A full-automatic recognition and classification method for deep-sea geomorphology based on contour pattern entropy

Abstract

This invention discloses a fully automated method for identifying and classifying deep-sea landforms based on isobath morphological entropy, relating to the fields of marine geological exploration and topographic analysis. The method includes: extracting isobaths from a digital elevation model of the target area at predetermined intervals and constructing an isobath topology map to identify closed loops and bifurcation points; and calculating multi-dimensional isobath morphological entropy sub-factors based on the isobath topology map, including isobath topological complexity index, isobath geometric entropy, isobath density gradient anomaly, and isobath curvature spectrum characteristics. This invention, by calculating multi-dimensional morphological entropy sub-factors encompassing isobath topological complexity, geometric entropy, density gradient anomaly, and curvature spectrum characteristics, provides a quantitative description of the topological structure, spatial distribution, and frequency domain characteristics of deep-sea landforms. Compared to traditional single-topographic factor analysis methods, it can capture multi-level topographic information from microscopic morphology to macroscopic patterns, significantly improving the comprehensive characterization ability of complex landform structures.

Description

Technical Field

[0001] This invention relates to the field of marine geological exploration and topographic analysis technology, specifically to a fully automatic method for identifying and classifying deep-sea landforms based on isobath morphological entropy. Background Technology

[0002] With the advancement of technology, especially the improvement of remote sensing technology and data processing capabilities, it is possible to acquire a large amount of deep-sea multi-band data, which usually contains complex spatial morphological information. As an effective tool for depicting seabed topography, isobaths can provide specific characteristics of deep-sea landforms. By analyzing the morphological entropy of isobaths, the complexity and diversity of landforms can be quantified, thereby providing a scientific basis for the classification of deep-sea landforms. As an indicator for measuring information uncertainty, morphological entropy can effectively reflect the characteristics of different seabed topography, thus playing a key role in the fully automated identification process.

[0003] The deep sea covers approximately 65% ​​of the Earth's surface, and its complex geomorphological features (such as seamounts, canyons, hydrothermal vents, cold seeps, and pockmarks) record crucial information about the Earth's tectonic evolution, resource distribution, and ecological environment. Current deep-sea geomorphological identification technologies face the following bottlenecks: First, they are highly dependent on human intervention: traditional methods rely heavily on marine geologists manually interpreting multibeam and side-scan sonar data, resulting in low efficiency, strong subjectivity, and difficulty in processing massive amounts of data. The manual interpretation cycle typically takes weeks to months, failing to meet the real-time requirements of modern marine exploration. Second, they lack sufficient feature representation: existing automated methods are mostly based on simple local topographic factors (such as slope and curvature), which can only characterize the terrain. The methods suffer from several drawbacks: First, they fail to capture the macroscopic structure and topological features of landforms, resulting in low accuracy in identifying complex landforms. Second, they lack multi-scale adaptability: deep-sea landforms exhibit significant multi-scale characteristics, and existing methods lack effective multi-scale analysis mechanisms, making it difficult to simultaneously identify large-scale basin features and fine micro-landform structures. Third, they are inaccurate in boundary identification: the boundaries of landform units are often key areas where topographic changes occur, and existing methods are not sensitive enough to boundaries, making it difficult to accurately delineate transition zones between different landform types. Fourth, they have weak interpretability: although deep learning-based "black box" methods perform well in certain scenarios, their decision-making process lacks clear physical and geological significance, making it difficult to combine with professional geological knowledge.

[0004] In existing technologies, traditional methods struggle to effectively distinguish subtle geomorphic differences, such as seamounts and hydrothermal vent areas, during fully automated identification and classification of deep-sea landforms. This leads to inaccurate identification of resource exploration targets. Furthermore, even with similar topological structures, the delineation of fault zones and sedimentary basin boundaries remains ambiguous, affecting the precise location of tectonic units. Moreover, for canyon systems formed by multiple superpositions, existing methods struggle to separate geomorphic units from different evolutionary stages, hindering the detailed reconstruction of the seafloor geomorphic evolution history. To address these issues, a fully automated deep-sea landform identification and classification method based on isobath morphological entropy is proposed. Summary of the Invention

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a fully automatic identification and classification method for deep-sea landforms based on isobath morphological entropy, comprising the following steps:

[0006] Step S1: Extract contour lines from the digital elevation model of the target area at set intervals, construct a topological relationship map of contour lines, identify closed loops and bifurcation points, realize the preliminary quantification of the terrain structure, and provide a spatial topological basis for subsequent morphological entropy calculation;

[0007] Step S2: Based on the isobath topology map, calculate the multi-dimensional isobath morphology entropy (CME) sub-factor, which includes the isobath topology complexity index, isobath geometric entropy, isobath density gradient anomaly, and isobath curvature spectrum characteristics. This multi-dimensional analysis of terrain complexity and spatial variation enhances the ability to distinguish landforms.

[0008] Step S3: Calculate the isobath morphological entropy sub-factors at multiple analysis window scales, and obtain the comprehensive isobath morphological entropy index through weighted fusion. This integrates multi-scale topographic information to improve the recognition adaptability of geomorphic units of different scales.

[0009] Step S4: Integrate the morphological entropy of the comprehensive isobaths with traditional topographic factors, and organize them into a high-dimensional geomorphic feature matrix after standardization to construct a comprehensive and comparable feature system to support high-precision automatic classification;

[0010] Step S5: Principal component analysis is used to reduce the dimensionality of the high-dimensional geomorphic feature matrix, retaining the main components whose cumulative contribution rate exceeds the threshold, removing redundant information, and improving computational efficiency and model stability.

[0011] Step S6: Based on the dimensionality-reduced high-dimensional landform features, unsupervised clustering is performed using a density-based clustering algorithm to adaptively determine the number of landform categories, thereby achieving automatic discovery of natural landform groupings and avoiding the subjectivity of manually pre-setting categories.

[0012] Step S7: Based on the feature vectors corresponding to each cluster center, semantic annotation is performed in conjunction with prior geomorphological knowledge. Then, image processing methods such as morphological opening and closing operations are used to optimize the spatial morphology of the classification map to obtain automated geomorphological classification results.

[0013] Preferably, step S1 specifically includes the following steps:

[0014] The system acquires the digital elevation model of the target area, extracts contour lines from the digital elevation model at preset depth intervals, and records the geometric coordinates and depth attributes of each contour line, thereby systematically extracting discrete contour information reflecting depth changes from the continuous terrain surface.

[0015] Based on the extracted geometric coordinates and depth attributes of the isobaths, a topological relationship map of the isobaths is constructed, closed isobath loops and bifurcation nodes are identified, and their spatial locations and connectivity are recorded. A topological network between isobaths is established, clearly revealing the structural characteristics and spatial relationships of geomorphic units.

[0016] The geometric properties of each contour line are calculated, including length, average curvature, and discretized orientation angle sequence, to provide basic data for subsequent morphological entropy calculation, quantify the basic morphological characteristics of the contour lines, and provide a computable standardized input for complex terrain information measurement.

[0017] Preferably, step S2 specifically includes the following steps:

[0018] Based on the topological relationship map of isobaths, and combined with the number of closed loops and the density of bifurcation points per unit area, the topological complexity index of isobaths is calculated to reflect the topological complexity of the terrain structure and is used to distinguish the topological connection structure of positive landforms (seamounts) and negative landforms (depressions).

