Three-dimensional complex micro-terrain identification method and device, computer device and storage medium

Abstract

This invention discloses a method, apparatus, computer device, and storage medium for three-dimensional complex micro-topography recognition. The method includes: acquiring DEM data to be identified; inputting the DEM data to a classification model to identify micro-topography categories, thereby obtaining the corresponding micro-topography categories; wherein the classification model is obtained by training a convolutional neural network using a DEM encoder and a data dimensionality reduction method based on Transformation Groups to reduce the DEM data as a sample set, combined with a parameterized micro-topography classification system and an automated pre-clustering classification algorithm for complex micro-topography. By implementing the method of this invention, problems such as incomplete and inflexible classification system coverage, complex and redundant DEM data, easy misclassification of micro-topography, imbalanced sample size, need for manual intervention, and difficulty in determining the number of clusters can be solved.

Description

Technical Field

[0001] This invention relates to terrain identification methods, and more specifically to three-dimensional complex micro-terrain identification methods, devices, computer equipment, and storage media. Background Technology

[0002] A small section of a transmission line, or even just between one or two towers, may experience heightened climatic factors due to systemic weather patterns and specific topographical influences, exceeding the designed ice and wind conditions for the area. This could potentially jeopardize the operation of the transmission line. Such areas are called "micro-topographic and micro-meteorological zones." In recent years, many researchers have recognized the significant impact of micro-topography and micro-meteorology on transmission towers and have attempted to analyze these factors from meteorological and topographical perspectives. However, few researchers have conducted comprehensive classification and analysis of micro-topographic and micro-meteorological zones.

[0003] Geomorphology refers to the geometric shape characteristics of the Earth's surface, and along with the composition, formation, and evolution of surface materials, it forms the core of geomorphological research. Geomorphological forms exhibit typical spatial differences in distribution, yet also display relatively independent characteristics within a certain range. Therefore, in research and application, they are often classified into different geomorphological types. Geomorphological types possess typical scale and hierarchical characteristics. Depending on the research and application objectives, there is no unified and definitive classification system for geomorphological types across different fields. However, for ease of understanding, they can be broadly divided into basic morphological types and composite morphological types. The former are the basic unit types of geomorphology, such as slope position and geomorphological elements; while the latter are combinations of basic morphological types and lower-level morphological types according to a certain spatial pattern. For example, a slope is composed of different slope positions, and a multi-level terrace is composed of several single terraces. As a basic spatial unit of geographical elements, geomorphological types play an important role in the study of micro-topography. For instance, geomorphological research, based on geomorphological types, further considers non-morphological factors such as material composition, formation, and evolutionary characteristics to identify geomorphological types. Landform types are frequently used as important input data in many applications such as ecological research and speculative soil mapping. Therefore, it is essential to efficiently and accurately extract the spatial distribution of landform types.

[0004] Micro-topography classification techniques include existing micro-topography classification systems, micro-topography DEM (Digital Elevation Model) encoding, and systems that combine clustering and classification. Current micro-topography classification systems cannot cover all possible terrain features, only those micro-topography features that researchers believe may affect power transmission lines. Similar terrain cannot be parameterized or quantified for differentiation, leading to misclassification and requiring manual sample finding for supervised classification, which is very time-consuming. For DEM encoding, the Geomorphons algorithm is widely used, allowing for further micro-topography classification. However, the Geomorphons algorithm only considers the elevation of each surrounding point relative to the center point in a 3x3 grid, neglecting the pairwise size of the nine points within the 3x3 grid. This results in the loss of much DEM texture. However, over-considering every possible texture of the DEM would require an enormous sample size. For the classification of DEMs, most scholars currently use a method of clustering first and then classifying, which requires manual intervention in the process. For similar categories, misclassification is prone to occur. In actual terrain, the number of DEM samples of different categories is different, which can easily affect the performance of the classification algorithm.

[0005] Therefore, it is necessary to design a method to address the problems of incomplete and inflexible classification system coverage, complex texture and redundancy of DEM data, easy misclassification of micro-topography, imbalanced sample size, need for manual intervention, and difficulty in determining the number of clusters. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, device, computer equipment and storage medium for three-dimensional complex micro-terrain recognition.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a three-dimensional complex micro-terrain identification method, comprising:

[0008] Obtain the DEM data to be identified;

[0009] The DEM data to be identified is input into a classification model to identify the micro-topography category, so as to obtain the corresponding micro-topography category;

[0010] The classification model is obtained by using a DEM encoder and a data dimensionality reduction method based on Transformation Group to reduce DEM data as a sample set, combined with a parameterized micro-topography classification system and an automated pre-clustering classification algorithm for complex micro-topography to train a convolutional neural network.

[0011] Its further technical solution is as follows: The classification model is obtained by using a DEM encoder and a data dimensionality reduction method based on TransformationGroup to reduce DEM data as a sample set, combined with a parameterized micro-topography classification system and an automated pre-clustering classification algorithm for complex micro-topography to train a convolutional neural network, including:

[0012] Construct a parametric micro-topography classification system;

[0013] Select DEM data that meets the requirements;

[0014] The DEM data is encoded using an improved Geomorphons algorithm to obtain DEM codes;

[0015] The DEM encoding is dimensionality reduced based on the Transformation Group algorithm to obtain dimensionality-reduced data;

[0016] Specify a number of categories and use the k-means algorithm to perform unsupervised classification on the dimensionality-reduced data.

[0017] Input parameters, construct fake terrain according to the parameterized micro-terrain classification system, and output the DEM code corresponding to the fake terrain;

[0018] Based on the DEM code corresponding to the fake terrain, determine whether there is only one parameterized micro-terrain classification category under each category of the corresponding unsupervised k-means classification;

[0019] If there is only one parameterized micro-topography classification category under each category of the corresponding k-means unsupervised classification, then geographic information features are extracted from the DEM data with the determined micro-topography categories using GIS spatial analysis. The extracted features and corresponding categories are then combined to form sample pairs.

