Underwater terrain adaptation area selection method based on terrain subdivision and Gaussian mixture model
By converting the underwater terrain adaptation area selection method from two-dimensional to three-dimensional, and using Gaussian mixed model cluster analysis, the problem that traditional methods cannot effectively reflect the three-dimensional characteristics of the terrain is solved, and adaptation area selection with higher accuracy and larger area is achieved.
Patent Information
- Application Number
- CN202510229082.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-28
- Publication Date
- 2025-06-13
AI Technical Summary
The traditional method of selecting adaptive zones for underwater terrain is based on two-dimensional grid maps, which cannot effectively reflect the three-dimensional characteristics of the terrain, resulting in the possibility of misalignment and omission of the adaptive zones.
Using a method based on topographic segmentation and Gaussian mixed model, a two-dimensional raster map is converted into a three-dimensional topographic segmentation model, a multi-dimensional feature parameter matrix of the terrain triangle surface is calculated, and the adaptation area is divided through Gaussian mixed model cluster analysis.
The mining capacity of the adapter area has been improved, and the accuracy and area of the selected adapter area have been significantly improved, reducing the impact of artificial weight setting and experience thresholds on the selection results.
Smart Images
Figure CN120141436A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of navigation, guidance and control technology, and in particular to a method for selecting an underwater terrain-assisted navigation adaptation area based on terrain segmentation and a Gaussian mixture model. Background Art
[0002] In the process of marine environment exploration and resource development, underwater vehicles are crucial carrier equipment. Inertial navigation systems are often used as the main system for underwater navigation because of their completely autonomous and independent operation and comprehensive output of navigation information. However, the inertial navigation system has the characteristic of error divergence over time, and its errors need to be corrected with the aid of other technologies. Terrain matching navigation uses bathymetric sonar to obtain measured water depth data and matches and locates it with a pre-stored terrain reference map. Among them, terrain adaptation area selection is one of the core technologies. How to search and find reliable terrain adaptation areas directly affects the matching positioning accuracy.
[0003] The traditional adaptation area selection method is based on the grid point map, using characteristic parameters such as terrain standard deviation, terrain longitude and latitude correlation, terrain longitude and latitude roughness and skewness coefficient for feature description, and setting thresholds according to experience to achieve adaptation area selection for the search area. Its disadvantages are: the traditional terrain characteristic parameters mainly express the two-dimensional characteristics of the terrain, and cannot reflect the three-dimensional characteristics of the terrain; different weight assignments and threshold settings between characteristic parameters will make the adaptation area likely to be misclassified and missed. Summary of the invention
[0004] In view of this, the present invention provides a method for selecting underwater terrain adaptation areas based on terrain subdivision and Gaussian mixture model, which can improve the ability to mine adaptation areas, and the accuracy and area of the selected adaptation areas are significantly improved.
[0005] In order to solve the above technical problems, the present invention is implemented as follows.
[0006] A method for selecting underwater terrain adaptation area based on terrain subdivision and Gaussian mixture model, comprising:
[0007] Step 1: Convert the topographic reference map of the operating waters from a two-dimensional grid map to a three-dimensional terrain subdivision model;
[0008] Step 2: Calculate the terrain characteristic parameter vector of each terrain triangle in the terrain subdivision model to form a multi-dimensional characteristic parameter matrix;
[0009] Step 3: Use Gaussian mixture model clustering to perform adaptability analysis on the multi-dimensional feature parameter matrix of the terrain subdivision model, and divide various adaptation areas according to the different clustering indexes of the terrain triangles.
[0010] Preferably, in step 1, Delaunay triangulation is used to construct a spatial triangular network for the grid point set of the terrain reference map, and continuous triangular surfaces are used to represent the underwater terrain surface, forming a terrain triangulation model.
[0011] Preferably, in step 2, the terrain feature parameter vector includes the terrain feature of the triangular surface at the triangular surface scale, the terrain change trend feature at the triangular surface neighborhood scale, the severity feature of the terrain change of the triangular surface, and the terrain complexity feature where the triangular surface is located.