[0019] Each isobath is resampled to generate a direction angle sequence, and its Shannon entropy is calculated to obtain the geometric entropy of the isobath, which characterizes the randomness and irregularity of the isobath direction distribution and can effectively identify the morphological differences between linear structures (fracture zones) and meandering river systems.

[0020] The gradient magnitude is calculated based on the isobath density field, and the isobath density gradient anomaly is obtained through normalization to indicate the local variation in the spatial aggregation of isobaths, which helps to accurately locate transition zones such as canyon sidewalls, slope toes, and fan fronts.

[0021] Preferably, step S2 further includes:

[0022] Fourier transform is performed on the high-order sequence of a single isobath to calculate its power spectrum, and the low-frequency energy ratio and high-frequency energy ratio are extracted. Based on the distribution of the power spectrum in the logarithmic coordinate system, the spectral slope is calculated by fitting, which characterizes the scale characteristics of the isobath undulation, quantitatively separates the macroscopic pattern and microscopic details of the landform, and provides frequency domain basis for evolution stage analysis.

[0023] The low-frequency energy ratio, high-frequency energy ratio and spectral slope are combined to form the isobath curvature spectrum feature, which is used to describe the multi-scale frequency domain structure of isobath morphology, forming a compact digital fingerprint that can comprehensively reflect the multi-scale undulation structure of the terrain.

[0024] By integrating the calculated isobath topological complexity index, isobath geometric entropy, isobath density gradient anomaly, and isobath curvature spectrum features, a multi-dimensional isobath morphological entropy sub-factor sequence is formed, constructing a complete feature set from topology, geometry, statistics to the frequency domain, and comprehensively characterizing terrain complexity.

[0025] Preferably, step S3 specifically includes the following steps:

[0026] Multiple analysis windows of different scales are set up, and multi-dimensional isobath morphological entropy sub-factors are calculated in each window to comprehensively capture multi-level topographic information from local micro-topography to regional macro-structure.

[0027] We assign weight coefficients to the isobath morphology entropy sub-factors at various scales to reflect the differences in contributions of different scales to landform identification. Through local statistical aggregation, we effectively extract the dominant features of terrain structure at different spatial scales. By weighted fusion of the isobath morphology entropy sub-factors at various scales, we generate a comprehensive isobath morphology entropy index, realizing the integrated expression of multi-scale terrain information and enhancing the comprehensive representation ability of complex landforms.

[0028] Preferably, step S4 specifically includes the following steps:

[0029] Traditional topographic factors of the target area are extracted, including slope, aspect, profile curvature, plane curvature, roughness, and topographic relief, to comprehensively depict the local geometry and macroscopic undulation characteristics of the land surface and provide basic topographic information for landform classification.

[0030] By integrating the comprehensive contour morphological entropy index with traditional topographic factors, a feature set containing multi-source topographic description information is formed, which enhances the multidimensionality and complementarity of feature expression and improves the ability to distinguish complex landforms.

[0031] The fused features are standardized and organized into a high-dimensional landform feature matrix by pixel, which serves as the input data for landform classification. This eliminates dimensional differences, builds a unified and comparable numerical basis, and ensures the stable operation of subsequent classification models.

[0032] Preferably, step S5 specifically includes the following steps:

[0033] Principal component analysis was performed on the high-dimensional geomorphic feature matrix to extract the feature principal components and calculate the eigenvalues ​​and contribution rates of each principal component. The extracted principal components effectively revealed the most important variation directions in the original dataset.

[0034] Calculate the variance contribution rate of each principal component, set a cumulative contribution rate threshold, and select the number of principal components to retain based on whether the cumulative contribution rate exceeds the preset cumulative contribution rate threshold. Adaptively determine the number of principal components based on the threshold to ensure that the data retains most of the effective information after dimensionality reduction and avoid information loss.

[0035] Principal components exceeding the cumulative contribution rate threshold are retained, and the original features are projected onto the feature subspace formed by the selected principal components to achieve data dimensionality reduction and redundancy removal, thereby obtaining the dimensionality-reduced feature dataset. The projection operation maps high-dimensional data to a low-dimensional space, eliminates redundant features, and significantly improves the computational efficiency and stability of subsequent clustering.

[0036] Preferably, step S6 specifically includes the following steps:

[0037] Based on the dimensionality-reduced feature dataset, a density-based spatial clustering algorithm is used to perform unsupervised clustering analysis on the dimensionality-reduced feature data, so as to automatically discover the hidden natural grouping structure in the data when the landform type is unknown.

[0038] By optimizing clustering parameters that encompass clustering radius and minimum sample size, the spatial clustering algorithm adaptively determines the natural number of groupings for landform categories, i.e., the optimal number of categories. This avoids over-segmentation or under-segmentation problems that may result from artificially pre-setting the number of categories. Furthermore, spatial neighborhood constraints are introduced during the clustering process to ensure spatial continuity of the same landform category, preventing fragmented classification patches and improving the rationality and interpretability of the classification results in geospatial context.

[0039] Preferably, step S7 specifically includes the following steps:

[0040] The feature vectors of each cluster center are extracted and semantic matching and type labeling are performed in combination with the prior geomorphological knowledge base. This effectively overcomes the limitation of unsupervised clustering results lacking actual geomorphological significance and realizes the geological interpretability transformation of data-driven classification results.

[0041] The numerical clustering results are transformed into classification maps with landform type labels, completing the transformation from data clustering to semantic interpretation. Abstract numerical groups are mapped to specific landform type maps, providing an intuitive and understandable base map for subsequent geological analysis and applications. Morphological opening and closing operations are performed on the preliminary classification results to identify spatial regions that need post-processing, so as to smooth the boundaries between classes and suppress small noise patches, significantly improving the spatial continuity and visual quality of the classification patches, and eliminating the interference of discrete noise points and small holes on the mapping effect.

[0042] Preferably, step S7 further includes:

[0043] Based on the morphologically processed classification map, region merging and boundary optimization are performed to improve the internal homogeneity and boundary consistency of each category region. The classification results are spatially overlaid and verified with the original terrain data to ensure that the classification results match the actual terrain features, effectively eliminate salt and pepper noise and broken patches in the classification map, and improve the overall map and visual coherence.

[0044] Based on spatial overlay and verification, the final output is an automated geomorphological classification result with clear geomorphological semantics, spatial continuity and optimized morphology. It provides a standardized geomorphological classification product and supporting documents that are structurally complete, semantically clear and can be directly used in business.

[0045] This invention provides a fully automated method for identifying and classifying deep-sea landforms based on isobath morphological entropy. It offers the following advantages:

[0046] (I) This fully automatic identification and classification method for deep-sea landforms based on isobath morphological entropy achieves a comprehensive quantitative description of the topological structure, spatial distribution and frequency domain characteristics of deep-sea landforms by comprehensively calculating multi-dimensional morphological entropy sub-factors of isobath topological complexity, geometric entropy, density gradient anomaly and curvature spectrum characteristics. Compared with traditional single topographic factor analysis methods, it can capture multi-level topographic information from micromorphology to macro pattern more precisely, and significantly improve the comprehensive representation ability of complex landform structures.