[0020] The sample pairs are input into the convolutional neural network for training, and the trained convolutional neural network is saved to obtain a classification model.

[0021] The further technical solution is as follows: after determining whether there is only one parameterized micro-topography classification category under each category of the corresponding k-means unsupervised classification based on the DEM encoding of the fake terrain, it also includes:

[0022] If each category in the corresponding unsupervised k-means classification does not have only one parameterized micro-topography classification category, then adjust the number of classification categories and execute the specified number of categories, using the k-means algorithm to perform unsupervised classification on the dimensionality-reduced data.

[0023] The further technical solution is as follows: Based on the parameterized micro-topography classification system, the input parameters are used to construct a fake terrain and output the DEM code corresponding to the fake terrain, including:

[0024] Based on the parametric micro-terrain classification system, terrain parameters can be set to obtain a variety of different fake terrains;

[0025] An improved Geomorphons algorithm was used to encode the DEM data of various fake terrains to obtain the corresponding DEM codes.

[0026] The further technical solution is as follows: the convolutional neural network includes the AlexNet model.

[0027] The further technical solution is as follows: the AlexNet model includes five convolutional layers and three fully connected layers; and the AlexNet model uses stacked convolutional layers to extract image features, uses Dropout and data augmentation to suppress overfitting, and the AlexNet model uses ReLU as the activation function.

[0028] The further technical solution is as follows: the DEM data is encoded using the improved Geomorphons algorithm to obtain DEM encoding, including:

[0029] The DEM data is compared pairwise and encoded to obtain a base-9 encoding.

[0030] The 9-base code is converted to decimal code to obtain the DEM code.

[0031] The present invention also provides a three-dimensional complex micro-terrain recognition device, comprising:

[0032] The acquisition unit is used to acquire the DEM data to be identified;

[0033] The identification unit is used to input the DEM data to be identified into the classification model to identify the micro-topography category, so as to obtain the corresponding micro-topography category; wherein, the classification model is obtained by using a DEM encoder and a data dimensionality reduction method based on TransformationGroup to reduce the DEM data as a sample set, combined with a parameterized micro-topography classification system and a complex micro-topography automated pre-clustering classification algorithm to train a convolutional neural network.

[0034] The present invention also provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the above-described method.

[0035] The present invention also provides a storage medium storing a computer program that, when executed by a processor, can implement the above-described method.

[0036] The beneficial effects of this invention compared with the prior art are as follows: This invention identifies micro-topography categories by inputting the DEM data to be identified into a classification model. The classification model is obtained by using a DEM encoder and a data dimensionality reduction method based on TransformationGroup to reduce the DEM data as a sample set, combined with a parameterized micro-topography classification system and a complex micro-topography automated pre-clustering classification algorithm to train a convolutional neural network. This solves the problems of incomplete and inflexible classification system coverage, complex texture and redundancy of DEM data, easy misclassification of micro-topography, unbalanced sample size, need for manual intervention, and difficulty in determining the number of clusters.

[0037] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0038] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram illustrating an application scenario of the three-dimensional complex micro-terrain recognition method provided in this embodiment of the invention.

[0040] Figure 2 A flowchart illustrating the three-dimensional complex micro-terrain recognition method provided in an embodiment of the present invention;

[0041] Figure 3 A schematic diagram of the sub-processes of the three-dimensional complex micro-terrain recognition method provided in the embodiments of the present invention;

[0042] Figure 4 A schematic diagram of the sub-processes of the three-dimensional complex micro-terrain recognition method provided in the embodiments of the present invention;

[0043] Figure 5 A schematic diagram of the sub-processes of the three-dimensional complex micro-terrain recognition method provided in the embodiments of the present invention;

[0044] Figure 6 A schematic diagram of the geometric parameters of a Y-shape provided in an embodiment of the present invention;

[0045] Figure 7 A schematic diagram of DEM geometric invariants provided in an embodiment of the present invention;

[0046] Figure 8This is a schematic diagram of the clustering results of Y-type categories provided in an embodiment of the present invention;

[0047] Figure 9 A schematic diagram illustrating the process of automatically selecting training samples according to an embodiment of the present invention;

[0048] Figure 10 A schematic diagram of the AlexNet model provided in an embodiment of the present invention;

[0049] Figure 11 This is a schematic diagram of a network with dropout added, provided in an embodiment of the present invention.

[0050] Figure 12 A schematic diagram of the maximum pooling layer structure provided in an embodiment of the present invention;

[0051] Figure 13 A schematic block diagram of a three-dimensional complex micro-terrain recognition device provided in an embodiment of the present invention;

[0052] Figure 14 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0055] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0056] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0057] Please see Figure 1 and Figure 2 , Figure 1 This is a schematic diagram illustrating an application scenario of the three-dimensional complex micro-terrain recognition method provided in an embodiment of the present invention. Figure 2 This is a schematic flowchart illustrating the three-dimensional complex micro-topography recognition method provided in an embodiment of the present invention. The method is applied in a server. The server interacts with a terminal, using a trained classification model to identify the DEM (Digital Elevation Model) data input from the terminal. A parametric micro-topography classification system is employed to flexibly adjust various micro-topography types. A DEM encoder and Transformation Group-based data dimensionality reduction are used to reduce the complexity and redundancy of the DEM data. Finally, an automated pre-clustering classification algorithm for complex micro-topography is used for micro-topography classification.

[0058] Figure 2 This is a flowchart illustrating the three-dimensional complex micro-terrain recognition method provided in an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S120.

[0059] S110. Obtain the DEM data to be identified.

[0060] In this embodiment, the DEM data to be identified refers to the DEM data of the micro-topography that needs to be categorized.

[0061] S120. Input the DEM data to be identified into the classification model to identify the micro-topography category, so as to obtain the corresponding micro-topography category.

[0062] In this embodiment, micro-topography categories are classified according to the contour of the mountain as cosine, Gaussian, triangular, bell-shaped, etc.; according to the number of mountains as single mountains, double mountains, and mountain ranges; according to the size of the mountains as peaks and mountain ranges; and according to the trend of the mountains and the relative relationship between the mountains as passes, valleys, and funnel-shaped passes, etc.

