Cited By
Simulated natural assembly type treatment planning method for river
CN121052677A
Method for monitoring activity of high-order loose body based on multi-source SAR data
CN121069388A
Method for monitoring high-position loose body activity based on multi-source SAR data
CN121069388B