[0047] (II) This fully automatic identification and classification method for deep-sea landforms based on isobath morphological entropy introduces a multi-scale analysis mechanism, calculates and integrates morphological entropy indices at multiple characteristic window scales, effectively solves the problem of the difficulty in uniformly representing the multi-scale characteristics of deep-sea landforms, and can simultaneously identify large-scale ocean basin structural features and local micro-landform morphology, thereby enhancing the adaptability of the landform classification system to topographic patterns at different scales.

[0048] (III) This fully automatic identification and classification method for deep-sea landforms based on isobath morphological entropy organically integrates isobath morphological entropy features with traditional topographic factors to construct a high-dimensional feature set with complementary information. It retains the basic descriptive ability of traditional topographic factors and incorporates the unique topological and structural information of isobaths, which significantly improves the discriminative power and classification accuracy of landform features.

[0049] (iv) This fully automatic identification and classification method for deep-sea landforms based on isobath morphological entropy adopts a density-based unsupervised clustering algorithm, combined with spatial neighborhood constraints and adaptive parameter optimization mechanism, to realize the autonomous discovery and division of landform categories. It does not require pre-setting the number of categories and can naturally form landform groups according to the distribution characteristics of the data itself. At the same time, spatial constraints ensure the geographical rationality of the classification results. Attached Figure Description

[0050] Figure 1 This is a schematic diagram illustrating the workflow of a fully automatic deep-sea landform identification and classification method based on isobath morphological entropy according to the present invention.

[0051] Figure 2 This is the original DEM data diagram of the present invention;

[0052] Figure 3 This is a diagram showing the calculation results of the comprehensive contour morphological entropy index of the present invention;

[0053] Figure 4 The image shows the automatic landform classification results of this invention, where Type I represents foothills, Type II represents depressions, Type III represents ridges, Type IV represents shoulders, and Type V represents slopes. Detailed Implementation

[0054] 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 embodiments of the present invention, and not all embodiments. 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.

[0055] Example 1, please refer to Figure 1 This invention provides a technical solution: a fully automatic identification and classification method for deep-sea landforms based on isobath morphological entropy, comprising the following steps:

[0056] Step S1: Extract isobaths from the digital elevation model of the target area at set intervals, construct an isobath topology map, identify closed loops and bifurcation points, and achieve preliminary quantification of the terrain structure, providing a spatial topological basis for subsequent morphological entropy calculation. Obtain the digital elevation model of the target area, extract isobaths from the digital elevation model at preset depth intervals, and record the geometric coordinates and depth attributes of each isobath, realizing the systematic extraction of discrete contour information reflecting depth changes from the continuous terrain surface. Based on the extracted isobath geometric coordinates and depth attributes, construct an isobath topology map, identify closed isobath loops and bifurcation nodes, and record their spatial location and connectivity, establish a topological network between isobaths, clearly reveal the structural characteristics and spatial relationships of geomorphic units, calculate the geometric attributes of each isobath, including length, average curvature, and discretized orientation angle sequence, providing basic data for subsequent morphological entropy calculation, quantifying the basic morphological characteristics of isobaths, and providing calculable standardized input for complex terrain information measurement.

[0057] The specific work involves: First, acquiring a high-resolution digital elevation model (DEM) covering the target sea area. Its planar resolution is typically 10 to 100 meters, with a vertical accuracy of no less than 0.1% of the water depth. Based on a preset isobath interval (dynamically adjusted according to terrain complexity and data resolution), contour line extraction algorithms are used to extract isobaths layer by layer from the DEM. The geometric coordinates of each isobath are stored as a vector sequence and associated with its corresponding depth attribute value. To ensure topological integrity, linear interpolation is used to handle missing data areas during extraction, and the Douglas-Peucker algorithm is used to compress redundant nodes, controlling the data volume while maintaining morphological accuracy. This results in an output isobath vector dataset containing complete geometric information and attribute tables. Based on the extracted isobath vector data, a planar graph theory method is used to construct an isobath topology graph. Spatial indexing accelerates neighborhood queries, identifying closed isobath loops (such as seamount tops and basin centers) and bifurcation nodes (such as canyon confluences and fault zone branch points), recording their spatial locations and connections. Based on this, the geometric attributes of each contour line are calculated, including length, average curvature, and orientation angle sequence. Length is obtained by accumulating the Euclidean distance between adjacent nodes. Average curvature is calculated segment by segment using the three-point method and then averaged. The orientation angle sequence is generated by resampling the contour lines at a fixed step size and discretizing the azimuth angles of adjacent segments (dividing them into 8 or 16 orientation intervals). All geometric attributes are organized and stored according to the contour line ID, forming an attribute matrix. Multi-level quality control is implemented during the topology construction and geometric calculation stages. Consistency checks are performed on the topology graph to eliminate anomalies such as self-intersections and overlapping segments. A sliding window filter (window size of 3-5 nodes) is used to smooth the orientation angle sequence and suppress the influence of local noise on entropy calculation. During geometric attribute calculation, a curvature threshold is set to filter abnormally curved segments, and contour lines with a length less than three times the sampling interval are marked. All processing steps retain original data and correction records, supporting result traceability and parameter optimization. The final output is a structured dataset containing a topology graph, geometric attribute table, and quality control report, meeting the data requirements for multi-dimensional morphological entropy calculation.

[0058] Step S2: Based on the isobath topology map, calculate the multi-dimensional isobath morphological entropy (CME) sub-factor, which includes the isobath topology complexity index, isobath geometric entropy, isobath density gradient anomaly, and isobath curvature spectrum characteristics. This multi-dimensional model characterizes terrain complexity and spatial variation, enhancing the ability to distinguish landforms. Based on the isobath topology map, combined with the number of closed loops and the density of bifurcation points per unit area, calculate the isobath topology complexity index to reflect the topological complexity of the terrain structure. This index is used to distinguish the topological connection structure of positive landforms (seamounts) and negative landforms (depressions). Resample each isobath to generate a direction angle sequence, calculate its Shannon entropy, and obtain the isobath geometric entropy. This entropy characterizes the randomness and irregularity of the isobath direction distribution and can effectively identify the morphological differences between linear structures (fault zones) and meandering river systems. Calculate the gradient magnitude based on the isobath density field, and further obtain the isobath density gradient anomaly through standardization to indicate the local variation in the spatial aggregation degree of isobaths. This helps to accurately locate transition zones such as canyon sidewalls, slope toes, and fan fronts.