[0063] In addition, the classification model is obtained by using a DEM encoder and a data dimensionality reduction method based on Transformation Group to reduce DEM data as a sample set, combined with a parameterized micro-topography classification system and an automated pre-clustering classification algorithm for complex micro-topography to train a convolutional neural network.

[0064] In one embodiment, the above-mentioned classification model is obtained by using a DEM encoder and a data dimensionality reduction method based on Transformation Group to reduce DEM data as a sample set, and training a convolutional neural network by combining a parameterized micro-topography classification system and an automated pre-clustering classification algorithm for complex micro-topography, and may include steps S121 to S1210.

[0065] S121. Construct a parametric micro-topography classification system;

[0066] In this embodiment, the parametric micro-topography classification system refers to a system that uses parameters to classify micro-topography. In this system, each category can cover a variety of seemingly different DEMs simply by adjusting different parameters. The geometric parameters of the terrain category can also be considered as the geometric features of that category.

[0067] The parameterized micro-topography classification system is as follows: First, the canyon-pass-mountain type: Passes formed between continuous mountain ranges exhibit concentrated acceleration of airflow. When transmission lines cross these passes, the funneling effect leads to a significant increase in lateral wind speed and a possible slight increase in icing. When a line crosses a canyon with high, steep banks, the funneling effect generates large and prolonged local lateral wind speeds, significantly increasing the wind load on the transmission line. Second, the cosine mountain range: Cosine mountains are a common terrain feature and a component of various micro-topographical systems. They block upwind winds, resulting in lower temperatures on the leeward slope and a higher likelihood of circuit icing. Third, the π-shaped mountain range: When a line crosses a canyon with high, steep banks, the funneling effect generates large and prolonged local lateral wind speeds, significantly increasing the wind load on the transmission line and potentially causing conductor galloping. Fourth, the Y-shaped mountain range: When a transmission line corridor faces a Y-shaped mountain range, the windward slope's topographic lifting effect generates strong, moisture-laden winds, easily leading to tower collapse and line breakage on the windward slope. At the mountaintop and windward slope, the adiabatic expansion process occurs as the water rises along the slope under the influence of wind speed. This causes the supercooled water to cool out of the air and adhere to the conductor, exacerbating icing. Each terrain has corresponding geometric parameters, such as the geometric parameters of a Y-shaped structure. Figure 6 As shown.

[0068] Building upon the aforementioned parametric classification system, we will proceed with micro-topographic classification of geographic information feature data. First, DEM encoding will be performed, followed by micro-topographic classification. DEM encoding uses a unique string of codes to represent DEM grid data containing complex textures.

[0069] S122. Select DEM data that meets the requirements.

[0070] In this embodiment, it is necessary to select high-precision DEM data that meets the requirements, including accuracy, precision, and completeness.

[0071] S123. The DEM data is encoded using the improved Geomorphons algorithm to obtain the DEM code.

[0072] In this embodiment, DEM encoding refers to the result obtained by encoding the DEM data using the improved Geomorphons algorithm.

[0073] The improved Geomorphons algorithm is a DEM encoder for micro-terrain recognition. It can unify multiple DEM image blocks with different elevations into a fixed code, while preserving the texture details inside the data block and reducing the complexity of DEM data.

[0074] In one embodiment, please refer to Figure 4 The above step S123 may include steps S1231 to S1232.

[0075] S1231. The DEM data is compared pairwise and encoded to obtain a base-9 encoding.

[0076] In this embodiment, 9-base encoding refers to encoding the 9x9 DEM into a 9-base number.

[0077] S1232. Convert the 9-base code to a decimal code to obtain the DEM code.

[0078] For example: Suppose we have 3x3 DEM data. Perform pairwise comparisons and encode the data; specifically, sort the DEM data by size to obtain -12 1 4 5 5 5 7 13 97; convert these into data numbers. When forming the DEM code, the smallest number is taken for the same DEM, 6 1 2 3 3 3 7 9 8, finally resulting in a base-9 DEM code: 501222687, which is further converted to a decimal number: 215898289.

[0079] The above encoding method encodes a 9x9 DEM into a base-9 number, which is then converted to a decimal number. Unlike the ternary encoding of the Geomorphons algorithm, this encoding method can preserve the texture information of the DEM, the size relationship between each grid point, their relative position, and topological structure. Users can set the size of the DEM themselves, resulting in base-25 encoding (5x5 DEM), base-49 encoding (7x7 DEM), etc., all of which can be converted to decimal numbers and used as unique codes for the DEM data block within a specified row and column size. Traversing the base-9 encodings in the above examples reveals nearly 30 million possible cases, with a sample size on the order of 10^9. 7 If the user increases the row and column size of the DEM, the number of samples will increase exponentially, which is very detrimental to training. Therefore, the samples will be dimensionality reduced next.

[0080] S124. The DEM encoding is reduced in dimensionality based on the Transformation Group algorithm to obtain dimensionality-reduced data.

[0081] In this embodiment, the dimensionality-reduced data refers to the result of dimensionality reduction of the DEM encoding based on the Transformation Group algorithm. Dimensionality reduction based on Transformation Group reduces the redundancy of the DEM data.

[0082] Transformation groups are a concept in abstract algebra, a crucial subject in geometry. They are groups formed by transformations. Let G be the set of all transformations of a set S. If G satisfies the following conditions: 1) the product of any two transformations within the set belongs to this set; 2) the inverse transformation of any transformation within the set also belongs to this set, then G is called a transformation group of S. For example, the transformation group of all orthogonal transformations in the plane is called the orthogonal group; the transformation group of all affine transformations in the plane is called the affine group; and the transformation group of all projective transformations in the plane is called the projective group. In the Erlangen Program, transformation groups can be used to classify geometry. In this embodiment, this theory is used to classify and reduce the dimensionality of DEM samples.

