Terrain feature recognition method, device and equipment based on remote sensing technology and medium

Abstract

The invention relates to a topographic feature recognition method and device based on a remote sensing technology, equipment and a medium. The method comprises the following steps: acquiring remote sensing image data, and preprocessing the remote sensing image data to obtain surface reflectance data; based on the surface reflectance data, carrying out ground feature category preliminary classification on each pixel in the remote sensing image data to obtain initial ground feature classification data; according to the initial ground feature classification data and the digital elevation model data, carrying out gradient and slope direction analysis on the elevation information to obtain topographic feature data; based on the initial ground feature classification data and the topographic feature data, optimizing and classifying the boundary of each ground feature type to obtain fine ground feature classification data; and generating a terrain recognition result according to the fine terrain classification data and the terrain feature data. By adopting the method, the reliability and practicability of a terrain recognition result can be improved.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing technology, and in particular relates to methods, devices, equipment and media for terrain feature identification based on remote sensing technology. Background Technology

[0002] With the rapid development of remote sensing technology in the field of geospatial information extraction, remote sensing image data, with its wide coverage, high acquisition efficiency, and rich information, has become the core data source for terrain feature identification. Traditional terrain feature identification methods typically use remote sensing imagery as a basis, combined with digital elevation model data, to classify land cover categories through simple spectral analysis, and to identify geomorphic units and land cover structures based on basic terrain parameters.

[0003] In practical applications, traditional methods first perform simple processing on remote sensing images to obtain basic spectral information, then classify land features in each pixel through single feature analysis, subsequently calculate basic terrain parameters by combining digital elevation model data, and finally integrate relevant data to obtain terrain identification results.

[0004] However, existing methods have significant shortcomings: First, the processing of remote sensing images lacks a systematic approach, resulting in limited accuracy of the acquired surface feature data and affecting the accuracy of land cover classification. Second, land cover classification relies solely on single-dimensional features, failing to fully integrate topographic information, leading to coarse initial classification results. Third, topographic analysis is limited to basic parameter calculations, making it difficult to comprehensively characterize the distribution characteristics of surface undulation and orientation. Fourth, the initial land cover classification boundaries are not effectively optimized, and deep integration of land cover classification data and topographic feature data is not achieved, ultimately resulting in blurred geomorphic unit divisions and inaccurate representation of land cover structure in the topographic identification results, failing to meet the practical application requirements of high-precision topographic feature identification. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, equipment, and medium for terrain feature recognition based on remote sensing technology that can improve the reliability and practicality of terrain recognition results, addressing the aforementioned technical problems.

[0006] Firstly, this application provides a terrain feature recognition method based on remote sensing technology, including:

[0007] Remote sensing image data is acquired and preprocessed to obtain surface reflectance data; surface reflectance data is used to characterize the reflectance characteristics of ground objects in different bands.

[0008] Based on surface reflectance data, preliminary classification of land cover categories is performed on each pixel in remote sensing image data to obtain initial land cover classification data.

[0009] Based on the initial land feature classification data and digital elevation model data, slope and aspect analysis is performed on the elevation information to obtain topographic feature data; the topographic feature data is used to characterize the distribution of surface undulation and orientation within the region.

[0010] Based on the initial land cover classification data and terrain feature data, the boundaries of each land cover type are optimized and classified to obtain refined land cover classification data.

[0011] Based on detailed land cover classification data and terrain feature data, terrain identification results are generated; the terrain identification results include geomorphic unit division and land cover structure.

[0012] In one embodiment, the remote sensing image data is preprocessed to obtain surface reflectance data, including:

[0013] Radiometric calibration processing is performed on remote sensing image data to convert the raw digital quantization values ​​into radiance values, thus obtaining radiometric calibration data;

[0014] Based on radiometric calibration data, atmospheric scattering and absorption effects are eliminated by using a correction model to obtain atmospheric correction data;

[0015] Geometric correction is performed on the atmospheric correction data to obtain geometrically corrected data.

[0016] Based on geometrically corrected data, the surface reflectance is calculated using the following formula to obtain the surface reflectance data:

[0017]

[0018] in, For surface reflectance, For band The radiance value, This is the Earth-Sun distance correction factor. For band Solar irradiance at the top of the atmosphere, This is the solar zenith angle.

[0019] In one embodiment, based on surface reflectance data, preliminary land cover classification is performed on each pixel in the remote sensing image data to obtain initial land cover classification data, including:

[0020] Vegetation cover was calculated using the normalized vegetation index based on surface reflectance data, and vegetation characteristic data were obtained.

[0021] Based on surface reflectance data, water body information is extracted through normalized differential water body index to obtain water body characteristic data;

[0022] Texture feature data is obtained by calculating texture features based on surface reflectance data and gray-level co-occurrence matrix.

[0023] Based on vegetation feature data, water body feature data, and texture feature data, a random forest classifier is used to perform preliminary classification of land cover categories, resulting in initial land cover classification data.

[0024] In one embodiment, based on initial land cover classification data and digital elevation model data, slope and aspect analysis is performed on the elevation information to obtain terrain feature data, including:

[0025] Based on digital elevation model data, the variable window search method considering terrain undulation is used to calculate the elevation change rate of pixels in the east-west and north-south directions in the elevation matrix, and obtain elevation gradient component data.

[0026] Based on the elevation gradient component data, slope data and aspect data are obtained; slope data is used to describe the degree of inclination of the ground surface; aspect data is used to describe the azimuth angle of the slope.

[0027] Based on the elevation gradient component data, the profile curvature and plane curvature are obtained using the following formulas:

[0028]

[0029]

[0030] in, The curvature of the profile is used to characterize the surface's erosion and deposition capacity. Planar curvature is used to characterize the surface runoff capacity. Indicates elevation In the east-west direction rate of change on Indicates elevation In the north-south direction rate of change on East-west direction Elevation curvature, For mixed directional curvature, North-South Direction Elevation curvature;

[0031] Curvature characteristic data are obtained based on the profile curvature and the planar curvature;

[0032] Slope data, aspect data, and curvature feature data are fused at multiple scales to generate terrain feature data.

[0033] In one embodiment, based on initial land cover classification data and terrain feature data, the boundaries of each land cover type are optimized and classified to obtain refined land cover classification data, including:

[0034] Based on the initial land cover classification data and topographic feature data, a spectral-topographic correlation map is obtained;

[0035] Based on the spectral-topography correlation map, the initial land cover category labels are globally optimized to obtain optimized category label data;

[0036] Based on the optimized category label data, a conditional random field model is used to integrate spatial context information and perform semantic consistency smoothing on the boundaries of ground features to obtain refined ground feature classification data.