[0059] The specific work involves: based on the isobath topology map, constructing an isobath topology complexity index by quantifying the spatial distribution characteristics of closed loops and bifurcation points. Specifically, a basic statistical unit is used (1 square kilometer). The number of closed isobath loops and bifurcation points within each unit is calculated. Bifurcation points are identified using a neighborhood search algorithm with a search radius twice the current isobath interval. The number of closed loops is normalized to the unit area. The number of bifurcation points is further combined with the total length of isobaths within the unit area to calculate a density index. These two indices are linearly weighted and fused, with weighting coefficients preset according to terrain type, ultimately generating a topology complexity index map with a spatial resolution of 100 meters. The isobath topology complexity index directly reflects the complexity of the topological connections of the seabed topography, providing a quantitative basis for identifying tectonic landforms such as fault zones and seamounts. To characterize the irregularity of the isobath geometry, geometric entropy is calculated for each isobath. Fixed-step resampling is used, with the step size set to 1.2 times the average node spacing of the isobaths to ensure uniform sampling and preserve morphological features. After resampling, a direction angle sequence is obtained, divided into 16 equal partitions. Discretization was performed with an interval spanning 22.5 degrees. The frequency distribution of each directional angle interval was statistically analyzed. Shannon entropy values ​​were calculated using a base-2 logarithm. During the calculation, contour lines with fewer than 10 sampling points were discarded to avoid statistical errors. The entropy values ​​of all valid contour lines were recorded according to their spatial location. Geometric entropy of the contour lines was obtained and a geometric entropy distribution map was generated to effectively distinguish between smooth depositional landforms and eroded landforms. By analyzing the spatial clustering characteristics of contour lines, the density gradient anomaly index of contour lines was calculated. First, a density field was constructed based on the contour line vector data, using a circular... The neighborhood statistical method was used, with a neighborhood radius of 500 meters. The density field grid was generated by the ratio of the total length of the isobaths to the area within the statistical unit, with a spatial resolution of 50 meters. The Sobel operator gradient was used to calculate the gradient magnitude map of the density field. Then, the gradient magnitude of the entire map was normalized by Z-Score with a window size of 3×3 pixels to generate the final density gradient anomaly map. This map highlights the transition zone between dense and sparse isobaths and has a good indicative effect on identifying geomorphic boundaries with significant gradient changes, such as the sidewalls of submarine canyons and the front edges of sedimentary fans.

[0060] The expression for calculating the topological complexity index of contour lines is as follows:

[0061] ;

[0062] in: It is an index of topological complexity of contour lines; The number of closed isobath loops per unit area; The number of isobath bifurcation points per unit area; The total length of isobaths per unit area; The branch point weight coefficient; To prevent division by zero of small constants;

[0063] The geometric entropy of contour lines is obtained by resampling each contour line to obtain a fixed number of orientation angle sequences, and then calculating the Shannon entropy of the orientation angle sequences. The calculation expression is as follows:

[0064] ;

[0065] in: The geometric entropy of the contour lines; For the direction of the corner into the first The probability of each interval;

[0066] The formula for calculating the density gradient anomaly along the isobath is as follows:

[0067] ;

[0068] in: This is an isobath density gradient anomaly. and They are respectively The global mean and standard deviation; The density gradient magnitude is calculated using the following formula:

[0069] ;

[0070] in: Let be the gradient vector of the density field. For density field in Partial derivatives in direction, For density field in Partial derivatives in direction, The linear density field at isobath depth is calculated using the following formula:

[0071] ;

[0072] in: For the point Linear density at depth, For Center, radius The circular neighborhood, The Dirac function is used to select the depth. contour lines, For the area of ​​a microelement, The neighborhood radius, neighborhood radius The neighborhood, For depth surface variables, This represents the current depth value of the iso-depth surface;

[0073] Furthermore, step S2 also includes: performing a Fourier transform on the high-order sequence of a single isobath, calculating its power spectrum, and extracting the low-frequency energy ratio and high-frequency energy ratio. Based on the distribution of the power spectrum in the logarithmic coordinate system, fitting and calculating the spectral slope to characterize the scale characteristics of the isobath undulations, quantitatively separating the macroscopic pattern and microscopic details of the landform, providing frequency domain basis for evolution stage analysis, and using the low-frequency energy ratio, high-frequency energy ratio, and spectral slope together to form the isobath curvature spectrum feature, which is used to describe the multi-scale frequency domain structure of the isobath morphology, forming a compact digital fingerprint that can comprehensively reflect the multi-scale undulation structure of the terrain. Integrating the calculated isobath topological complexity index, isobath geometric entropy, isobath density gradient anomaly, and isobath curvature spectrum feature, forming a multi-dimensional isobath morphological entropy sub-factor sequence, and constructing a complete feature set from topology, geometry, statistics to the frequency domain to comprehensively characterize the complexity of the terrain.

[0074] The specific work involves: extracting the high-order sequence of contour lines one by one and performing spectral analysis to quantitatively characterize the multi-scale structural features of contour line morphology in the frequency domain; resampling the contour lines at a fixed step size to ensure that the number of sampling points N is not less than 128 points to guarantee spectral resolution; adaptively determining the sampling step size based on the original contour line node density, controlling it between 20 meters and 100 meters; performing a fast Fourier transform on the high-order sequence of each contour line to obtain its complex spectrum; calculating the power spectral density based on the spectral amplitude; setting the low-frequency band to the first 1 / 8 of the total frequency band and the high-frequency band to the last 1 / 8; and calculating the cumulative proportion of low-frequency energy and the cumulative proportion of high-frequency energy to characterize the contribution of macroscopic trends and microscopic details of topographic relief. Furthermore, in a double logarithmic coordinate system, least-squares linear fitting was performed on the power spectrum in the 1 / 8 to 1 / 2 frequency band. The negative slope of the resulting fitted line is the spectral slope parameter, effectively reflecting the attenuation characteristics of topographic relief with spatial scale changes. All spectral parameter calculations used Hanning window preprocessing to suppress spectral leakage. Based on the frequency domain analysis results, the low-frequency energy ratio, high-frequency energy ratio, and spectral slope of each isobath are used to construct the curvature spectrum feature vector of that isobath. A low-frequency energy ratio greater than 0.6 indicates that the topography is dominated by long-wave undulations, commonly found in large gentle slopes or basins; a high-frequency energy ratio greater than 0.3 indicates rich local details, often seen in eroded areas or complex tectonic zones; the spectral slope is between -1.5 and - Between 3.5 and 3.5, a larger negative value indicates more significant small-scale fluctuations. After extracting features from all isobaths, the curvature spectrum feature vectors are associated with geographic coordinates according to the spatial distribution of isobaths. Low-frequency energy ratio distribution maps, high-frequency energy ratio distribution maps, and spectral slope distribution maps are generated using the Kriging spatial interpolation method. The interpolation grid resolution is consistent with the original DEM, and the interpolation search radius is set to 3 times the average spacing of isobaths to ensure spatial continuity and avoid excessive smoothing. The calculated isobath curvature spectrum features are integrated with the three isobath morphological entropy sub-factors—the isobath topological complexity index, the isobath geometric entropy, and the isobath density gradient anomaly—all of which are uniformly sampled to the same spatial grid. The system uses a uniform grid size of 50m x 50m. Non-grid data is resampled using bilinear interpolation. Each grid cell corresponds to a four-dimensional feature vector, which includes the topological complexity index, geometric entropy, density gradient outliers, and curvature spectrum features composed of low-frequency energy ratio, high-frequency energy ratio, and spectral slope. All features need to be standardized before integration. The Z-score method is used to normalize the data with the entire target area as the statistical range. The resulting multi-dimensional isobath morphological entropy sub-factor sequence is stored in the form of a three-dimensional array. The data structure includes a planar coordinate index and a four-dimensional feature vector, along with complete metadata descriptions, including the calculation methods of various parameters, spatial reference information, and standardized parameters.