[0083] A set of transformations, the group formed by the products of transformations, is called a transformation group on M if G is a finite or infinite set of transformations that satisfies the following two conditions: first, the product of any two transformations in G is still a transformation of G; second, every transformation in G has an inverse transformation, and this inverse transformation is also a transformation of G. For example, translation transformations can form a group: the product of any two translation transformations on a plane is still a translation transformation; each translation transformation has an inverse transformation, which is the transformation in the opposite direction of the original transformation, so it is still a translation transformation. Given a set of arbitrary geometric objects M, M is conventionally called space. The process of transforming each geometric object or element in M ​​onto another geometric object is called a geometric transformation on M, or simply a transformation. Let α represent a geometric object or a figure composed of many objects, and let T represent a geometric transformation. Then, transforming α into another object or figure b under T is denoted as T(α) = b, where b is called the image of α, and α is called the image source of b. Let another transformation S be applied to b, and let S(b) = c. If these two transformations are applied consecutively, then α transforms to c. Therefore, the process of α transforming to c is also a transformation, denoted as P, i.e., P(α) = c. P is called the product of S and T, denoted as P = ST. The order of transformation products is generally non-commutative, i.e., ST ≠ TS. If there are three transformations T, S, and R, and T is applied first, then S, and finally R, the result is RST. This notation indicates that the order of application is from right to left. The associative law of transformation products holds: (RS)T = R(ST) = RST. If a transformation T makes each element b a unique image of some element α, then T is called a one-to-one transformation. In this case, T has a definite inverse transformation, denoted as T-1. The product of T and T-1 keeps each element unchanged, which is called the identity transformation, denoted as E, i.e., TT-1 = T-1T = E.

[0084] Let G be a set of finite or infinite transformations on M, satisfying the following two conditions: first, the product of any two transformations in G still belongs to G; second, every transformation in G must have its inverse transformation, and this inverse transformation also belongs to G. Then G is called a transformation group on M. If a subset of transformations is taken from a known transformation group G, and their entirety also constitutes a transformation group G1, then G1 is called a transformation subgroup of G. It is easy to see from the definition that sets of motion, affine transformations, projective transformations, etc., on a plane or in space each constitute a transformation group, respectively called the motion group, affine group, projective group, etc.; the motion group is a subgroup of the affine group, and both the motion group and the affine group are subgroups of the projective group. Given a space M and a transformation group G, if there is a transformation in G that transforms figure α to figure b, then α and b are said to be equivalent. From the definition of a transformation group, it can be deduced that if figure α is equivalent to figure b, then figure b is also equivalent to figure α. In fact, if figures α and b are equivalent, then there must be a transformation T in group G such that T(α) = b; thus T⁻¹(b) = α. However, T⁻¹ belongs to G, which indicates that there is a transformation in G that transforms b to α. Therefore, b and α are equivalent. If two figures α and b are both equivalent to a third figure c, then α and b are also mutually equivalent. In fact, if α and c are equivalent, then there must be a transformation T in group G such that T(α) = c; and if b and c are equivalent, then there must be a transformation S in G such that S(b) = c, thus S⁻¹(c) = b. Therefore, S⁻¹T(α) = b, so figures α and b are equivalent. The equivalence properties of figures in space M are called geometric properties or invariant properties. Moreover, combining geometric properties with quantities that remain unchanged under any transformation in a known group G, these invariants are obviously common to all equivalent figures. All invariant properties under all transformations in a certain group G are called properties belonging to G, and the geometry that studies properties belonging to G is called the geometry belonging to G. Properties that remain unchanged after motion are called metric properties, and the geometry that studies metric properties is called metric geometry (Euclidean geometry). Properties that remain unchanged after affine transformations are called affine properties, and the geometry that studies affine properties is called affine geometry. Properties that remain unchanged after projective transformations are called projective properties, and the geometry that studies projective properties is called projective geometry; and so on. Under a motion group, distance, angle, area, parallelism, single ratio, and cross ratio remain unchanged. Under affine transformations, distance, angle, and area change, but single ratio, parallelism, collinearity, and cross ratio (of line segments in the same direction) remain unchanged. For a projective group, single ratio and parallelism change, but collinearity and cross ratio remain unchanged. This is because a motion group is a subgroup of an affine group, and an affine group is a subgroup of a projective group.

[0085] According to transformation group theory, every "graph," a data block with a topological or spatial structure (referred to here as a DEM), possesses geometric invariance. For example, if a DEM data block retains its geometric properties after rotation or flipping, then DEMs with the same properties can be removed from the overall sample set. Users can customize the shape of the DEM; it can be a nonagon or a 25-sided polygon. The more sides, the greater the dimensionality reduction.

[0086] Based on the above examples, when considering that all DEMs are quadrilaterals, the DEM has the following geometric invariants: Figure 7 As shown, the advantage of this method is that if the user extracts a set of irregular polygonal DEMs, the DEM can be described in a rasterized form (matrix). Then, using invariant theory, different degrees of data dimensionality reduction can be performed by setting the "number of edges of the graphic". Furthermore, during training, the distance between different samples will be much larger than before dimensionality reduction, thus reducing misclassification between similar categories. After removing the DEM invariants, the final DEM is obtained by traversing the sample set for K-means training.

[0087] S125. Specify a number of categories and use the kmeans algorithm to perform unsupervised classification on the dimensionality-reduced data.

[0088] In this embodiment, the number of categories refers to the total number of categories of the final micro-topography.

[0089] The K-means algorithm is a classic algorithm in clustering and one of the top ten classic algorithms in data mining. By inputting the dimensionality-reduced data from the example above, we obtained 32 DEM classifications. This shows that the categories that need to be classified in the classification system are basically reflected in the results of unsupervised classification. For example, the results corresponding to the Y-type category are as follows... Figure 8 As shown, the K-means algorithm is used to classify the DEM traversal samples. Based on these classification results, the DEM of the study area can be matched to different categories, and the correctly classified samples can be automatically selected as samples for subsequent supervised classification based on AlexNet.

[0090] S126. Input parameters, construct fake terrain according to the parameterized micro-terrain classification system, and output the DEM code corresponding to the fake terrain.