[0037] In one embodiment, a terrain identification result is generated based on detailed land cover classification data and terrain feature data, including:

[0038] Based on detailed land cover classification data, the area proportion of each land cover type is calculated using spatial statistical methods to obtain land cover structure data;

[0039] Based on topographic feature data, geomorphic units are divided through cluster analysis to obtain geomorphic unit division data;

[0040] Based on land cover structure data and geomorphic unit division data, terrain identification results are generated through association rule mining.

[0041] In one embodiment, based on terrain feature data, geomorphic units are divided through cluster analysis to obtain geomorphic unit division data, including:

[0042] Cluster analysis was performed on the terrain feature data to obtain the initial clustering results;

[0043] The initial clustering results are evaluated to assess the clustering effectiveness, and clustering evaluation data is obtained.

[0044] Based on the cluster evaluation data, the cluster centers are iteratively optimized and adjusted to obtain the geomorphic unit division data.

[0045] Secondly, this application also provides a terrain feature recognition device based on remote sensing technology, comprising:

[0046] The data acquisition and preprocessing module is used to acquire remote sensing image data and preprocess the remote sensing image data to obtain surface reflectance data; the surface reflectance data is used to characterize the reflectance characteristics of ground objects in different bands.

[0047] The initial land cover classification module is used to perform preliminary land cover classification on each pixel in the remote sensing image data based on the surface reflectance data, and obtain the initial land cover classification data.

[0048] The terrain feature module is used to analyze the slope and aspect of elevation information based on the initial land feature classification data and digital elevation model data to obtain terrain feature data; the terrain feature data is used to characterize the distribution of surface undulation and orientation within the region;

[0049] The land cover classification optimization module is used to optimize and classify the boundaries of various land cover types based on the initial land cover classification data and terrain feature data, so as to obtain refined land cover classification data.

[0050] The terrain recognition module is used to generate terrain recognition results based on detailed land feature classification data and terrain feature data; the terrain recognition results include geomorphic unit division and land feature cover structure.

[0051] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0052] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0053] The aforementioned terrain feature recognition method, device, equipment, and medium based on remote sensing technology integrate remote sensing image data and digital elevation model data. Through standardized preprocessing, high-quality surface reflectance data is obtained, laying a reliable foundation for subsequent analysis. Multi-dimensional spectral and textural features are combined to achieve preliminary and accurate classification of land cover categories. Then, through variable window search and multi-scale fusion, terrain slope, aspect, and curvature features are comprehensively mined. A spectral-terrain correlation map is constructed, and land cover boundaries are optimized using a conditional random field model. Finally, the land cover structure and geomorphic unit division are integrated to generate recognition results, achieving deep synergy between land cover classification and terrain features. This effectively solves the problems of insufficient data accuracy, coarse classification, one-sided terrain representation, and blurred boundaries in traditional methods, improving the comprehensiveness, accuracy, and reliability of terrain recognition. The obtained results clearly reflect surface undulation, land cover distribution, and the correlation between the two, meeting the practical application needs of high-precision geospatial information extraction. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1This is a flowchart illustrating a terrain feature recognition method based on remote sensing technology in one embodiment;

[0056] Figure 2 This is a schematic diagram of a terrain feature recognition device based on remote sensing technology in one embodiment. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0058] In one embodiment, such as Figure 1 As shown, a terrain feature recognition method based on remote sensing technology is provided. This embodiment illustrates the application of this method to a terrain recognition terminal (hereinafter referred to as the terminal). It can be understood that this method can also be applied to a server, and can also be applied to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0059] S1. Acquire remote sensing image data and preprocess the remote sensing image data to obtain surface reflectance data.

[0060] For example, the terrain recognition terminal first acquires remote sensing image data, which can come from satellite remote sensing systems (such as Landsat and Sentinel series satellites) or airborne remote sensing equipment. This data contains raw response information of ground features in multiple spectral bands, including visible light, near-infrared, and shortwave infrared. The data format is typically TIFF or ENVI standard format. To eliminate the influence of various interference factors during data acquisition on subsequent analysis, the terminal preprocesses the remote sensing image data: first, radiometric calibration eliminates errors caused by sensor response differences and equipment aging; then, atmospheric correction removes scattering and absorption by gas molecules and aerosols in the atmosphere; subsequently, geometric correction corrects spatial position deviations caused by sensor attitude shifts, Earth curvature, and terrain undulations; finally, surface reflectance data is calculated using a standardized formula. Surface reflectance data objectively characterizes the reflectivity of ground features to electromagnetic radiation in different bands. Its value is directly related to the physical properties of ground features (such as water content, chlorophyll content, roughness, and material composition), and is unaffected by lighting conditions or atmospheric conditions, providing a standardized and highly reliable spectral feature basis for subsequent ground feature classification.

[0061] S2. Based on surface reflectance data, preliminary classification of land cover categories is performed on each pixel in the remote sensing image data to obtain initial land cover classification data.

[0062] For example, the terrain recognition terminal performs preliminary classification of land cover categories for each pixel in a remote sensing image based on surface reflectance data. Different types of land cover exhibit significantly specific spectral curves in different spectral bands. For instance, vegetation has high reflectance in the near-infrared band due to the influence of leaf cell structure, and low reflectance in the red band due to chlorophyll absorption; water bodies have extremely low reflectance in the infrared band due to strong absorption characteristics, and relatively high reflectance in the green band; bare soil and buildings each have their own unique spectral curve shapes, and their reflectance varies significantly across different bands. The terminal extracts these specific spectral features and combines them with the spatial distribution patterns of land cover in the image—such as buildings often clustered and water bodies often distributed in continuous patches—using specialized classification algorithms to determine the land cover category of each pixel, obtaining preliminary land cover classification results, i.e., initial land cover classification data. This data includes vegetation, water bodies, bare soil, building cover types, etc.

[0063] S3, based on the initial land feature classification data and digital elevation model data, performs slope and aspect analysis on the elevation information to obtain topographic feature data. This topographic feature data is used to characterize the distribution of surface undulation and orientation within the region.