[0075] The expression for calculating the curvature spectrum characteristics of isobaths is as follows:

[0076] Let the sampling of a single contour line be the contour line height sequence. After Fourier transform, we can obtain:

[0077] ;

[0078] Then calculate the power spectrum. Its formula is:

[0079] ;

[0080] Next, calculate the spectral characteristic quantities (take...). ; );

[0081] The low-frequency energy ratio for:

[0082] ;

[0083] High frequency energy ratio for:

[0084] ;

[0085] The spectral slope can then be calculated. :

[0086] ;

[0087] in , Pick , where is the number of frequency points used for fitting; The sequence represents the elevation sequence of contour lines, indicating the th contour line. Elevation values ​​of each sampling point; The sampling point number is from 0 to 1. ; This represents the total number of sampling points; The complex value after the discrete Fourier transform represents the frequency. spectral components; For frequency index (wavenumber), indicating spatial frequency; The imaginary unit; For frequency The corresponding power spectral density (energy); For complex numbers The magnitude (amplitude); Low-frequency energy accounts for a significant portion of the total energy (previous) The proportion of (each frequency); This is a low-frequency cutoff index; For Nyquist frequency index (highest analyzable frequency); The proportion of high-frequency energy to total energy; The slope of the power spectrum in log-log coordinates reflects the scale characteristics of topographic relief.

[0088] Step S3: Calculate the isobath morphological entropy sub-factors at multiple analysis window scales, and obtain the comprehensive isobath morphological entropy index through weighted fusion. This integrates multi-scale topographic information to improve the recognition adaptability of landform units of different scales. Multiple analysis windows at different scales are set, and multi-dimensional isobath morphological entropy sub-factors are calculated in each window to comprehensively capture multi-level topographic information from local micro-landforms to regional macro-structures. Weight coefficients are assigned to the isobath morphological entropy sub-factors at each scale to reflect the differences in contribution of different scales to landform recognition. Through local statistical aggregation, the dominant features of topographic structure at different spatial scales are effectively extracted. By weighted fusion of the isobath morphological entropy sub-factors at each scale, a comprehensive isobath morphological entropy index is generated, realizing the integrated expression of multi-scale topographic information and enhancing the comprehensive representation ability of complex landforms.

[0089] The specific work involves: to achieve multi-scale perception of deep-sea topography from microscopic morphology to macroscopic patterns, three characteristic analysis window scales are set: 500-meter, 2000-meter, and 5000-meter square windows. In practice, each pixel of the target area's digital elevation model is used as the center, and moving windows of the three scales are applied sequentially. Within each window, based on the generated standardized multi-dimensional isobath morphological entropy sub-factor raster data, the in-window statistics of each sub-factor are calculated. For the isobath topological complexity index and isobath geometric entropy, their in-window mean is calculated to reflect the overall characteristics of that scale; for the isobath density gradient anomaly, its in-window standard deviation is calculated to highlight its variability; for the low-frequency energy ratio, high-frequency energy ratio, and spectral slope of the isobath curvature spectrum characteristics, their window values ​​are calculated respectively. The median was used to enhance robustness to outliers. Boundary padding was employed in all calculations to ensure the integrity of regional edge information. Finally, three sets of multi-dimensional morphological entropy feature vectors corresponding to different scales were generated for each pixel. An adaptive weighting strategy was adopted to assign weight coefficients to the isobath morphological entropy sub-factors at each scale, including the isobath topological complexity index, isobath geometric entropy, isobath density gradient anomaly, isobath curvature spectrum feature-low frequency energy, and isobath curvature spectrum feature-high frequency energy ratio. This quantified the difference in contribution of different analysis scales to landform identification. The values ​​of each sub-factor corresponding to the three scales on the same pixel were multiplied by their corresponding scale weights and then summed one by one to generate the comprehensive isobath morphological entropy index value of the pixel, forming a new feature vector that integrates multi-scale information.

[0090] The formula for calculating the entropy index of contour lines is as follows:

[0091] ;

[0092] In the formula: It is the morphological entropy of isobaths, a fusion index that integrates multi-scale and multi-dimensional terrain features; The weight coefficient represents the importance weight of each sub-feature in the fusion process. It is the topological complexity index of isobaths, reflecting the topological complexity of isobath closed loops and bifurcation points; The geometric entropy of isobaths, based on the Shannon entropy of the isobath direction angle sequence, describes the randomness of the geometric morphology; This is an isobath density gradient anomaly, reflecting an anomalous spatial gradient variation of the isobath density field; The low-frequency energy ratio of the curvature spectrum of isobaths reflects the energy proportion of low-frequency components in the high-order sequence of isobaths, corresponding to large-scale undulations. The high-frequency energy ratio of the curvature spectrum of isobaths reflects the energy proportion of high-frequency components in the isobath high-degree sequence, corresponding to small-scale details.

[0093] Step S4: Integrate the morphological entropy of the comprehensive isobaths with traditional topographic factors, and organize them into a high-dimensional geomorphic feature matrix after standardization to construct a comprehensive and comparable feature system to support high-precision automatic classification;

[0094] Step S5: Principal component analysis is used to reduce the dimensionality of the high-dimensional geomorphic feature matrix, retaining the main components whose cumulative contribution rate exceeds the threshold, removing redundant information, and improving computational efficiency and model stability.

[0095] Step S6: Based on the dimensionality-reduced high-dimensional landform features, unsupervised clustering is performed using a density-based clustering algorithm to adaptively determine the number of landform categories, thereby achieving automatic discovery of natural landform groupings and avoiding the subjectivity of manually pre-setting categories.

[0096] Step S7: Based on the feature vectors corresponding to each cluster center, semantic annotation is performed in conjunction with prior geomorphological knowledge. Then, image processing methods such as morphological opening and closing operations are used to optimize the spatial morphology of the classification map, resulting in automated geomorphological classification results. The output is a classification map with clear geomorphological significance, spatial continuity, and reasonable morphology, which is convenient for practical applications.

[0097] Example 2, as Figure 1As shown, based on Embodiment 1, the present invention provides a technical solution: Step S4 specifically includes the following steps: extracting traditional topographic factors of the target area, including slope, aspect, profile curvature, plane curvature, roughness, and topographic relief, comprehensively depicting the local geometric shape and macroscopic undulation characteristics of the land surface, providing basic topographic information for landform classification, fusing the comprehensive isobath morphological entropy index with traditional topographic factors to form a feature set containing multi-source topographic description information, enhancing the multidimensionality and complementarity of feature expression, improving the ability to distinguish complex landforms, standardizing the fused features, and organizing them into a high-dimensional landform feature matrix according to pixels, as input data for landform classification, eliminating dimensional differences, constructing a unified and comparable numerical basis, and ensuring the stable operation of subsequent classification models;