[0012] Preferably, the terrain feature of the triangular surface at the triangular surface scale includes the slope and aspect of the terrain triangular surface;
[0013] The expression of the terrain triangular surface is: Δ m = E 1 (m)x + E 2 (m)y + E 0 (m);
[0014] The slope S of the terrain triangular surface is:
[0015] The aspect A of the terrain triangular surface is: A(m) = arctan(E 1 (m) / E 2 (m));
[0016] Among them, E 0 (m), E 1 (m), E 2 (m) are the surface coefficients of the terrain triangular surface Δ m ; x, y represent two axes in the spatial rectangular coordinate system.
[0017] Preferably, the terrain change trend feature at the triangular surface neighborhood scale includes the slope deviation coefficient and aspect deviation coefficient of the terrain triangular surface;
[0018] The slope deviation coefficient S dc of the terrain triangular surface is:
[0019]
[0020] Among them
[0021]
[0022] The aspect deviation coefficient A dc of the terrain triangular surface is:
[0023]
[0024] Among them
[0025]
[0026] wherein, μ S (m) is the average slope of the terrain triangular facets in the neighborhood, and σ S (m) is the mean square deviation of the slope of the terrain triangular facets in the neighborhood; μ A (m) is the average aspect of the terrain triangular facets in the neighborhood, and σ A (m) is the mean square deviation of the aspect of the terrain triangular facets in the neighborhood; S(i) is the slope of the terrain triangular facet Δ i , and A(i) is the aspect of the terrain triangular facet Δ i , and N Δ is the number of terrain triangular facets in the neighborhood.
[0027] Preferably, the severity characteristics of the terrain change of the triangular facet include the terrain height difference and the terrain fragmentation degree;
[0028] The terrain height difference T df (m) is:
[0029] T df (m) = |Z(m) max - Z(m) min |
[0030] wherein, T df (m) is the terrain height difference of the terrain triangular facet Δ m , Z(m) max is the depth value of the deepest part of the terrain in the terrain triangular facet Δ m , and Z(m) min is the depth value of the shallowest part of the terrain in the terrain triangular facet Δ m ;
[0031] The terrain fragmentation degree T L is:
[0032] T L = L 12 + L 23 + L 13
[0033] wherein, T L is the terrain fragmentation degree, and L 12 , L 23 , L 13 are the side lengths of the three sides of the terrain triangular facet, respectively, and the longer the sum of the side lengths.
[0034] Preferably, the terrain complexity characteristics of the triangular facet include terrain entropy and terrain Fisher information;
[0035] The terrain entropy H(m) is: For a terrain triangular facet Δ m , the sum of the water depth depths of the three vertices is D Δm , Δm Select the range of N1×N2 for neighborhood, then the terrain entropy H(m) is:
[0036]
[0037] where Δ(i,j) is the terrain triangular facet Δ m the neighbor triangular facet at the i-th row and j-th column in the neighborhood; D Δ (i,j) is the sum of the water depths of the three vertices corresponding to the neighbor triangular facet Δ(i,j);
[0038] The terrain Fisher information content TFIC(m) is:
[0039]
[0040] where represents the partial derivative of the neighbor triangular facet, represents the partial derivative of D Δ(i,j) of.
[0041] Preferably, in step 3, the Gaussian mixture model unsupervised learning algorithm is used to perform clustering analysis on the terrain feature parameter vector; the Gaussian mixture model is composed of sub-models subject to multiple Gaussian distributions; the input of the Gaussian mixture model is the multi-dimensional feature parameter matrix of all terrain triangular facets in the terrain dissection model, and the output is the clustering result of the terrain triangular facets; terrain adaptability analysis is performed based on the clustering result to determine the adaptation area.
[0042] Preferably, when the number of clusters of the Gaussian mixture model is set to 3, the Gaussian mixture model divides into a strong adaptation area, a weak adaptation area, and a non-adaptation area; when the number of clusters of the Gaussian mixture model is set to 2, the Gaussian mixture model divides into an adaptation area and a non-adaptation area.
[0043] Preferably, the method further includes: after obtaining the multi-dimensional feature parameter matrix in step 2, further performing standardization processing on different spatial feature parameters.