[0064] For example, the terrain recognition terminal performs slope and aspect analysis on elevation information based on initial land feature classification data and Digital Elevation Model (DEM) data. DEM data is a digital terrain model obtained through topographic surveying, remote sensing stereo mapping, and other technologies. It stores the elevation value of each pixel within a region in raster form, intuitively reflecting the undulations of the land surface. Initial land feature classification data provides the distribution range and category information of different land feature types within the region. Combining the two makes terrain analysis more targeted and reasonable. For example, water features are usually associated with low slopes and low-lying terrain, while buildings are mostly distributed in gently sloping terrain areas. The terminal uses professional terrain analysis algorithms to calculate key terrain parameters such as slope, aspect, and curvature. These parameters comprehensively characterize the undulations, slope variation patterns, and orientation distribution characteristics of the land surface within the region, obtaining terrain feature data. Specifically, slope reflects the steepness of the land surface, aspect reflects the orientation of the slope, and curvature reflects the degree of curvature of the land surface.

[0065] S4, based on the initial land cover classification data and terrain feature data, optimizes and classifies the boundaries of each land cover type to obtain refined land cover classification data.

[0066] For example, the terrain recognition terminal optimizes and classifies the boundaries of various feature types based on initial feature classification data and terrain feature data. The initial feature classification data, based solely on spectral features, may suffer from issues such as blurred boundaries and isolated misclassified points. Terrain feature data, however, provides crucial auxiliary constraints, as the distribution of different feature types is often associated with specific terrain conditions. For instance, water bodies are mostly distributed in low-slope, low-lying areas, buildings in gently sloping areas, and vegetation is found in various terrains but is more concentrated in areas with moderate slopes. The terminal deeply fuses these two types of data, using terrain features to verify and adjust the initial classification results. This makes the location and shape of feature boundaries more consistent with actual geographical distribution patterns. Furthermore, it further classifies and integrates feature categories with similar features and continuous distribution, resulting in more precise, clearer-boundary, and more accurate feature classification data. This provides a reliable foundation of feature information for the final terrain recognition results. Blurred boundaries include transitional areas between vegetation and bare soil, and boundaries between water bodies and wetlands. Isolated misclassified points include individual building pixels being misclassified as bare soil due to spectral similarity.

[0067] S5 generates terrain recognition results based on detailed land feature classification data and terrain feature data.

[0068] For example, the terrain recognition terminal generates terrain recognition results based on detailed land cover classification data and terrain feature data. Detailed land cover classification data provides precise distribution range, boundary locations, and category information for different land cover types (vegetation, water bodies, bare soil, building cover) within a region, clearly reflecting the spatial distribution pattern of land cover. Terrain feature data comprehensively reflects the undulations, slope and aspect distribution, and curvature characteristics of the land surface, serving as the core basis for dividing geomorphic units. The deep integration of these two data points enables correlation analysis between land cover and terrain morphology, revealing the distribution patterns of land cover under different terrain conditions. The terminal uses professional spatial analysis algorithms to first statistically analyze the spatial proportion of each land cover type to obtain the land cover structure, then clusters similar terrain feature regions to obtain geomorphic unit divisions, and finally integrates the two types of information and mines their inherent correlations to form terrain recognition results that include geomorphic unit divisions and land cover structure, comprehensively representing the relationship between regional terrain features and land cover distribution.

[0069] The aforementioned terrain feature recognition method based on remote sensing technology improves the comprehensiveness and accuracy of terrain recognition by integrating remote sensing imagery and digital elevation model data. First, high-quality surface reflectance data is obtained through standardized preprocessing, laying a reliable foundation for subsequent analysis. Then, multi-dimensional spectral and texture features are combined to complete preliminary land cover classification. Combined with comprehensive terrain analysis of slope, aspect, and curvature, the information related to surface space and terrain is fully explored. Through graph model construction, global optimization, and conditional random field smoothing, land cover boundaries are effectively optimized, improving classification accuracy. Finally, the land cover structure and geomorphic unit division are integrated to generate comprehensive terrain recognition results. This avoids the problems of unsystematic data processing, coarse classification, and one-sided terrain representation in traditional methods, improving the reliability and practicality of terrain recognition and better adapting to the application requirements of high-precision geospatial information extraction.

[0070] In an optional embodiment, the remote sensing image data is preprocessed to obtain surface reflectance data, including the following steps:

[0071] S11. Radiometric calibration processing is performed on the remote sensing image data, converting the original digital quantization values ​​into radiance values ​​to obtain radiometric calibration data.

[0072] For example, the terrain recognition terminal performs radiometric calibration on remote sensing image data. The raw remote sensing image data is stored as a Digital Number (DN) value, which is the output of the sensor after converting the received radiant energy into a digital signal. This value only reflects the sensor's relative response intensity to radiant energy and has no physical meaning; furthermore, DN values ​​from different bands and imaging times are not comparable. The radiometric calibration process uses the calibration coefficients preset at the sensor's factory, including gain and offset coefficients, and employs a linear or nonlinear conversion model to convert the DN value into a radiance value with physical dimensions (units are watts / (square meter·steradian·micrometer)). This process effectively eliminates the sensor's response deviations at different times and in different bands, ensuring that the converted radiance value accurately reflects the radiative emission level of ground objects in a specific band. This provides a unified radiometric benchmark for subsequent atmospheric correction and other processing, ultimately resulting in standardized radiometric calibration data.

[0073] S12, based on radiometric calibration data, eliminates the effects of atmospheric scattering and absorption by using a correction model to obtain atmospheric correction data.

[0074] For example, the terrain recognition terminal performs atmospheric correction processing based on radiometric calibration data. As solar radiation travels from the top of the atmosphere to the surface of land features, it is scattered and absorbed by gas molecules and aerosols in the atmosphere. This causes the radiance values ​​received by the sensor to deviate from the true radiation characteristics of the land feature surface, directly affecting the accuracy of the spectral features. The terminal employs a mature atmospheric correction model. By inputting atmospheric parameters at the time of observation, such as aerosol optical thickness, water vapor content, atmospheric pressure, and temperature, which can be obtained from meteorological station data or metadata accompanying images, a radiative transfer model is constructed to invert the influence of the atmosphere on radiation. This influence is then precisely removed from the radiometric calibration data, thereby restoring the true radiation information of the land feature surface and obtaining atmospheric correction data that accurately reflects the inherent spectral characteristics of the land feature. Among them, gas molecules include oxygen, water vapor, carbon dioxide, etc., and aerosols include dust, smoke particles, etc. Atmospheric correction models can use MODTRAN model (MODerate spectral resolution atmospheric TRANsmission, atmospheric radiative transfer model), 6S model (SECOND SIMULATION OF THE SATELLITE SIGNAL IN THE SOLARSPECTRUM, radiative transfer model), FLAASH model (Fast line-of-sight atmospheric analysis of spectral hypercubes), etc.