[0091] In one embodiment, please refer to Figure 5 The above step S126 may include steps S1261 to S1262.

[0092] S1261. Based on the parametric micro-terrain classification system, terrain parameters are set to obtain various different fake terrains;

[0093] S1262. The DEM data of various fake terrains are encoded using the improved Geomorphons algorithm to obtain the DEM codes corresponding to the fake terrains.

[0094] Based on the parametric classification system, various fake terrains are generated by setting terrain parameters. These fake terrains are then encoded using the DEM method described above, and their categories in unsupervised classification are examined. If two fake terrains are classified into the same category, the number of categories in the k-means algorithm is adjusted, and unsupervised classification is performed again. Finally, the number of unsupervised categories and the corresponding unsupervised classification number for each category of fake terrain are determined.

[0095] S127. Based on the DEM code corresponding to the fake terrain, determine whether there is only one parameterized micro-terrain classification category under each category of the corresponding unsupervised k-means classification.

[0096] S128. If there is only one parameterized micro-topography classification category under each category of the corresponding unsupervised k-means classification, then use GIS spatial analysis to extract geographic information features from the DEM data with the determined micro-topography categories, and combine the extracted features with the corresponding categories to form sample pairs.

[0097] Based on the definitions of different micro-topography features in the recommended micro-topography classification system, and combined with the acquired geographic information features, sample pairs can be generated. Specifically, the micro-topography category corresponding to different DEMs is used as the output, and the geographic information features corresponding to each DEM are used as the input, forming sample pairs. Thus, the problem is transformed into a classic image classification problem, which can be solved using the AlexNet image classification model based on convolutional neural networks. By feeding the samples into the AlexNet model for training, the functional relationship between geographic information features and micro-topography categories can be obtained. In subsequent work, only the geographic information features corresponding to new DEMs need to be input into the trained model to obtain the micro-topography classification.

[0098] S129. Train the sample pairs into the convolutional neural network and save the trained convolutional neural network to obtain the classification model.

[0099] In this embodiment, the convolutional neural network includes the AlexNet model. The AlexNet model includes five convolutional layers and three fully connected layers; and the AlexNet model uses stacked convolutional layers to extract image features, uses Dropout and data augmentation to suppress overfitting, and uses ReLU as the activation function.

[0100] AlexNet deepens the network structure based on LeNet, learning richer and higher-dimensional image features. AlexNet has the following characteristics: First, a deeper network structure; the entire AlexNet network structure consists of 5 convolutional layers and 3 fully connected layers, with a total depth of 8 layers, such as... Figure 10 As shown; secondly, it uses stacked convolutional layers, i.e., convolutional layer + convolutional layer + pooling layer, to extract image features, and uses Dropout to suppress overfitting. The AlexNet model contains a total of 8 weight layers, including 5 convolutional layers and 3 fully connected layers. Convolutional layers 1 and 2 are followed by an LRN layer. The LRN is not a mandatory layer in a CNN. Each LRN and the last convolutional layer is followed by a max pooling layer, and each weight layer is connected to the ReLU activation function. Dropout is used after the fully connected layers to solve overfitting. Dropout can be used as an option for training deep neural networks. In each training batch, by ignoring half of the feature detectors and setting half of the hidden layer nodes to 0, overfitting can be significantly reduced. This method can reduce the interaction between feature detectors (hidden layer nodes). Detector interaction refers to the fact that some detectors depend on other detectors to function. Simply put, Dropout means that during forward propagation, the activation value of a neuron stops working with a certain probability p. This makes the model more generalizable because it does not rely too much on certain local features, such as... Figure 11As shown; thirdly, data augmentation is used to suppress overfitting. The first form of data augmentation is to use random cropping and flipping to expand the number of training samples by thousands of times. Assuming the original input sample is 256×256, to obtain more samples, a 224x224 template is used to crop the original image, so that each image can generate many similar images, thus obtaining sufficient samples. The second form of data augmentation is to change the intensity of the RGB channels in the training image. Specifically, we perform PCA to reduce the dimensionality of the RGB pixel value set in the entire image, and generate more images by changing the brightness of the main colors. Fourthly, ReLU is used to replace the previous sigmoid as the activation function, in conjunction with LRN; neural networks need to use activation functions, which generally have two functions. First, whether it is convolution or fully connected layers, they are essentially linear operations. By adding non-linear activation functions, the non-linear mapping ability of the network can be enhanced, which is equivalent to the kernel method. Second, neural networks are generally optimized by the backpropagation algorithm, and activation functions can speed up the convergence speed of the network. Traditional neural network activation functions are typically arctangent or sigmoid. AlexNet uses ReLU as its activation function, which improves training speed by approximately 6 times compared to arctangent. Adding LRN after the ReLU activation enhances the network's generalization ability. Based on the characteristics of convolutional neural networks, AlexNet can consider the surrounding DEM distribution when classifying micro-terrain at a given grid point. The network contains multiple feature extractors composed of convolutional and pooling layers. In a convolutional layer, a neuron is connected only to a subset of its neighboring neurons. A convolutional layer typically contains several feature maps, each composed of neurons arranged in a rectangular pattern. Neurons within the same feature map share weights, which are the convolutional kernel. The convolutional kernel is generally initialized as a random fractional matrix, and during network training, it learns appropriate weights. The direct benefit of shared weights (convolutional kernels) is reduced connections between network layers, while also lowering the risk of overfitting. Subsampling, also known as pooling, typically takes two forms: mean pooling and max pooling. Figure 12 As shown, subsampling can be viewed as a special type of convolution process. Convolution and subsampling greatly simplify model complexity and reduce the number of model parameters.

[0101] S1210. If each category of the corresponding unsupervised k-means classification does not have only one parameterized micro-topography classification category, then adjust the number of classification categories and execute step S125.

[0102] In this embodiment, adjusting the number of classification categories refers to the total number of micro-topography classifications.