[0044] Beneficial effects:
[0045] (1) The present invention innovatively processes the data structure of the pre-stored terrain reference map based on the computational geometry principle dissection method, and converts the terrain grid map into a terrain dissection model. The terrain dissection model can process terrain reference maps with unevenly distributed sampling points. The shapes and sizes of the terrain triangular facets are flexible and variable, and can restore the underwater terrain surface with high accuracy and accurately represent terrain changes.
[0046] (2) Since the feature vectors of terrain triangles in feature-rich and feature-sparse areas in the terrain subdivision model belong to different Gaussian distributions, the present invention uses the Gaussian mixture model clustering method to process the multi-dimensional feature parameter matrix of the terrain subdivision model, realizes the adaptability analysis of the terrain triangles and automatically selects the adaptation area within the entire map. The area of the selected adaptation area is larger than that of the traditional algorithm and has better accuracy.
[0047] (3) In the method of the present invention, based on the terrain triangle, a total of eight spatial characteristic parameters describing terrain changes are designed in four aspects. The four aspects of features cover the small scale of the triangle and the medium scale of the triangle neighborhood in terms of scale; in terms of change, they cover the terrain change trend and the severity of terrain change; and the terrain complexity is also selected. Parameters that can well express the above characteristics and are easy to calculate are selected or designed in a targeted manner. The designed three-dimensional terrain features have better description capabilities for terrain than the two-dimensional features used in traditional algorithms.
[0048] (4) The spatial characteristic parameters selected by the present invention to describe terrain changes are conducive to the clustering of the Gaussian mixture model. Different adaptation areas can be better distinguished through clustering, which can improve the ability to mine adaptation areas. The accuracy and area of the selected adaptation areas are significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] Figure 1 The present invention is a flow chart of the underwater terrain adaptation area selection method based on terrain subdivision and Gaussian mixture model.
[0050] Figure 2 A schematic diagram for converting a terrain reference map into a terrain subdivision model;
[0051] Figure 3 It is a schematic diagram for describing the spatial characteristics of the terrain subdivision model;
[0052] Figure 4 This is a schematic diagram of the selection results of the underwater terrain-assisted navigation adaptation area based on the terrain subdivision model;
[0053] Figure 5 Schematic diagram of the relationship between fragmentation and terrain change degree. DETAILED DESCRIPTION
[0054] The present invention is described in detail below with reference to the accompanying drawings and embodiments.
[0055] The adaptation zone division of underwater terrain-assisted navigation usually adopts the method of comparing two-dimensional terrain features with thresholds. Among them, using terrain grid maps, setting a certain range of search range to calculate several two-dimensional feature parameters of each grid point, the ability to describe terrain undulation is limited; summing different feature parameters according to certain weights, because the representation between feature parameters cannot be unified, the selected adaptation zone is incoherent; artificially setting thresholds for feature parameters based on experience makes it impossible to find more adaptation areas due to threshold settings.
[0056] Therefore, the present invention first breaks through the limitation of two-dimensional grids, and innovatively processes the data structure of the pre-stored terrain reference map based on the computational geometry principle partitioning method, converting the terrain grid map into a three-dimensional terrain partitioning model. Extracting terrain features from a three-dimensional perspective can improve the ability to describe terrain undulations. The terrain partitioning model can process terrain reference maps with uneven distribution of sampling points. The shape and size of the terrain triangles are flexible and changeable, and the underwater terrain surface can be restored with high precision, accurately describing terrain changes.
[0057] Secondly, through research, it is found that the feature vectors of terrain triangles in feature-rich and feature-sparse areas in the terrain subdivision model belong to different Gaussian distributions. That is, areas with drastic terrain changes and areas with gentle terrain changes can be distinguished by the different distribution properties of their terrain triangle feature vectors. Therefore, the Gaussian mixture model is used to perform machine learning clustering on some representative features, which can obtain a good adaptation area division effect. Moreover, the machine learning method does not require artificial threshold setting, which can effectively reduce the impact of artificially set weights and empirical thresholds on the selection results, and can discover more adaptation areas.