[0075] S13, Perform geometric correction processing on the atmospheric correction data to obtain geometric correction data.

[0076] For example, the terrain recognition terminal performs geometric correction processing on atmospheric correction data. During the imaging process, remote sensing images are subject to geometric distortion due to a combination of factors, specifically manifested as a mismatch between the pixel positions on the image and their actual geographical locations. Common distortion types include translation, rotation, scaling, affine distortion, and nonlinear distortion, primarily caused by sensor platform attitude changes, Earth's curvature, terrain undulations, and Earth's rotation. The terminal determines ground control points—clearly identifiable feature points on the image with known actual geographical locations, such as road intersections, bridge endpoints, and mountain peaks—through manual selection or automatic matching. Subsequently, methods such as polynomial fitting, affine transformation, or rational function models are used to establish a mathematical mapping relationship between the image coordinate system and the target geographic coordinate system. Based on this mapping relationship, the position of each pixel is corrected point-by-point, ensuring that the corrected image accurately corresponds to the actual geographic spatial location. This ensures that images from different sources and at different times can be spatially overlaid for analysis, resulting in geometrically corrected data with satisfactory accuracy.

[0077] S14. Based on geometric correction data, the surface reflectance is calculated using the following formula to obtain surface reflectance data:

[0078]

[0079] in, For surface reflectance, For band The radiance value, This is the Earth-Sun distance correction factor. For band Solar irradiance at the top of the atmosphere, This is the solar zenith angle.

[0080] In the above formula, Surface reflectance is the ratio of the radiation flux reflected by a ground object to the incident radiation flux. It is used to characterize the ability of a ground object to reflect electromagnetic radiation in a specific band. It is not affected by external factors such as light, atmosphere, and observation distance, and is a core spectral indicator for the classification and identification of ground objects. For band The radiance value of a ground object is the radiance value of a ground object under wave. Radiometric radiance after radiometric calibration, atmospheric correction, and geometric correction is expressed in watts per square meter (m²·steradian·micrometer), reflecting the radiative emission intensity of ground features in that wavelength band. This is a correction factor for the Earth-Sun distance. Because the Earth's orbit around the Sun is elliptical, the distance between the Earth and the Sun varies at different times, causing changes in solar irradiance. To correct for this distance discrepancy, its value can be calculated using an astronomical formula based on the imaging date. For band Solar irradiance at the top of the atmosphere refers to the wavelengths reaching the top of Earth's atmosphere. Solar radiation intensity, measured in watts per square meter (µm), can be obtained from solar spectral databases or remote sensing image metadata. The solar zenith angle is the angle between the sun's rays and the Earth's surface normal. It affects the angle of incidence of solar radiation and is used to correct the impact of the solar incidence angle on radiation reception. It can be calculated from the imaging time and geographical location. For example, the terrain recognition terminal, based on the geometric correction data terminal, uses the above formula to convert the radiance value, which has physical dimensions, into the dimensionless surface reflectance, obtaining surface reflectance data that can objectively characterize the spectral differences of different land features.

[0081] In an optional embodiment, based on surface reflectance data, preliminary land cover classification is performed on each pixel in the remote sensing image data to obtain initial land cover classification data, including the following steps:

[0082] S21. Based on surface reflectance data, vegetation cover is calculated using the normalized vegetation index to obtain vegetation characteristic data.

[0083] For example, the terrain recognition terminal calculates vegetation feature data based on surface reflectance data. The terminal uses the Normalized Difference Vegetation Index (NDVI) as the core calculation indicator, and its calculation formula is as follows: ,in, The surface reflectance is in the near-infrared band. NDVI is the surface reflectance in the red light band. The design principle of this index is based on the physiological characteristics of plant leaves: chlorophyll strongly absorbs solar radiation in the red light band, while the cellular structure inside the leaf strongly reflects radiation in the near-infrared band. This characteristic is not possessed by other land features (such as soil, buildings, and water bodies). NDVI values ​​typically range from -1 to 1. Negative values ​​often correspond to areas with water bodies or cloud shadows, while areas near 0 are mostly bare soil or rock. Positive values ​​represent vegetation-covered areas, and higher values ​​indicate higher vegetation coverage and better growth. The terminal constructs a vegetation index image by calculating the NDVI value of each pixel in the image, obtaining vegetation characteristic data that clearly reflects the distribution range, coverage intensity, and spatial differences of vegetation within the region.

[0084] S22, based on surface reflectance data, extracts water body information through normalized differential water body index to obtain water body characteristic data.

[0085] For example, the terrain recognition terminal extracts water body feature data based on surface reflectance data. The terminal uses the Normalized Difference Water Index (NDWI) as the core extraction indicator. Its core design principle utilizes the spectral differences between water bodies and other land features in specific wavelength bands: water bodies have relatively high reflectance in the green band, while their reflectance is extremely low in the near-infrared band due to strong absorption; conversely, non-water features such as vegetation and soil have high reflectance in the near-infrared band and relatively low reflectance in the green band. The commonly used calculation formula is... ,in, The surface reflectance is in the green band. This refers to the surface reflectance in the near-infrared band. After calculation using this formula, water areas will exhibit higher NDWI values, while non-water areas will exhibit lower NDWI values. By setting a reasonable threshold, the terminal can effectively distinguish between water areas and non-water areas, further eliminating small noise interference areas and extracting water feature data that accurately reflects the distribution range, morphological characteristics, and boundary locations of water bodies.

[0086] S23, based on surface reflectance data, texture features are calculated through gray-level co-occurrence matrix to obtain texture feature data.

[0087] For example, the terrain recognition terminal calculates texture feature data based on surface reflectance data. Texture features are the spatial distribution patterns and combinations of gray values ​​on the surface of land features, effectively reflecting information such as the physical roughness, structural arrangement, and spatial distribution density of land features. For instance, built-up areas typically have regular grid-like or blocky textures, bare soil areas have relatively uniform and smooth textures, and vegetated areas exhibit irregular flocculent textures. The terminal uses a Gray Level Co-occurrence Matrix (GLCM) as the texture feature extraction algorithm. This algorithm constructs a GLCM by statistically analyzing the probability of gray value combinations occurring between adjacent pixels at specific distances (e.g., 1 pixel, 3 pixels) and specific directions (e.g., 0°, 45°, 90°, 135°) in the image. Based on this GLCM, key texture parameters such as contrast, correlation, energy, and entropy are further calculated. These texture parameters effectively supplement the shortcomings of spectral features in distinguishing similar spectral features, providing an important basis for accurately distinguishing features with similar spectral characteristics but significant differences in texture features (such as bare soil and ungreened building areas), ultimately yielding texture feature data. Among them, contrast reflects the clarity and sharpness of the texture, correlation reflects the consistency of the texture, energy reflects the uniformity of the texture, and entropy reflects the complexity of the texture.