[0098] The specific work involves extracting six core traditional topographic factors based on a digital elevation model covering the target sea area: slope, aspect, profile curvature, plane curvature, roughness, and topographic relief. Slope and aspect are calculated using the Horn algorithm within a 3×3 moving window. The aspect calculation results are then cosine transformed to eliminate directional periodicity. Profile curvature and plane curvature are obtained using a quadratic surface fitting method with a 5×5 pixel fitting window to accurately reflect the vertical and horizontal concavity and convexity characteristics of the terrain. Roughness is obtained by calculating the standard deviation between the elevation of the central pixel and the surrounding eight pixels within the 3×3 window, directly characterizing the degree of micro-irregularity of the surface. Topographic relief is defined as the difference between the maximum and minimum elevations within a circular neighborhood with a radius of 500 meters, used to describe the range of topographic elevation changes at a mesoscale. All calculations use double-precision floating-point operations, and edge pixels are mirror-filled during the calculation process to ensure the integrity of the entire area's data and the absence of null values. The calculation results are generated in conjunction with the original DEM. Six raster layers with identical spatial resolution and range constitute a traditional topographic factor dataset. The calculated traditional topographic factors are then fused with the comprehensive isobath morphological entropy index generated through multi-scale fusion at the feature level. Before fusion, all feature data must be unified to the same spatial reference coordinate system and grid system, with a standard grid size of 50 meters. For data layers with inconsistent resolutions, bilinear interpolation is used for resampling. The comprehensive isobath morphological entropy index and traditional factors are spatially aligned pixel-by-pixel. The fusion process generates a comprehensive feature set containing multi-source descriptive information, consisting of more than ten feature dimensions, ensuring comprehensive coverage of geomorphological description information in terms of geometric morphology, statistical distribution, topological structure, and frequency domain features. The fused comprehensive feature set undergoes standardization preprocessing to eliminate differences in units and numerical ranges between different features. Z-score standardization is used, with the entire target sea area as the statistical range, and the global mean of each feature dimension is calculated independently. ) and standard deviation ( ), and perform the transformation: All standardized feature data are organized by spatial pixels. Each valid pixel corresponds to a vector containing all feature values. All pixel vectors together constitute a high-dimensional landform feature matrix.

[0099] Step S5 specifically includes the following steps: performing principal component analysis on the high-dimensional geomorphic feature matrix, extracting feature principal components and calculating the eigenvalues ​​and contribution rates of each principal component. The extracted principal components effectively reveal the most important variation directions in the original dataset. The variance contribution rate of each principal component is calculated, a cumulative contribution rate threshold is set, and the number of principal components to be retained is selected based on whether the cumulative contribution rate exceeds the preset cumulative contribution rate threshold. The number of principal components is adaptively determined according to the threshold to ensure that the data still retains most of the effective information after dimensionality reduction, avoiding information loss. Principal components exceeding the cumulative contribution rate threshold are retained. The original features are projected onto the feature subspace composed of the selected principal components to achieve data dimensionality reduction and redundancy removal, thereby obtaining the dimensionality-reduced feature dataset. The projection operation maps high-dimensional data to a low-dimensional space, eliminates redundant features, and significantly improves the computational efficiency and stability of subsequent clustering.

[0100] The specific work involves performing principal component analysis (PCA) on the constructed high-dimensional geomorphic feature matrix. First, each feature dimension of the matrix is ​​centered. Then, its covariance matrix is ​​calculated. By performing eigenvalue decomposition on the covariance matrix, a descending sequence of eigenvalues ​​and corresponding eigenvectors are obtained. Each eigenvector defines a new orthogonal coordinate axis, i.e., a principal component. The variance contribution rate of each principal component is calculated, i.e., the proportion of its eigenvalue to the sum of all eigenvalues. This contribution rate directly quantifies the explanatory power of the principal component for the total variance of the original data. To balance computational efficiency and information preservation, the maximum number of principal components analyzed is set to no more than 80% of the original feature dimensions or fixed at 30 components to initially control the computational scale. Based on the calculated variance contribution rates of each principal component, their cumulative contribution rate is further calculated. According to the requirements of deep-sea geomorphic classification tasks for feature representativeness and model complexity, the preset cumulative contribution rate threshold is between 85% and 95%. In practice, the threshold is fine-tuned based on data characteristics and classification accuracy requirements. It automatically accumulates the contribution rate starting from the first principal component. When the accumulated value first reaches or exceeds the preset threshold, the accumulation stops, and the principal components included up to this point are determined as the retention set. This achieves data-driven adaptive determination of the number of principal components, ensuring that the feature subspace after dimensionality reduction can cover most of the information with geomorphological significance in the original data. Each sample vector in the original high-dimensional geomorphological feature matrix is ​​projected onto the feature subspace spanned by the selected retention principal components. This projection is achieved by linearly transforming the centered original feature vector with the feature vector matrix corresponding to the retention principal components. After projection, each sample is represented by its coordinates on each retention principal component (i.e., principal component score), thereby transforming the data from the original high-dimensional space to a new space with significantly reduced dimensionality and uncorrelated dimensions. Finally, the dimensionality-reduced feature dataset is output, which will serve as the direct input for unsupervised clustering analysis.

[0101] Step S6 specifically includes the following steps: Based on the dimensionality-reduced feature dataset, an unsupervised clustering analysis is performed on the dimensionality-reduced feature data using a density-based spatial clustering algorithm. This enables the automatic discovery of the implicit natural grouping structure in the data when the landform type is unknown. By optimizing the clustering parameters covering the clustering radius and the minimum number of samples, the spatial clustering algorithm adaptively determines the number of natural groups for landform categories, i.e., the optimal number of categories. This avoids the problem of oversegmentation or undersegmentation that may be caused by manually presetting the number of categories. Furthermore, spatial neighborhood constraints are introduced during the clustering process to ensure that the same landform category has spatial continuity, avoid fragmented classification patches, and improve the rationality and interpretability of the classification results in geographic space.

[0102] The specific work involves: Based on the dimensionality-reduced feature dataset, implementing unsupervised clustering using a density-based spatial clustering algorithm. In practice, the cluster radius is determined by analyzing the k-distance plot between samples (where k is the minimum sample size minus 1) and observing the inflection point of the curve. This radius reflects the maximum allowable distance between similar samples in the feature space. The minimum sample size is preset based on data density and noise level, set to 2 to 4 times the feature dimension. To address the spatial continuity requirements of deep-sea geomorphological data, spatial neighborhood relationships are treated as a hard constraint: clustering only occurs when two samples satisfy the density reachability condition in the feature space, and... Only when the Euclidean distance in the original geospace does not exceed the preset geographical constraint radius can they be classified into the same class. This dual constraint mechanism effectively prevents spatially fragmented clustering results caused by pure feature similarity. A two-stage strategy is employed to adaptively determine the optimal number of classifications. In the first stage, preliminary density clustering is performed in the feature space to identify core points, boundary points, and noise points. To optimize parameters, a silhouette coefficient is introduced as an internal evaluation index. The clustering radius is increased from 0.1 to 0.5 in 0.05 steps, and the minimum sample size is increased from 10 to 30 in 5 steps. The algorithm selects the parameter combination that maximizes the contour coefficient as the final running parameters. No preset number of categories is required; clusters naturally form based on the density distribution of the data itself. For sample points marked as noise, they are not directly discarded but undergo a second stage of processing: they are reassigned to the category of the nearest core point, provided the aforementioned spatial distance constraints are met. This suppresses random noise while preserving effective geomorphic information to the greatest extent possible. After initial clustering, spatial post-processing further enhances the spatial consistency of the results. For any isolated small patches with an area smaller than a preset threshold, a forced merging operation is performed, merging them into the spatially adjacent and most similar main categories. Simultaneously, morphological closing operations (using 3×3 rectangular structural elements) are used to process the binary images of the spatial distribution of each category to fill in small holes caused by noise within similar regions and smooth irregular boundaries. Finally, the algorithm outputs the geomorphic category label for each pixel, the coordinate vector of the cluster center in the reduced-dimensional feature space, and the spatial statistical information of each category, achieving the recognition of natural geomorphic grouping. The algorithm adaptively determines the number of categories while ensuring the spatial continuity and morphological rationality of the category regions.