[0058] In addition, when selecting terrain features, the present invention covers the small scale of the triangle face and the middle scale of the triangle face neighborhood in terms of scale; covers the terrain change trend and the severity of terrain change in terms of change; and also selects the terrain complexity. And targeted selection or design of parameters that can well express the above features and are easy to calculate. Moreover, the coupling degree between these features is not high, which is conducive to the clustering of Gaussian mixture models. Through clustering, different adaptation areas can be better distinguished, which can improve the mining ability of the adaptation area, and the accuracy and area of the selected adaptation area are significantly improved.
[0059] Based on the above analysis, the present invention provides a method for selecting underwater terrain adaptation area based on terrain subdivision and Gaussian mixture model, such as Figure 1 As shown, it includes the following steps:
[0060] Step 1: Convert the topographic reference map of the operating waters from a two-dimensional raster map to a three-dimensional terrain subdivision model.
[0061] Step 2: Calculate the terrain feature parameter vectors of each terrain triangle in the terrain dissection model to form a multi-dimensional feature parameter matrix.
[0062] Step 3: Use Gaussian mixture model clustering to perform adaptability analysis on the multi-dimensional feature parameter matrix of the terrain dissection model, and divide various adaptation areas according to different clustering indexes of terrain triangles.
[0063] Taking the underwater topographic map of a certain area in the IBCAO dataset as an example, the implementation process of the preferred embodiment of the present invention is described in detail, including the following steps:
[0064] Step 1: Data structure conversion.
[0065] In this step, the topographic reference map of the operation water area is converted from a two-dimensional raster map to a three-dimensional terrain dissection model.
[0066] In this embodiment, Delaunay dissection is used to construct a spatial triangular network for the raster point set of the topographic reference map, and continuous triangular faces are used to represent the underwater terrain surface to form a terrain dissection model. The Delaunay triangulation method follows the maximum minimum angle criterion and the empty circumcircle criterion, that is, the closest three nodes form a triangle; any four points are not concyclic; the minimum angle of the formed triangle is the largest. Applying the triangulation algorithm to the terrain dissection of the pre-stored topographic reference map (raster map), the obtained terrain dissection model is unique. That is, starting from any area, the triangles forming the triangular network do not intersect or contain each other, and the generated triangular network can finally reach consistency. The obtained terrain dissection model M T is a set of triangular faces Δ.
[0067] Specifically, after clarifying the pre-stored topographic reference map corresponding to the operation water area, all raster sampling points (x, y, z) in the spatial rectangular coordinate system are put into the triangle list; the sampling points are inserted in turn, and the triangle whose circumcircle contains the insertion point is found in the triangle list (referred to as the auxiliary triangle of this point). Delete the common sides of the auxiliary triangle, and connect the insertion point with all vertices of the auxiliary triangle. Complete the insertion in the Delaunay triangle list; optimize the locally formed triangulation through a local optimization program, and put the formed triangles into the Delaunay triangle list; loop through the above operations until all raster sampling points are successfully inserted.
[0068] After all raster sampling point data is inserted, the terrain dissection model is completed, and the data structure of the pre-stored topographic reference map is converted from a raster map to a terrain dissection model. This model consists of a series of points and the edges connecting the points within the region D under the constraint of the dissection rule ξ.
[0069] Such as Figure 2As shown, the terrain subdivision model is essentially a terrain digital modeling process that uses surface geometry to mathematically describe the underwater terrain.
[0070] X, Y, and Z are the three-axis coordinates of the terrain subdivision model coordinate system. After being processed by the terrain subdivision algorithm, the representation form of the underwater terrain changes from discrete grid points to interconnected triangular faces. Each triangular face contains the coordinate information of three vertices and related attribute information. A schematic diagram of the spatial structure of the terrain subdivision triangular network is shown in Figure 3 As shown, this information can more accurately describe the local undulations and changes of the terrain. This transformation of the data structure from grid points to triangular faces provides the ability to further perform vector feature analysis and spatial correlation analysis on the terrain.