[0088] S24. Based on vegetation feature data, water feature data, and texture feature data, a random forest classifier is used to perform preliminary classification of land cover categories, resulting in initial land cover classification data.

[0089] For example, the terrain recognition terminal performs preliminary classification of land cover categories based on vegetation feature data, water body feature data, and texture feature data using a random forest classifier. Random Forest (RF) is a classification algorithm based on ensemble learning. It improves classification accuracy and stability by constructing multiple decision trees and combining the outputs of all decision trees. It has advantages such as resistance to overfitting, insensitivity to noisy data, and strong ability to handle high-dimensional data. The terminal fuses the above three types of feature data to form a high-dimensional feature vector for each pixel. At the same time, it selects sample data of known land cover categories (obtained through field surveys or visual interpretation of high-resolution images) as the training set to train the random forest model. During training, each decision tree extracts a portion of samples from the training set based on the Bootstrap resampling technique, and randomly selects some features for node splitting to ensure the independence of each decision tree. After training, the feature vector of the pixel to be classified is input into each decision tree, and each decision tree outputs a category judgment result. The terminal performs a vote on the output results of all decision trees and uses the category with the most votes as the initial category of the pixel. By initially classifying land cover types, the terminal can accurately classify all pixels in the image into vegetation, water bodies, bare soil, and building cover types, thus obtaining initial land cover classification data.

[0090] In an optional embodiment, based on the initial land feature classification data and digital elevation model data, slope and aspect analysis is performed on the elevation information to obtain terrain feature data, including the following steps:

[0091] S31, based on digital elevation model data, uses a variable window search method that takes into account terrain undulations to calculate the elevation change rate of pixels in the east-west and north-south directions in the elevation matrix, and obtains elevation gradient component data.

[0092] For example, the terrain recognition terminal calculates elevation gradient component data based on digital elevation model (DEM) data. The terminal employs a variable window search method that considers terrain undulations. The core advantage of this method is its ability to dynamically adjust the size of the analysis window according to the terrain complexity of different regions, avoiding the loss of details in complex terrain areas or the introduction of excessive noise in flat terrain areas by using a fixed window. In specific implementation, the terminal first preprocesses the DEM data, filling in holes and noise points, and then constructs an elevation matrix. For each pixel in the elevation matrix, the terrain complexity is determined by analyzing the elevation changes within a certain range around it: in flat terrain areas with small elevation changes (such as plains), a larger analysis window is used to reduce the impact of random noise; in steep terrain areas with drastic elevation changes (such as mountains and canyons), a smaller analysis window is used to accurately capture terrain details. Using this method, the terminal calculates the elevation change rate (i.e., the elevation gradient component) of each pixel in the east-west (x-axis) and north-south (y-axis) directions of the elevation matrix. and (where z is the elevation value of a pixel). These rates of change directly reflect the tilt trend and intensity of the land surface in the horizontal direction, forming elevation gradient component data, which provides basic data support for subsequent calculations of slope, aspect and curvature.

[0093] S32, based on elevation gradient component data, yields slope and aspect data.

[0094] For example, the terrain recognition terminal obtains slope data and aspect data based on elevation gradient component data. Slope data is obtained by calculating the magnitude of the elevation gradient component and performing an angle transformation; the specific formula is: Slope... ,slope Slope aspect data is typically expressed as an angle (0°~90°) or a percentage, used to quantitatively describe the steepness of the land surface slope. A larger value indicates a steeper slope, and vice versa. Slope aspect data is obtained by calculating the direction angle of the elevation gradient component; the specific formula is: Slope Aspect... Its numerical range is 0° to 360°, where 0° represents true north, 90° represents true east, 180° represents true south, and 270° represents true west. It is used to accurately describe the azimuth of a slope. Slope aspect not only affects the intensity and duration of solar radiation received by the earth's surface, but also indirectly affects environmental factors such as precipitation and temperature, thereby influencing the distribution patterns of land features. Combined with slope data, it can comprehensively reflect the basic topographic features of the earth's surface.

[0095] S33, based on elevation gradient component data, uses the following formula to obtain the profile curvature and planar curvature:

[0096]

[0097]

[0098] in, The curvature of the profile is used to characterize the surface's erosion and deposition capacity. Planar curvature is used to characterize the surface runoff capacity. Indicates elevation In the east-west direction rate of change on Indicates elevation In the north-south direction rate of change on East-west direction Elevation curvature, For mixed directional curvature, North-South Direction The elevation curvature.

[0099] In the above formula, The curvature of the profile is used to characterize the surface's erosion and deposition capacity. Planar curvature is used to characterize the surface runoff capacity. Indicates elevation In the east-west direction The rate of change of slope, i.e., the slope at... The directional component is used to reflect the east-west slope of the Earth's surface. Indicates elevation In the north-south direction The rate of change of slope, i.e., the slope at... The directional component is used to reflect the north-south tilt of the Earth's surface. East-west direction The elevation curvature reflects The degree of curvature of the earth's surface in a certain direction. The curvature is a mixture of directions, reflecting the surface curvature. , Combined bending characteristics in the direction, North-South Direction The elevation curvature reflects The degree of curvature of the earth's surface in a certain direction.

[0100] Specifically, the terrain recognition terminal calculates profile curvature and planar curvature based on elevation gradient component data and the formula described above. Profile curvature and plane curvature The second derivative of the elevation gradient components is obtained by performing a second derivative calculation on them. The second derivative of the elevation involved in the formula in the east-west direction is also included. Second derivative in the north-south direction and the second derivative in the mixing direction All values ​​are calculated through convolution operations on the elevation matrix or the finite difference method. Profile curvature characterizes the degree of surface curvature in the slope direction (i.e., the direction of maximum slope descent), and its positive and negative values ​​have clear geographical significance: positive values ​​indicate convex slopes, which are prone to accelerated water erosion; negative values ​​indicate concave slopes, which are prone to sediment deposition; and zero values ​​indicate straight slopes, where erosion and deposition are in a relative equilibrium. Planar curvature characterizes the degree of surface curvature perpendicular to the slope direction; positive values ​​indicate convex slopes with weaker runoff collection capacity; negative values ​​indicate concave slopes with stronger runoff collection capacity; and zero values ​​indicate planar slopes with stable runoff collection direction. Both reflect the complex curvature characteristics of the Earth's surface from different dimensions and are important parameters for analyzing topographic erosion, deposition processes, and hydrological processes.