[0103] Step S7 specifically includes the following steps: extracting the feature vectors of each cluster center, combining them with the prior geomorphological knowledge base for semantic matching and type labeling, effectively overcoming the limitation of unsupervised clustering results lacking actual geomorphological significance, realizing the geological interpretability transformation of data-driven classification results, transforming numerical clustering results into classification maps with geomorphological type labels, completing the transformation from data clustering to semantic interpretation, mapping abstract numerical groupings to specific geomorphological type maps, providing intuitive and understandable result base maps for subsequent geological analysis and applications, and performing morphological opening and closing operations on the preliminary classification results to identify spatial regions that need post-processing, so as to smooth inter-class boundaries and suppress small noise patches, significantly improving the spatial continuity and visual quality of classification map patches, and eliminating the interference of discrete noise points and small holes on the mapping effect;

[0104] The specific work involves: matching and labeling the coordinate vectors of each category center in the reduced-dimensional feature space obtained from unsupervised clustering with an established deep-sea geomorphological prior knowledge base. This knowledge base is stored in a structured form and includes the theoretical or empirical feature vector range, statistical distribution, and spatial configuration patterns of typical geomorphological units (seamounts, hydrothermal vents, fault zones, sedimentary basins, canyons, etc.) in their corresponding feature spaces. The matching process uses the least Euclidean distance criterion, measuring the similarity between the feature vector of each cluster center and the feature templates of each predefined geomorphological type in the knowledge base. If the feature matching degree with a certain geomorphological type exceeds a preset threshold (set to similarity ≥ 0.75), the cluster is initially labeled as that geomorphological type. For categories with a matching degree below the threshold, they are marked as unidentified types, and their abnormal feature combinations are recorded for subsequent expert evaluation or knowledge base iteration updates, achieving a transition from unlabeled to unidentified types. The process involves transforming numerical clustering into a category system with preliminary geomorphic semantics. Semantically matched category labels are assigned to each pixel to generate a preliminary geomorphic classification map. Subsequently, to improve image quality and spatial consistency, morphological post-processing is employed. Morphological closing operations (using 3×3 or 5×5 rectangular structural elements) are sequentially performed on the binary images corresponding to each category to fill small holes caused by sporadic noise within the category, connect narrow breaks, and smooth irregular boundaries. Then, morphological opening operations (using the same structural elements) are performed to eliminate isolated small patches with areas smaller than a preset threshold and separate unnaturally adhered regions caused by over-connectivity. During processing, the size of the structural elements and the order of operations need to be adjusted according to the typical scale of the geomorphic unit to ensure that noise is suppressed without destroying the morphological details of the real geomorphic boundaries. After processing, the spatial distribution of each category is more continuous and morphologically reasonable.

[0105] Step S7 also includes: based on the morphologically processed classification map, performing region merging and boundary optimization to improve the internal homogeneity and boundary consistency of each category of region; spatially overlaying and verifying the classification results with the original terrain data to ensure that the classification results match the actual terrain features; effectively eliminating salt-and-pepper noise and broken patches in the classification map; improving the overall integrity and visual coherence of the map; and, based on spatial overlay and verification, outputting the final automated geomorphological classification results with clear geomorphological semantics, spatial continuity and morphological optimization, providing a standardized geomorphological classification product and supporting documents that are structurally complete, semantically clear and can be directly used in business.

[0106] The specific work involved: Based on morphological processing, to further improve the spatial quality of the classification map, region merging and boundary optimization operations were implemented. Using a connected component analysis algorithm, isolated small patches within each category with an area smaller than a preset threshold (set to 50 pixels, corresponding to an actual area of ​​0.125 square kilometers) were forcibly merged and assigned to the main category that was spatially adjacent and had the most similar spectral and morphological characteristics, thus eliminating classification fragmentation. A sliding window-based local boundary optimization algorithm was then used, with a window size of 7×7 pixels. Category boundary pixels were iteratively reclassified, adjusting their assignment based on their Mahalanobis distance from the centers of neighboring categories in the feature space, until the category change rate of the boundary pixels was less than 1% or the maximum number of iterations (set to 10) was reached, making the inter-class boundaries smoother and more continuous, conforming to the transition characteristics of natural land features. After spatial optimization, the final classification results were spatially overlaid and their consistency verified with the original high-resolution digital elevation model and multi-source marine geological data. The overlay analysis was performed in the GIS platform. The process involves calculating the spatial distance deviation between the classification boundary and topographic abrupt change zones and known geological structural lines to quantify classification accuracy. The average positional deviation is required to be controlled within 2 pixels (i.e., 100 meters). Simultaneously, the classification is validated using a confusion matrix and field-measured samples or expert-interpreted sample areas to ensure that the identification accuracy of each landform category is not less than 85%, and that the classification results can reasonably reflect the actual topographic structure and spatial distribution patterns of seamounts, fault zones, sedimentary basins, etc. After validation, the final automated landform classification results are output. The results are based on GeoTIFF format raster data, with spatial resolution consistent with the input DEM, and include a vector boundary layer with complete attributes. Each landform patch records its type code, area, average elevation, main morphological entropy characteristic values, and spatial statistical information. At the same time, a comprehensive technical report is generated, covering processing methods, parameter settings, accuracy evaluation indicators, and uncertainty explanations, forming a standardized and traceable landform classification product that can be directly used for marine geological mapping, resource exploration, and environmental assessment.

[0107] Example 3, as Figures 1 to 4As shown, based on Examples 1-2, this invention provides a technical solution: taking a 100-meter resolution digital elevation model (DEM) data of a typical sea area as an example, the proposed automatic landform classification method is applied to conduct an experiment. The specific implementation process is as follows:

[0108] 1. Experimental area and data preparation:

[0109] A typical sea area containing various landform types such as seamounts, sea hills, and sea valleys was selected as the experimental area. 100m×100m grid DEM data measured in this area were used as input, and preprocessing such as outlier removal and smoothing filtering was performed in advance to ensure data quality.

[0110] 2. Feature extraction and matrix construction:

[0111] First, multi-scale convolutional kernels are used to extract features from DEM data. The kernel sizes are set to 3×3, 7×7, and 15×15 to generate CME feature maps that reflect the local to meso-level topographic structure. At the same time, traditional topographic factors such as slope, profile curvature, planar curvature, topographic roughness, and undulation are calculated for the region. After normalizing all features, they are integrated into a high-dimensional feature matrix according to pixel location. The rows of this matrix correspond to each pixel, and the columns correspond to topographic features.