[0103] First, unsupervised clustering is performed based on k-means to obtain pre-classification results. Then, a pseudo-terrain is constructed based on a parameterized classification system. Using this as a benchmark, samples with accurate pre-classification are automatically selected, and relevant geographic feature parameters are calculated. These parameters are then substituted into AlexNet for supervised classification to obtain the final micro-terrain classification model.

[0104] Using AlexNet, a relationship model (function) between different geographic information features and different micro-topography was obtained, serving as a tool for subsequent geographic information data processing. Combining the above, a 30m high-precision NASADEM was used, and after GIS spatial analysis, the geographic information features corresponding to the DEM were obtained. Samples were generated using a small number of DEMs, micro-topography features, and geographic information features, and then fed into the convolutional neural network from the previous section for training, resulting in a general micro-topography classification model. The DEM and geographic information features of the study area were then input into this model to obtain the micro-topography classification results for the study area, thereby acquiring the micro-topography of each region and providing data reserves for subsequent research on wind speed correction coefficients.

[0105] In this embodiment, while traversing all DEM cases, the texture details of the DEM are preserved. Furthermore, dimensionality reduction is performed on a large number of samples using transformation group-invariant theory, with the degree of dimensionality reduction selectable by the user. Utilizing a parametric micro-topography classification system eliminates the manual intervention required in the original classification process. By using parametric pseudo-topography to judge, evaluate, and adjust the pre-classification accuracy, the quality of training samples is improved, and the manual selection of samples is eliminated.

[0106] The aforementioned three-dimensional complex micro-terrain identification method identifies micro-terrain categories by inputting the DEM data to be identified into a classification model. The classification model is obtained by using a DEM encoder and a data dimensionality reduction method based on Transformation Group to reduce the DEM data as a sample set, combined with a parameterized micro-terrain classification system and an automated pre-clustering classification algorithm for complex micro-terrain to train a convolutional neural network. This addresses the problems of incomplete and inflexible classification system coverage, complex texture and redundancy of DEM data, easy misclassification of micro-terrain, imbalanced sample size, need for manual intervention, and difficulty in determining the number of clusters.

[0107] Figure 13 This is a schematic block diagram of a three-dimensional complex micro-terrain recognition device 300 provided in an embodiment of the present invention. Figure 13As shown, corresponding to the above-described three-dimensional complex micro-terrain recognition method, the present invention also provides a three-dimensional complex micro-terrain recognition device 300. This three-dimensional complex micro-terrain recognition device 300 includes a unit for performing the above-described three-dimensional complex micro-terrain recognition method, and the device can be configured in a server. Specifically, please refer to... Figure 13 The three-dimensional complex micro-terrain recognition device 300 includes an acquisition unit 301 and a recognition unit 302.

[0108] Acquisition unit 301 is used to acquire DEM data to be identified;

[0109] The identification unit 302 is used to input the DEM data to be identified into the classification model to identify the micro-topography category, so as to obtain the corresponding micro-topography category; wherein, the classification model is obtained by using a DEM encoder and a data dimensionality reduction method based on Transformation Group to reduce the DEM data as a sample set, combined with a parameterized micro-topography classification system and a complex micro-topography automated pre-clustering classification algorithm to train a convolutional neural network.

[0110] In one embodiment, the three-dimensional complex micro-terrain recognition device 300 further includes a model generation unit, which is used to reduce the DEM data as a sample set by using a DEM encoder and a data dimensionality reduction method based on Transformation Group, and train a convolutional neural network by combining a parameterized micro-terrain classification system and an automated pre-clustering classification algorithm for complex micro-terrain to obtain a classification model.

[0111] In one embodiment, the model generation unit includes a system construction subunit, a data selection subunit, an encoding subunit, a dimensionality reduction subunit, a classification subunit, a fake terrain construction subunit, a judgment subunit, an analysis subunit, a training subunit, and an adjustment subunit.

[0112] The system comprises the following sub-units: a construction sub-unit for building a parametric micro-topography classification system; a data selection sub-unit for selecting DEM data that meets the requirements; an encoding sub-unit for encoding the DEM data using an improved Geomorphons algorithm to obtain DEM codes; a dimensionality reduction sub-unit for reducing the dimensionality of the DEM codes based on the Transformation Group algorithm to obtain dimensionality-reduced data; a classification sub-unit for specifying a number of categories and performing unsupervised classification of the dimensionality-reduced data using the k-means algorithm; a fake terrain construction sub-unit for inputting parameters, constructing fake terrain according to the parametric micro-topography classification system, and outputting the corresponding DEM codes of the fake terrain; a judgment sub-unit for determining whether there is only one parametric micro-topography classification category under each category of the corresponding k-means unsupervised classification based on the DEM codes of the fake terrain; and an analysis sub-unit for extracting geographic information features from the DEM data with determined micro-topography categories using GIS spatial analysis if there is only one parametric micro-topography classification category under each category of the corresponding k-means unsupervised classification, and combining the extracted features with the corresponding categories to form sample pairs. The training subunit is used to train the input convolutional neural network with samples and save the trained convolutional neural network to obtain a classification model. The adjustment subunit is used to adjust the number of classification categories if there is not only one parameterized micro-topography classification category under each category of the corresponding unsupervised k-means classification, and to perform unsupervised classification of the dimensionality-reduced data using the specified number of categories and the k-means algorithm.

[0113] In one embodiment, the fake terrain construction subunit includes a parameter setting module and a fake terrain encoding module.

[0114] The parameter setting module is used to set terrain parameters according to the parameterized micro-terrain classification system to obtain various different fake terrains; the fake terrain encoding module is used to encode the DEM data of various different fake terrains using the improved Geomorphons algorithm to obtain the DEM code corresponding to the fake terrain.

[0115] In one embodiment, the encoding subunit includes a comparison module and a conversion module.

[0116] The comparison module is used to perform pairwise comparisons of the DEM data and encode them to obtain a base-9 encoding; the conversion module is used to convert the base-9 encoding to a decimal encoding to obtain the DEM encoding.