[0101] S34. Based on the profile curvature and the plane curvature, the curvature characteristic data is obtained.

[0102] Specifically, the terrain recognition terminal obtains curvature feature data based on profile curvature and planar curvature. This curvature feature data is an integrated and quantified representation of profile curvature and planar curvature. The terminal first standardizes the values ​​of profile curvature and planar curvature to eliminate the influence of differences in their dimensions. Then, based on the actual needs of terrain analysis, it classifies the numerical ranges of both. For example, profile curvature is divided into five levels: strong erosion, weak erosion, neutral, weak deposition, and strong deposition; planar curvature is divided into five levels: strong confluence, weak confluence, neutral, weak divergence, and strong divergence. Finally, combining the classification results and the spatial distribution patterns of both, it constructs feature information that comprehensively reflects the type, intensity, and spatial differences of surface curvature, forming curvature feature data. This data supplements the deficiencies of slope and aspect in describing terrain details, making the representation of terrain features more comprehensive and accurate.

[0103] S35 integrates slope data, aspect data, and curvature feature data at multiple scales to generate terrain feature data.

[0104] For example, the terrain recognition terminal fuses slope data, aspect data, and curvature feature data at multiple scales to generate terrain feature data. The core idea of ​​multi-scale fusion is to integrate and optimize these three types of terrain data at different spatial scales (including local microscale, regional mid-scale, and global macroscale): at the local microscale (e.g., a single pixel and its surrounding small area), it focuses on preserving detailed terrain information, such as small-scale slope abrupt changes and local curvature variations; at the regional mid-scale (e.g., townships or watersheds), it highlights the local overall trend of the terrain, such as the dominant slope aspect and slope grade distribution in a certain area; at the global macroscale (e.g., the entire study area), it reflects the overall pattern and distribution law of the terrain, such as the approximate distribution range of mountains and plains within the area. The terminal uses a weighted fusion algorithm or wavelet transform fusion algorithm to assign corresponding weights according to the importance of different scales, organically combining terrain parameters at different scales to ultimately form comprehensive terrain feature data that accurately reflects local terrain features and clearly demonstrates the overall regional terrain law, comprehensively characterizing the surface undulations and orientation distribution within the region.

[0105] In an optional embodiment, based on initial land cover classification data and terrain feature data, the boundaries of each land cover type are optimized and classified to obtain refined land cover classification data, including the following steps:

[0106] S41. Based on the initial land cover classification data and terrain feature data, a spectral-terrain correlation map is obtained.

[0107] For example, the terrain recognition terminal constructs a spectral-terrain association map based on initial land cover classification data and terrain feature data. This graph model uses each pixel in the remote sensing image as an independent node. The attributes of each node contain two core pieces of information: first, spectral feature attributes, which come from the category label information and corresponding surface reflectance spectral data in the initial land cover classification data; second, terrain feature attributes, which come from terrain parameters such as slope, aspect, and curvature in the terrain feature data. The edges in the graph model are used to represent the association between nodes. The weight of the edges is determined by both feature similarity and spatial adjacency: feature similarity is obtained by calculating the spectral feature distance (such as Euclidean distance or cosine distance) and terrain feature distance between two nodes. The smaller the distance, the more similar the features are, and the greater the edge weight. Spatial adjacency is determined based on the spatial position relationship of pixels. Adjacent pixels (such as 4-neighborhood or 8-neighborhood pixels) have strong spatial adjacency and are assigned higher edge weights, while non-adjacent pixels have weak spatial adjacency and are assigned lower or zero edge weights. By constructing this association map, the terminal can quantify the intrinsic relationships between pixels, providing a solid topological foundation for the subsequent global optimization of land cover category labels.

[0108] S42, based on the spectral-terrain correlation map, performs global optimization on the initial land cover category labels to obtain optimized category label data.

[0109] For example, the terrain recognition terminal performs global optimization on the initial land cover category labels based on the spectral-terrain association map to obtain optimized category label data. The global optimization process is implemented through an energy minimization algorithm. The core objective is to ensure that the category labels of adjacent nodes in the spectral-terrain association map are as consistent as possible, conforming to the spatial continuity of land cover distribution. Specifically, the terminal defines an energy function for the association map, which includes a data term and a smoothing term: the data term measures the degree of matching between the node's initial category label and feature attributes; the higher the matching degree, the lower the energy value. The smoothing term measures the consistency of the category labels of adjacent nodes; the higher the consistency, the lower the energy value. The terminal adjusts the node's category labels by iteratively calculating and minimizing the energy function: for edges with high weights (pixels with similar features and spatial proximity), if their initial category labels are different, they are adjusted to be consistent with the labels of most surrounding nodes; for edges with low weights (pixels with large feature differences or large spatial distances), their initial category labels are retained or adjusted according to the feature matching degree. Through this process, the terminal can effectively eliminate isolated misclassified points and category conflict areas in the initial classification, making the distribution of land cover categories more consistent with actual geographical patterns, and obtaining optimized category label data.

[0110] S43, based on the optimized category label data, adopts a conditional random field model, integrates spatial context information, and performs semantic consistency smoothing on the boundaries of ground features to obtain refined ground feature classification data.

[0111] For example, the terrain recognition terminal uses a Conditional Random Field (CRF) model to perform semantic consistency smoothing on feature boundaries based on optimized category label data, resulting in refined feature classification data. The CRF model is a supervised learning algorithm based on a probabilistic graphical model that fully utilizes spatial context information to optimize classification results, avoiding the problem of neglecting inter-pixel relationships in traditional classification algorithms. When applying the CRF model, the terminal uses the optimized category label data as observations to construct a probabilistic model that includes pixel features and spatial neighborhood relationships. By introducing semantic consistency constraints, such as the correlation between vegetation boundaries and terrain slope aspect changes, the continuous distribution of water body boundaries along low-slope lines, and the regular geometric shape of building area boundaries, the terminal calculates the posterior probability of each pixel belonging to a specific feature category. Based on this posterior probability, the terminal performs point-by-point correction on ambiguous feature boundaries, ensuring that the boundaries conform to both the spectral and topographical features of the pixels and the semantic consistency requirements (e.g., boundaries of the same feature type should be continuous, smooth, and logically reasonable), ultimately obtaining refined feature classification data with accurate boundaries, clear categories, and good spatial continuity.