[0112] 3. Unsupervised clustering analysis:

[0113] A density-based spatial clustering algorithm with noise was used to cluster the dimensionality-reduced features. After parameter tuning, with a neighborhood radius of 0.35 and a minimum sample size of 15, the algorithm automatically identified 5 spatial clusters and successfully separated noise points. The clustering process considered the spatial adjacency between pixels, ensuring the spatial continuity of each landform unit. The classification results are as follows: Figure 4 As shown, Type I is a foothill, Type II is a depression, Type III is a ridge, Type IV is a shoulder, and Type V is a slope.

[0114] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0115] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A fully automated method for identifying and classifying deep-sea landforms based on isobath morphological entropy, characterized in that, Includes the following steps: Step S1: Extract contour lines from the digital elevation model of the target area at set intervals, construct a topological relationship map of contour lines, and identify closed loops and bifurcation points; Step S2: Based on the isobath topology map, calculate the multi-dimensional isobath morphological entropy sub-factor, including the isobath topological complexity index, isobath geometric entropy, isobath density gradient anomaly, and isobath curvature spectrum characteristics. This specifically includes the following steps: Based on the topological relationship map of isobaths, and combined with the number of closed loops and the density of bifurcation points per unit area, the topological complexity index of isobaths is calculated to reflect the topological complexity of the terrain structure. Each contour line is resampled to generate a sequence of orientation angles, and its Shannon entropy is calculated to obtain the geometric entropy of the contour line, which characterizes the randomness and irregularity of the orientation distribution of the contour line. The gradient magnitude is calculated based on the isobath density field, and the isobath density gradient anomaly is obtained through normalization to indicate the local variation in the spatial aggregation of isobaths. Step S2 further includes: Fourier transform is performed on the high-order sequence of a single contour line to calculate its power spectrum, and the low-frequency energy ratio and high-frequency energy ratio are extracted. Based on the distribution of the power spectrum in the logarithmic coordinate system, the spectral slope is fitted and calculated to characterize the scale characteristics of the contour line undulation. The low-frequency energy ratio, high-frequency energy ratio and spectral slope are combined to form the curvature spectrum feature of the isobath, which is used to describe the multi-scale frequency domain structure of the isobath morphology. The calculated topological complexity index of isobaths, geometric entropy of isobaths, density gradient anomaly of isobaths and curvature spectrum characteristics of isobaths are integrated to form a multidimensional isobath morphological entropy sub-factor sequence. Step S3: Calculate the isobath morphology entropy sub-factors at multiple analysis window scales, and obtain the comprehensive isobath morphology entropy index through weighted fusion; Step S4: Integrate the morphological entropy of the comprehensive isobaths with traditional topographic factors, and organize them into a high-dimensional geomorphic feature matrix after standardization. Step S5: Use principal component analysis to reduce the dimensionality of the high-dimensional landform feature matrix and retain the main components whose cumulative contribution rate exceeds the threshold. Step S6: Based on the dimensionality-reduced high-dimensional landform features, unsupervised clustering is performed using a density-based clustering algorithm to adaptively determine the number of landform categories; Step S7: Based on the feature vectors corresponding to each cluster center, semantic annotation is performed in conjunction with prior geomorphological knowledge, and image processing methods are used to optimize the spatial morphology of the classification map to obtain automated geomorphological classification results.

2. The fully automatic identification and classification method for deep-sea landforms based on isobath morphological entropy according to claim 1, characterized in that: Step S1 specifically includes the following steps: Obtain the digital elevation model of the target area, extract contour lines from the digital elevation model at preset depth intervals, and record the geometric coordinates and depth attributes of each contour line; Based on the extracted geometric coordinates and depth attributes of the isobaths, a topological relationship map of the isobaths is constructed, closed isobath loops and bifurcation nodes are identified, and their spatial locations and connectivity are recorded. Calculate the geometric properties of each contour line, including length, average curvature, and discretized direction angle sequence.

3. The fully automatic identification and classification method for deep-sea landforms based on isobath morphological entropy as described in claim 1, characterized in that: Step S3 specifically includes the following steps: Multiple analysis windows of different scales are set up, and the multi-dimensional isobath morphological entropy sub-factor is calculated in each window; Weight coefficients are assigned to the isobath morphology entropy sub-factors at each scale to reflect the differences in the contribution of different scales to landform identification. The comprehensive isobath morphology entropy index is generated by weighted fusion of the isobath morphology entropy sub-factors at each scale.

4. The fully automatic identification and classification method for deep-sea landforms based on isobath morphological entropy as described in claim 1, characterized in that: Step S4 specifically includes the following steps: Extract traditional topographic factors of the target area, including slope, aspect, profile curvature, planar curvature, roughness, and topographic relief. The comprehensive contour line morphological entropy index is fused with traditional terrain factors to form a feature set containing multi-source terrain description information. The fused features are standardized and organized into a high-dimensional landform feature matrix by pixel, which serves as the input data for landform classification.

5. The fully automatic identification and classification method for deep-sea landforms based on isobath morphological entropy according to claim 1, characterized in that: Step S5 specifically includes the following steps: Principal component analysis is performed on the high-dimensional geomorphic feature matrix to extract characteristic principal components and calculate the eigenvalues ​​and contribution rates of each principal component. Calculate the variance contribution rate of each principal component, set a cumulative contribution rate threshold, and select the number of principal components to retain based on whether the cumulative contribution rate exceeds the preset cumulative contribution rate threshold. Principal components exceeding the cumulative contribution rate threshold are retained, and the original features are projected onto the feature subspace formed by the selected principal components to obtain the dimensionality-reduced feature dataset.

6. The fully automatic identification and classification method for deep-sea landforms based on isobath morphological entropy according to claim 1, characterized in that: Step S6 specifically includes the following steps: Based on the dimension-reduced feature dataset, a density-based spatial clustering algorithm is used to perform unsupervised clustering analysis on the dimension-reduced feature data. By optimizing the clustering parameters that cover the clustering radius and the minimum number of samples, the spatial clustering algorithm adaptively determines the number of natural groups for landform categories, i.e. the optimal number of categories, and introduces spatial neighborhood constraints during the clustering process to ensure that the same landform category has spatial continuity.

7. The fully automatic identification and classification method for deep-sea landforms based on isobath morphological entropy according to claim 6, characterized in that: Step S7 specifically includes the following steps: Extract the feature vectors of each cluster center and combine them with a prior geomorphological knowledge base for semantic matching and type labeling; The numerical clustering results are transformed into classification maps with landform type labels, completing the transformation from data clustering to semantic interpretation. Morphological opening and closing operations are performed on the preliminary classification results to identify spatial regions that need post-processing, so as to smooth the inter-class boundaries and suppress small noise patches.

8. The fully automatic identification and classification method for deep-sea landforms based on isobath morphological entropy according to claim 7, characterized in that: Step S7 further includes: Based on the morphologically processed classification map, region merging and boundary optimization are performed to improve the internal homogeneity and boundary consistency of each category region. The classification results are spatially overlaid and verified with the original terrain data to ensure that the classification results match the actual terrain features. Based on spatial overlay and verification, the final output is an automated geomorphological classification result with clear geomorphological semantics, spatial continuity and morphological optimization.

Images