[0117] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned three-dimensional complex micro-terrain recognition device 300 and each unit can be referred to the corresponding description in the foregoing method embodiments. For the sake of convenience and brevity, it will not be repeated here.

[0118] The aforementioned three-dimensional complex micro-terrain recognition device 300 can be implemented as a computer program, which can, for example... Figure 14 It runs on the computer device shown.

[0119] Please see Figure 14 , Figure 14 This is a schematic block diagram of a computer device provided in an embodiment of this application. The computer device 500 can be a server, wherein the server can be a standalone server or a server cluster composed of multiple servers.

[0120] See Figure 14 The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a system bus 501. The memory may include a non-volatile storage medium 503 and internal memory 504.

[0121] The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform a three-dimensional complex micro-terrain recognition method.

[0122] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0123] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can perform a three-dimensional complex micro-terrain recognition method.

[0124] This network interface 505 is used for network communication with other devices. Those skilled in the art will understand that... Figure 14 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 500 to which the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0125] The processor 502 is used to run a computer program 5032 stored in the memory to perform the following steps:

[0126] Obtain the DEM data to be identified; input the DEM data to be identified into the classification model to identify the micro-topography category, so as to obtain the corresponding micro-topography category;

[0127] The classification model is obtained by using a DEM encoder and a data dimensionality reduction method based on Transformation Group to reduce DEM data as a sample set, combined with a parameterized micro-topography classification system and an automated pre-clustering classification algorithm for complex micro-topography to train a convolutional neural network.

[0128] In one embodiment, when the processor 502 implements the step of using a DEM encoder and a Transformation Group-based data dimensionality reduction method to reduce DEM data as a sample set, and combining the results of training a convolutional neural network with a parameterized micro-topography classification system and an automated pre-clustering classification algorithm for complex micro-topography, the specific steps are as follows:

[0129] A parameterized micro-topography classification system is constructed; suitable DEM data is selected; the DEM data is encoded using an improved Geomorphons algorithm to obtain DEM codes; the DEM codes are dimensionality-reduced using the Transformation Group algorithm to obtain dimensionality-reduced data; a specified number of categories is assigned, and the dimensionality-reduced data is unsupervisedly classified using the k-means algorithm; input parameters are used, and fake terrain is constructed according to the parameterized micro-topography classification system, outputting the DEM codes corresponding to the fake terrain; based on the DEM codes corresponding to the fake terrain, it is determined whether there is only one parameterized micro-topography classification category under each category of the corresponding k-means unsupervised classification; if there is only one parameterized micro-topography classification category under each category of the corresponding k-means unsupervised classification, then geographic information features are extracted from the DEM data with determined micro-topography categories using GIS spatial analysis, and the extracted features are combined with the corresponding categories to form sample pairs. The sample pairs are input into a convolutional neural network for training, and the trained convolutional neural network is saved to obtain a classification model.

[0130] The convolutional neural network includes the AlexNet model.

[0131] The AlexNet model comprises five convolutional layers and three fully connected layers; it uses stacked convolutional layers to extract image features, employs Dropout and data augmentation to suppress overfitting, and uses ReLU as the activation function.

[0132] In one embodiment, after implementing the step of determining whether there is only one parameterized micro-topography classification category under each category of the corresponding k-means unsupervised classification based on the DEM encoding corresponding to the fake terrain, the processor 502 further implements the following steps:

[0133] If each category in the corresponding unsupervised k-means classification does not have only one parameterized micro-topography classification category, then adjust the number of classification categories and execute the specified number of categories, using the k-means algorithm to perform unsupervised classification on the dimensionality-reduced data.

[0134] In one embodiment, when the processor 502 implements the steps of inputting the input parameters, constructing a fake terrain based on the parameterized micro-terrain classification system, and outputting the DEM encoding corresponding to the fake terrain, it specifically implements the following steps:

[0135] Based on the parametric micro-terrain classification system, terrain parameters are set to obtain various different fake terrains; the DEM data of the various fake terrains are encoded using the improved Geomorphons algorithm to obtain the DEM codes corresponding to the fake terrains.

[0136] In one embodiment, when the processor 502 encodes the DEM data using the improved Geomorphons algorithm to obtain the DEM encoding step, it specifically implements the following steps:

[0137] The DEM data is compared pairwise and encoded to obtain a base-9 code; the base-9 code is then converted to a decimal code to obtain the DEM code.

[0138] It should be understood that in the embodiments of this application, the processor 502 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0139] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0140] Therefore, the present invention also provides a storage medium. This storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein when executed by a processor, the computer program causes the processor to perform the following steps:

[0141] Obtain the DEM data to be identified; input the DEM data to be identified into the classification model to identify the micro-topography category, so as to obtain the corresponding micro-topography category;

[0142] The classification model is obtained by using a DEM encoder and a data dimensionality reduction method based on Transformation Group to reduce DEM data as a sample set, combined with a parameterized micro-topography classification system and an automated pre-clustering classification algorithm for complex micro-topography to train a convolutional neural network.

[0143] In one embodiment, when the processor executes the computer program to implement the classification model, which involves using a DEM encoder and a Transformation Group-based data dimensionality reduction method to reduce DEM data as a sample set, and combining the results of training a convolutional neural network with a parameterized micro-topography classification system and an automated pre-clustering classification algorithm for complex micro-topography, the specific steps are as follows:

[0144] A parameterized micro-topography classification system is constructed; suitable DEM data is selected; the DEM data is encoded using an improved Geomorphons algorithm to obtain DEM codes; the DEM codes are dimensionality-reduced using the Transformation Group algorithm to obtain dimensionality-reduced data; a specified number of categories is assigned, and the dimensionality-reduced data is unsupervisedly classified using the k-means algorithm; input parameters are used, and fake terrain is constructed according to the parameterized micro-topography classification system, outputting the DEM codes corresponding to the fake terrain; based on the DEM codes corresponding to the fake terrain, it is determined whether there is only one parameterized micro-topography classification category under each category of the corresponding k-means unsupervised classification; if there is only one parameterized micro-topography classification category under each category of the corresponding k-means unsupervised classification, then geographic information features are extracted from the DEM data with determined micro-topography categories using GIS spatial analysis, and the extracted features are combined with the corresponding categories to form sample pairs. The sample pairs are input into a convolutional neural network for training, and the trained convolutional neural network is saved to obtain a classification model.