[0112] In an optional embodiment, a terrain identification result is generated based on detailed land cover classification data and terrain feature data, including the following steps:

[0113] S51, based on detailed land cover classification data, calculates the area proportion of each land cover type using spatial statistical methods to obtain land cover structure data.

[0114] For example, the terrain recognition terminal calculates the area proportion of each feature type based on refined feature classification data using spatial statistical methods to obtain feature cover structure data. The spatial statistical methods employed by the terminal include grid statistical methods and vector surface statistical methods. Specifically, the refined feature classification data in raster format is first converted into vector surface features, each representing a continuous distribution area of ​​a feature type and accompanied by corresponding category attribute information. Then, spatial geometric calculation algorithms are used to calculate the actual area of ​​each vector surface (the area unit can be set to square meters, hectares, etc., as needed). Finally, the ratio of the total area of ​​each feature type to the total area of ​​the study area is calculated to obtain the area proportion of each feature type. The feature cover structure data is presented in the form of percentages or area numerical tables, clearly reflecting the distribution ratio and spatial pattern of vegetation, water bodies, bare soil, buildings, and other types within the region. It is an important indicator characterizing the regional ecological environment, land use efficiency, and land cover characteristics.

[0115] S52, based on terrain feature data, divides landform units through cluster analysis to obtain landform unit division data.

[0116] For example, the terrain recognition terminal divides landform units based on terrain feature data through cluster analysis, obtaining landform unit division data. Cluster analysis is an unsupervised learning method whose core principle is to group samples with similar characteristics into the same category, minimizing the feature differences within the same category and maximizing the feature differences between different categories. The terminal uses mature clustering algorithms, with key parameters such as slope, aspect, and curvature from the terrain feature data as clustering input variables. First, the input variables are standardized to eliminate the influence of differences in the dimensions of different parameters. Then, a reasonable initial value for the number of clusters is set (which can be based on prior knowledge of the terrain in the study area or determined by the elbow rule). By iteratively calculating the distance between the samples and the cluster centers (such as Euclidean distance or Manhattan distance), the positions of the cluster centers are continuously adjusted until the clustering results converge. Finally, areas with similar terrain features are divided into the same landform unit (such as mountains, hills, plains, valleys, etc.), obtaining landform unit division data that clearly reflects the terrain differentiation patterns within the region. Clustering algorithms such as K-means, hierarchical clustering, and density clustering can be used.

[0117] S53, based on land cover structure data and geomorphic unit division data, generates terrain recognition results through association rule mining, thus obtaining terrain recognition results.

[0118] Specifically, the terrain recognition terminal generates terrain recognition results based on land cover structure data and geomorphic unit segmentation data through association rule mining. Association rule mining is a data mining technique used to discover inherent relationships between different datasets; commonly used algorithms include the Apriori algorithm and the FP-Growth algorithm. The terminal uses land cover structure data (area proportion and distribution location of each land cover type) and geomorphic unit segmentation data (type, extent, and terrain features of geomorphic units) as mining objects. By setting minimum support and minimum confidence thresholds, it filters out statistically significant association rules. For example, it mines association rules such as vegetation cover area exceeding 60% in mountainous geomorphic units, the sum of building cover and bare soil cover exceeding 50% in plain geomorphic units, and water cover significantly higher in valley geomorphic units than in other geomorphic units. The terminal organically integrates these association rules with geomorphic unit segmentation data and land cover structure data, ultimately forming terrain recognition results that include the spatial distribution of geomorphic units, the proportion of land cover within each unit, and the relationships between them, comprehensively meeting the application needs of terrain feature recognition.

[0119] In an optional embodiment, based on terrain feature data, geomorphic units are divided through cluster analysis to obtain geomorphic unit division data, including the following steps:

[0120] S61, perform cluster analysis on the terrain feature data to obtain the initial clustering results.

[0121] For example, the terrain recognition terminal performs cluster analysis on terrain feature data to obtain initial clustering results. The terminal first preprocesses the terrain feature data: integrating terrain parameters of different dimensions such as slope, aspect, and curvature into high-dimensional feature vectors. Standardization or normalization methods are then used to process the feature vectors, eliminating the influence of differences in dimensions and numerical ranges between different parameters, ensuring that each feature has equal weight in the clustering process. Subsequently, a suitable clustering algorithm is selected, commonly the K-means algorithm. Based on the terrain complexity of the study area and the actual application requirements, the initial number of cluster centers is preset (this can be initially determined based on prior knowledge of the terrain or through trial and error). During the algorithm execution, the distance (such as Euclidean distance) between each sample (i.e., the feature vector of each pixel) and each cluster center is calculated, and the sample is assigned to the category of the nearest cluster center. Then, the mean of each category is recalculated as the new cluster center, and the above assignment and update process is repeated until the position change of the cluster center is less than the preset threshold or the maximum number of iterations is reached. The clustering process terminates, and finally the initial clustering result of classifying pixels with similar terrain features into one category is obtained. Each category represents a potential geomorphic unit.

[0122] S62 evaluates the clustering effect of the initial clustering results and obtains clustering evaluation data.

[0123] For example, the terrain recognition terminal evaluates the clustering effect of the initial clustering results to obtain clustering evaluation data. The terminal uses various clustering effect evaluation indicators to quantify the rationality of the initial clustering results from different dimensions. Commonly used indicators include the Silhouette Coefficient, Davies-Bouldin Index (DBI), and Calinski-Harabasz Index (CHI). The Silhouette Coefficient calculates the silhouette value of each sample by measuring the average distance (cohesion) between each sample and other samples in the same class and the average distance (dissociation) with the nearest sample from a different class. The average of all sample silhouette values ​​is the overall Silhouette Coefficient, which ranges from -1 to 1. The closer the value is to 1, the better the clustering effect, the more compact the samples within each class, and the more obvious the separation between classes. The Davies-Bouldin Index calculates the ratio of the average intra-class distance to the inter-class distance of each class and takes the average of all ratios. The smaller the value, the better the clustering effect. The Calinski-Harabasz index is calculated by giving the ratio of the sum of squared deviations between clusters to the sum of squared deviations within clusters; a higher value indicates better clustering. By calculating these evaluation metrics, the system obtains comprehensive clustering evaluation data that reflects the quality of the initial clustering results, providing a clear basis for subsequent clustering optimization.