[0071] Figure 3 Points A, B, C, and D in are grid sampling points in the terrain field. A 1 , B 1 , C 1 , D 1 are the projections of A, B, C, and D on the plane XOY. ΔABC and ΔADC are terrain subdivision triangular faces. n 1 , n 2 are the normal vectors of the triangular faces ΔABC and ΔADC respectively. The terrain subdivision model is composed of a series of continuous terrain triangular faces. In the three-dimensional coordinate system (X, Y, Z), the expression of the triangular face is as follows:
[0072] z = E 1 (m)x + E 2 (m)y + E 0 (m)(2)
[0073] where E 0 , E 1 , E 2 are surface coefficients, and the value range of m is (1, 2,..., M Δ ), and M Δ is the number of triangular faces in the entire terrain subdivision model.
[0074] Step 2: Calculation of the spatial feature parameters of the terrain subdivision model.
[0075] The principle of feature parameter design should be able to reflect and embody the changing trend of the terrain as much as possible in multiple dimensions and be positively correlated with the operation of the matching algorithm. Based on the terrain triangulation network, a large amount of terrain spatial information can be obtained for analysis, such as: the vector feature of the triangulation surface (the terrain feature of the triangulation surface at the triangulation surface scale), the spatial correlation feature of the triangulation surface (the terrain change trend feature at the neighborhood scale of the triangulation surface), the geometric feature of the terrain triangulation surface (the feature of the intensity of terrain change of the triangulation surface), and the terrain complexity feature (the terrain complexity feature where the triangulation surface is located). The calculation and analysis of the above feature parameters are the basis for distinguishing the terrain adaptation area and the non-adaptation area.
[0076] 1) Vector feature of the terrain triangulation surface:
[0077] The vector feature describes the characteristics of a triangulation surface in the terrain triangulation model from two dimensions of magnitude and direction, and this feature reflects the characteristics of the terrain background map at a small scale. As Figure 3 shown in, the normal vector of each triangulation surface is an independent parameter. Through the normal vector, the slope and aspect of the terrain represented by the current triangulation surface can be calculated. The slope is the angle between the normal vector of the triangulation surface and the positive direction vector of the Z-axis, and the aspect is the angle between the projection of the normal vector of the triangulation surface on the horizontal plane (XOY plane) and the due north direction vector (positive direction of the Y-axis). Calculate the slope and aspect of each terrain triangulation surface according to the following calculation formulas:
[0078] The slope S of the terrain triangulation surface is:
[0079]
[0080] The aspect A of the terrain triangulation surface is:
[0081] A(m) = arctan(E 1 (m) / E 2 (m)) (4)
[0082] In the above formula is the normal vector of the current triangulation surface, is the positive direction vector of the Z-axis in the coordinate system, · is the dot product operator, E 1 (m) is the coefficient of x in the triangulation surface expression, and E 2 (m) is the coefficient of y in the triangulation surface expression. The value range of m is (1, 2,..., M Δ ), and M Δ is the number of triangulation surfaces in the entire terrain triangulation model.
[0083] 2) Spatial correlation feature of the terrain triangulation surface:
[0084] Different from the analysis of the characteristics of triangular face vectors at small scales, the spatial correlation characteristics of triangular meshes are used to describe the terrain characteristics at medium scales, that is, the correlation characteristics between adjacent triangular faces or within a certain neighborhood range. The slope and aspect calculated in 1) reflect the trend of each triangular face in the terrain dissection model. If the direction of the triangular face normal vector hardly changes within a certain area, then the terrain depth value shows a single change trend (continuously increasing / decreasing); on the contrary, if the direction of the triangular face normal vector changes significantly within the area, then the terrain here must have severe undulations and the depth value fluctuates significantly; if the normal vectors of all triangular faces in the area are nearly parallel to the Z-axis direction or have a very small angle, then this is a flat terrain. Calculate the slope deviation coefficient and aspect deviation coefficient of a terrain triangular face within the neighborhood according to the following calculation formulas:
[0085] The slope deviation coefficient of the terrain triangular face is:
[0086]
[0087] The aspect deviation coefficient of the terrain triangular face is:
[0088]
[0089] Among them,
[0090]
[0091] In the formula, μ S (m) is the average slope of the terrain triangular faces within the neighborhood, and σ S (m) is the mean square deviation of the slopes of the terrain triangular faces within the neighborhood; μ A (m) is the average aspect of the terrain triangular faces within the neighborhood, and σ A (m) is the mean square deviation of the aspects of the terrain triangular faces within the neighborhood; S(i) is the slope of the terrain triangular face Δ i , and A(i) is the aspect of the terrain triangular face Δ i , and N Δ is the number of terrain triangular faces within the neighborhood.