[0145] The convolutional neural network includes the AlexNet model.

[0146] The AlexNet model comprises five convolutional layers and three fully connected layers; it uses stacked convolutional layers to extract image features, employs Dropout and data augmentation to suppress overfitting, and uses ReLU as the activation function.

[0147] In one embodiment, after the processor executes the computer program to implement the step of determining whether there is only one parameterized micro-topography classification category under each category of the corresponding k-means unsupervised classification based on the DEM encoding of the fake terrain, the processor further implements the following steps:

[0148] If each category in the corresponding unsupervised k-means classification does not have only one parameterized micro-topography classification category, then adjust the number of classification categories and execute the specified number of categories, using the k-means algorithm to perform unsupervised classification on the dimensionality-reduced data.

[0149] In one embodiment, when the processor executes the computer program to implement the input parameters, constructs a fake terrain based on the parameterized micro-terrain classification system, and outputs the DEM encoding corresponding to the fake terrain, the processor specifically implements the following steps:

[0150] Based on the parametric micro-terrain classification system, terrain parameters are set to obtain various different fake terrains; the DEM data of the various fake terrains are encoded using the improved Geomorphons algorithm to obtain the DEM codes corresponding to the fake terrains.

[0151] In one embodiment, when the processor executes the computer program to implement the step of encoding the DEM data using the improved Geomorphons algorithm to obtain the DEM encoding, it specifically implements the following steps:

[0152] The DEM data is compared pairwise and encoded to obtain a base-9 code; the base-9 code is then converted to a decimal code to obtain the DEM code.

[0153] The storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0154] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0155] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0156] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the device of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0157] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0158] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A three-dimensional complex micro-terrain recognition method, characterized in that, include: Obtain the DEM data to be identified; The DEM data to be identified is input into a classification model to identify the micro-topography category, so as to obtain the corresponding micro-topography category; The classification model is obtained by using a DEM encoder and a data dimensionality reduction method based on Transformation Group to reduce DEM data as a sample set, combined with a parameterized micro-topography classification system and an automated pre-clustering classification algorithm for complex micro-topography to train a convolutional neural network. The classification model is obtained by using a DEM encoder and a data dimensionality reduction method based on Transformation Groups to reduce the DEM data as a sample set, combined with a parameterized micro-topography classification system and an automated pre-clustering classification algorithm for complex micro-topography to train a convolutional neural network, including: Construct a parametric micro-topography classification system; Select DEM data that meets the requirements; The DEM data is encoded using an improved Geomorphons algorithm to obtain DEM codes; The DEM encoding is dimensionality reduced based on the Transformation Group algorithm to obtain dimensionality-reduced data; Specify a number of categories and use the k-means algorithm to perform unsupervised classification on the dimensionality-reduced data. Input parameters, construct fake terrain according to the parameterized micro-terrain classification system, and output the DEM code corresponding to the fake terrain; Based on the DEM code corresponding to the fake terrain, determine whether there is only one parameterized micro-terrain classification category under each category of the corresponding unsupervised k-means classification; If there is only one parameterized micro-topography classification category under each category of the corresponding unsupervised k-means classification, then GIS spatial analysis is used to extract geographic information features from the DEM data with the determined micro-topography categories, and the extracted features and corresponding categories are combined to form sample pairs. The sample pairs are input into the convolutional neural network for training, and the trained convolutional neural network is saved to obtain a classification model.

2. The three-dimensional complex micro-terrain recognition method according to claim 1, characterized in that, After determining whether there is only one parameterized micro-topography classification category under each category of the corresponding k-means unsupervised classification based on the DEM encoding of the fake terrain, the method further includes: If each category in the corresponding unsupervised k-means classification does not have only one parameterized micro-topography classification category, then adjust the number of classification categories and execute the specified number of categories, using the k-means algorithm to perform unsupervised classification on the dimensionality-reduced data.

3. The three-dimensional complex micro-terrain recognition method according to claim 1, characterized in that, The input parameters, based on the parameterized micro-topography classification system, construct fake terrain and output the DEM code corresponding to the fake terrain, including: Based on the parametric micro-terrain classification system, terrain parameters can be set to obtain a variety of different fake terrains; An improved Geomorphons algorithm was used to encode the DEM data of various fake terrains to obtain the corresponding DEM codes.

4. The three-dimensional complex micro-terrain recognition method according to claim 1, characterized in that, The convolutional neural network includes the AlexNet model.

5. The three-dimensional complex micro-terrain recognition method according to claim 4, characterized in that, The AlexNet model comprises five convolutional layers and three fully connected layers; it uses stacked convolutional layers to extract image features, employs Dropout and data augmentation to suppress overfitting, and uses ReLU as the activation function.

6. The three-dimensional complex micro-terrain recognition method according to claim 1, characterized in that, The process of encoding the DEM data using the improved Geomorphons algorithm to obtain DEM encoding includes: The DEM data is compared pairwise and encoded to obtain a base-9 encoding. The 9-base code is converted to decimal code to obtain the DEM code.

7. A three-dimensional complex micro-terrain recognition device utilizing the three-dimensional complex micro-terrain recognition method according to any one of claims 1 to 6, characterized in that, include: The acquisition unit is used to acquire the DEM data to be identified; The identification unit is used to input the DEM data to be identified into the classification model to identify the micro-topography category, so as to obtain the corresponding micro-topography category; wherein, the classification model is obtained by using a DEM encoder and a data dimensionality reduction method based on Transformation Group to reduce the DEM data as a sample set, combined with a parameterized micro-topography classification system and a complex micro-topography automated pre-clustering classification algorithm to train a convolutional neural network.

8. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, can implement the method as described in any one of claims 1 to 6.

Images