[0092] The deviation coefficient reflects the relationship between the current triangular face and the normal vectors of the triangular faces within the neighborhood, and can effectively represent the spatial relationship and change trend of the triangular faces within the neighborhood.
[0093] 3) Geometric characteristics of terrain dissection triangular faces:
[0094] Within a terrain triangular facet, the maximum terrain height difference can reflect the severity of the terrain slope change represented by the triangular facet. A larger height difference indicates a steep terrain here, while a smaller height difference may indicate a gentle terrain here; the length of the boundary of each terrain triangular facet also contains rich terrain information. The boundary length is related to the degree of terrain fragmentation. A longer boundary means that the interaction between this triangular facet and other terrain units is more abundant, and the higher the terrain fragmentation, the more complex the terrain features.
[0095] Terrain height difference T df :
[0096] T df (m) = |Z(m) max - Z(m) min | (9)
[0097] Among them, T df (m) is the terrain height difference of the terrain triangular facet Δ m , Z(m) max is the depth value of the deepest part of the terrain within the terrain triangular facet Δ m , Z(m) min is the depth value of the shallowest part of the terrain within the terrain triangular facet Δ m .
[0098] Terrain fragmentation degree T L :
[0099] T L = L 12 + L 23 + L 13 (10)
[0100] Among them, T L is the terrain fragmentation degree, L 12 , L 23 , L 13 are the side lengths of the three sides of the terrain triangular facet respectively, and the longer the sum of the side lengths. Figure 5 It can be clearly seen that the terrain changes violently with a large fragmentation degree, and the terrain changes slowly with a small fragmentation degree.
[0101] 4) Complexity characteristics of the terrain triangulation network:
[0102] When analyzing the terrain complexity at a large scale, the characteristics of the normal vector of the triangular facet are no longer considered, but rather analyzed from the perspective of the amount of information contained in the terrain. In the context of terrain matching, the greater the terrain entropy, the more complex the terrain and the more information it contains; the smaller the terrain entropy, the flatter the terrain and the less information it contains. The calculation formula of terrain entropy is as follows:
[0103]
[0104] When the scale of the terrain subdivision model is selected as N Δ = N1×N2 triangular faces, for a subdivided triangular face Δ i the sum of the water depths at the three vertices is The calculation formula H(m) for the terrain entropy corresponding to the terrain triangular face is as follows:
[0105]
[0106] Based on the traditional feature parameter characterization, the concept of Fisher information is introduced into terrain analysis to propose the Fisher information of the terrain field. This concept represents the variance score of the likelihood function of the probability distribution of terrain data, and can further explore the distribution information of the terrain field to prevent misclassification due to extreme values of a certain feature parameter. Assuming that the scale for calculating Fisher information is selected as N1×N2 triangular faces, then the discrete form of the terrain Fisher information matrix can be expressed as:
[0107]
[0108] Among them, A is a constant, and the value ranges of i and j are 1≤i≤N1 - 1 and 1≤j≤N2 - 1 respectively. represents the partial derivative of the neighboring triangular face, represents the partial derivative of D Δ(i,j) Therefore, the calculation formula for the terrain Fisher information within the search window range is:
[0109]
[0110] Combine the P = 8 different feature parameters proposed and calculated at different scales above to form a multi-dimensional terrain feature parameter matrix T M×8 , and this matrix can reasonably and accurately describe the terrain background map.
[0111]
[0112] Step 3: Construction of the model feature parameter matrix.
[0113] The feature parameters of the model feature parameter matrix T M×8 obtained by merging in Step 2 are not on the same order of magnitude, and the units between the parameters are not unified. Now, perform standardization processing on the feature parameter matrix, and the calculation formula is as follows
[0114]
[0115] Step 4: Selection of the adaptation area.
[0116] To avoid the influence of the setting of artificial experience thresholds on the division of the adaptation area, the present invention uses the method of unsupervised machine learning clustering of Gaussian mixture models to perform clustering analysis on the terrain degree feature parameters, so as to effectively divide the dissected terrain model according to adaptability. The input of the Gaussian mixture model is the multi-dimensional feature parameter matrix of all terrain triangular faces in the terrain dissection model, and the output is the clustering result of the terrain triangular faces; terrain adaptability analysis is carried out based on the clustering result to determine the adaptation area.
[0117] The Gaussian mixture model assumes that the data is composed of sub-models that follow multiple Gaussian distributions. Each individual sub-model is a standard Gaussian model, and its mean μ i and variance σ i are parameters to be estimated. At the same time, each sub-model has a weight parameter π i .
[0118] θ represents the matrix of parameters to be estimated of the model. The formula of the Gaussian mixture model and the formula of the maximum likelihood function are as follows:
[0119]
[0120] The present invention points out that the characteristic parameters of the terrain triangular faces in the adaptation area and the characteristic parameters of the terrain triangular faces in the non-adaptation area belong to different sub-Gaussian models. According to the EM algorithm (Expectation-Maximization) framework, the characteristic parameter matrix of the model is processed by an iterative solution method until the optimal parameters of the Gaussian mixture model containing k sub-models are found. For complex terrain dissection models with different density regions and irregularly shaped triangular faces, good classification effects can be achieved. The implementation steps are as follows:
[0121] 1) Determination of the number of clusters K: The multi-dimensional terrain feature parameter matrix is divided into K categories. When the value of K is 3, three classifications of strong adaptation area, weak adaptation area, and non-adaptation area are realized for the terrain reference map; when the value of K is 2, two classifications of adaptation area and non-adaptation area are realized for the terrain reference map.
[0122] 2) E-step: According to the current parameters, calculate the possibility that the feature parameter vector data j of each triangular face comes from model k. The calculation formula is as follows:
[0123]
[0124] j = 1, 2,..., M Δ ; k = 1, 2,..., K, M Δ is the total number of triangular faces of the terrain dissection model, and K is the number of clusters.
[0125] 3) M-step: Calculate the model parameters of the new round of iteration;
[0126]
[0127] Iteratively calculate the E-step and M-step until convergence.
[0128] So far, the adaptability analysis of each terrain triangular surface has been completed. Taking the three-classification as an example, the triangular surface clustering index is k = 1 for the strong adaptation area, k = 3 for the weak adaptation area, and k = 2 for the non-adaptation area for the three-classification. The algorithm selection results are as Figure 4 shown, where green is the strong adaptation area, yellow is the weak adaptation area, and blue is the non-adaptation area. Thus, the process of selecting the underwater terrain adaptation area based on the terrain dissection model is completed.
[0129] In summary, the method for selecting the adaptation area proposed by the present invention based on the terrain dissection model can effectively describe the characteristic changes of the underwater terrain. The present invention uses the Gaussian mixture model clustering method to process the model feature parameter matrix, realizing the automatic division of the adaptation area and the non-adaptation area. It can effectively reduce the influence of artificially set weights and empirical thresholds on the selection results, improve the mining ability of the adaptation area, and at the same time, the accuracy and area of the selected adaptation area are significantly improved.
[0130] The above specific embodiments only describe the design principle of the present invention. The shapes and names of the components in this description can be different and are not limited. Therefore, those skilled in the art of the present invention can modify or equivalently replace the technical solutions recorded in the foregoing embodiments; and these modifications and replacements do not deviate from the spirit and technical solutions of the present invention, and shall all fall within the protection scope of the present invention.
Claims
1. A method for selecting underwater terrain adaptation area based on terrain subdivision and Gaussian mixture model, characterized in that: include: Step 1: Convert the topographic reference map of the operating waters from a two-dimensional grid map to a three-dimensional terrain subdivision model; Step 2: Calculate the terrain characteristic parameter vector of each terrain triangle in the terrain subdivision model to form a multi-dimensional characteristic parameter matrix; Step 3: Use Gaussian mixture model clustering to perform adaptability analysis on the multi-dimensional feature parameter matrix of the terrain subdivision model, and divide various adaptation areas according to the different clustering indexes of the terrain triangles.
2. The method according to claim 1, characterized in that In step 1, a spatial triangulation network is constructed on the grid point set of the terrain reference map using Delaunay subdivision, and a continuous triangular surface is used to represent the underwater terrain surface to form a terrain subdivision model.
3. The method according to claim 1, characterized in that In step 2, the terrain feature parameter vector includes the triangular face terrain features at the triangular face scale, the terrain change trend features at the triangular face neighborhood scale, the severity features of the triangular face terrain changes, and the terrain complexity features of the triangular face.
4. The method according to claim 3, characterized in that The triangular surface terrain features of the triangular surface scale include the slope and slope direction of the terrain triangle surface; The expression of terrain triangle is: Δ m =E1(m)x+E2(m)y+E0(m); The slope S of the terrain triangle is: The slope A of the terrain triangle is: A(m) = arctan(E1(m) / E2(m)); Among them, E0(m), E1(m), E2(m) are the terrain triangles Δ m The surface coefficient of ; x, y represent the two axes in the spatial rectangular coordinate system.
5. The method according to claim 3, characterized in that The terrain change trend characteristics at the triangular surface neighborhood scale include the slope deviation coefficient and the slope deviation coefficient of the terrain triangular surface; Slope deviation coefficient S of terrain triangle dc for: in Slope deviation coefficient A of terrain triangle dc for: in In the formula, μ S (m) is the mean slope of the terrain triangle in the neighborhood, σ S (m) is the mean square error of the slope of the terrain triangle in the neighborhood; μ A (m) is the mean slope of the terrain triangles in the neighborhood, σ A (m) is the mean square error of the slope of the terrain triangle in the neighborhood; S(i) is the slope of the terrain triangle Δ i The slope of the terrain triangle Δ i Slope direction, N Δ is the number of terrain triangles in the neighborhood.
6. The method according to claim 3, characterized in that The characteristics of the intensity of the triangular surface terrain changes include terrain height difference and terrain fragmentation; The terrain height difference T df (m) is: T df (m)=|Z(m) max -Z(m) min | Among them, T df (m) is the terrain triangle surface Δ m Terrain height difference, Z(m) max is the terrain triangle Δ m The deepest depth of the inner terrain, Z (m) min is the terrain triangle Δ m The depth value of the shallowest part of the inner terrain; The terrain fragmentation T L for: T L =L 12 +L 23 +L 13 Among them, T L is the terrain fragmentation, L 12 ,L 23 ,L 13 They are the side length, side length and the length of the three sides of the terrain triangle.
7. The method according to claim 3, characterized in that The terrain complexity characteristics of the triangular surface include terrain entropy and terrain Fisher information; The terrain entropy H(m) is: m , the sum of the water depths of the three vertices is Δ m The neighborhood of is selected as N1×N2, then the terrain entropy H(m) is: Among them, Δ(i,j) is the terrain triangle Δ m The neighbor triangle in the i-th row and j-th column in the neighborhood; D Δ (i, j) is the sum of the water depths of the three vertices corresponding to the neighboring triangle Δ(i, j); The terrain Fisher information TFIC(m) is: in, represents the partial derivative of the neighbor triangle, Indicates D Δ(i,j) The partial derivative of .
8. The method according to claim 1, characterized in that In step 3, a Gaussian mixture model unsupervised learning algorithm is used to perform cluster analysis on the terrain feature parameter vector; the Gaussian mixture model is composed of sub-models that obey multiple Gaussian distributions; the input of the Gaussian mixture model is a multi-dimensional feature parameter matrix of all terrain triangles in the terrain subdivision model, and the output is the clustering result of the terrain triangles; the terrain adaptability analysis is performed based on the clustering results to determine the adaptation area.
9. The method according to claim 1, characterized in that When the number of clusters of the Gaussian mixture model is set to 3, the Gaussian mixture model divides the area into a strong adaptation area, a weak adaptation area, and a non-adaptation area; when the number of clusters of the Gaussian mixture model is set to 2, the Gaussian mixture model divides the area into an adaptation area and a non-adaptation area.
10. The method according to claim 1, characterized in that The method further includes: after obtaining the multi-dimensional feature parameter matrix in step 2, further standardization processing of different spatial feature parameters is performed.