[0124] S63, based on cluster evaluation data, obtains geomorphic unit division data by iteratively optimizing and adjusting cluster centers.

[0125] For example, the terrain recognition terminal obtains landform unit division data by iteratively optimizing and adjusting cluster centers based on cluster evaluation data. If the cluster evaluation data shows that the initial clustering effect is poor (e.g., the silhouette coefficient is lower than the preset threshold, or the Davies-Bouldin index is higher than the preset threshold), the terminal adjusts the optimization strategy according to the evaluation results: if the difference between samples within a class is too large, the number of cluster centers can be appropriately increased; if the separation between classes is insufficient, the position of the initial cluster centers can be adjusted or the distance calculation method can be optimized; then the clustering analysis process is re-executed. If the evaluation results show that the clustering effect is good, but there is still room for local optimization, the position of the cluster centers is further fine-tuned to further reduce the feature differences between samples within a class and further increase the differences between classes. Through multiple iterative optimizations, the terminal continuously corrects the clustering results until the cluster evaluation data reaches the preset optimal threshold. At this point, the clustering results can accurately reflect the natural differentiation pattern of the terrain in the region, and each cluster category corresponds to a landform unit with unique and consistent terrain features, ultimately obtaining landform unit division data.

[0126] The aforementioned terrain feature recognition method based on remote sensing technology integrates remote sensing imagery and digital elevation model data, and obtains high-quality surface reflectance data through standardized preprocessing, laying a reliable foundation for subsequent analysis. It achieves preliminary land cover classification based on multi-dimensional spectral and texture features, and fully explores the intrinsic relationship between land cover and terrain by combining comprehensive terrain parameters such as slope, aspect, and curvature. Through graph model construction, global optimization, and conditional random field smoothing, it effectively solves problems such as blurred boundaries and misclassification in the initial classification, improving the accuracy and continuity of land cover classification. Finally, through spatial statistics, cluster analysis, and association rule mining, it achieves the organic integration of geomorphic unit division and land cover structure, comprehensively avoiding the shortcomings of traditional methods such as insufficient data accuracy, coarse classification, and one-sided terrain representation, improving the comprehensiveness, reliability, and practicality of terrain recognition results, and adapting to the actual needs of high-precision geospatial information extraction.

[0127] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0128] Based on the same inventive concept, this application also provides a terrain feature recognition device based on remote sensing technology for implementing the terrain feature recognition method based on remote sensing technology described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more terrain feature recognition device embodiments based on remote sensing technology provided below can be found in the limitations of the terrain feature recognition method based on remote sensing technology above, and will not be repeated here.

[0129] In one exemplary embodiment, such as Figure 2 As shown, a terrain feature recognition device 200 based on remote sensing technology is provided, comprising:

[0130] The data acquisition and preprocessing module 201 is used to acquire remote sensing image data and preprocess the remote sensing image data to obtain surface reflectance data; the surface reflectance data is used to characterize the reflectance characteristics of ground objects in different bands.

[0131] The initial land cover classification module 202 is used to perform preliminary land cover classification on each pixel in the remote sensing image data based on the surface reflectance data, so as to obtain the initial land cover classification data.

[0132] The terrain feature module 203 is used to perform slope and aspect analysis on elevation information based on the initial land feature classification data and digital elevation model data to obtain terrain feature data; the terrain feature data is used to characterize the surface undulation and orientation distribution within the region;

[0133] The land feature classification optimization module 204 is used to optimize and classify the boundaries of various land feature types based on the initial land feature classification data and terrain feature data, so as to obtain refined land feature classification data.

[0134] The terrain recognition module 205 is used to generate terrain recognition results based on detailed land feature classification data and terrain feature data; the terrain recognition results include landform unit division and land feature cover structure.

[0135] Furthermore, the data acquisition and preprocessing module 201 is also used for:

[0136] Radiometric calibration processing is performed on remote sensing image data to convert the raw digital quantization values ​​into radiance values, thus obtaining radiometric calibration data;

[0137] Based on radiometric calibration data, atmospheric scattering and absorption effects are eliminated by using a correction model to obtain atmospheric correction data;

[0138] Geometric correction is performed on the atmospheric correction data to obtain geometrically corrected data.

[0139] Based on geometrically corrected data, the surface reflectance is calculated using the following formula to obtain the surface reflectance data:

[0140]

[0141] in, For surface reflectance, For band The radiance value, This is the Earth-Sun distance correction factor. For band Solar irradiance at the top of the atmosphere, This is the solar zenith angle.

[0142] Furthermore, the initial feature classification module 202 is also used for:

[0143] Vegetation cover was calculated using the normalized vegetation index based on surface reflectance data, and vegetation characteristic data were obtained.

[0144] Based on surface reflectance data, water body information is extracted through normalized differential water body index to obtain water body characteristic data;

[0145] Texture feature data is obtained by calculating texture features based on surface reflectance data and gray-level co-occurrence matrix.

[0146] Based on vegetation feature data, water body feature data, and texture feature data, a random forest classifier is used to perform preliminary classification of land cover categories, resulting in initial land cover classification data.

[0147] Furthermore, the terrain feature module 203 is also used for:

[0148] Based on digital elevation model data, the variable window search method considering terrain undulation is used to calculate the elevation change rate of pixels in the east-west and north-south directions in the elevation matrix, and obtain elevation gradient component data.

[0149] Based on the elevation gradient component data, slope data and aspect data are obtained; slope data is used to describe the degree of inclination of the ground surface; aspect data is used to describe the azimuth angle of the slope.

[0150] Based on the elevation gradient component data, the profile curvature and plane curvature are obtained using the following formulas:

[0151]

[0152]

[0153] in, The curvature of the profile is used to characterize the surface's erosion and deposition capacity. Planar curvature is used to characterize the surface runoff capacity. Indicates elevation In the east-west direction rate of change on Indicates elevation In the north-south direction rate of change on East-west direction Elevation curvature, For mixed directional curvature, North-South Direction Elevation curvature;

[0154] Curvature characteristic data are obtained based on the profile curvature and the planar curvature;

[0155] Slope data, aspect data, and curvature feature data are fused at multiple scales to generate terrain feature data.

[0156] Furthermore, the land feature classification optimization module 204 is also used for:

[0157] Based on the initial land cover classification data and topographic feature data, a spectral-topographic correlation map is obtained;

[0158] Based on the spectral-topography correlation map, the initial land cover category labels are globally optimized to obtain optimized category label data;

[0159] Based on the optimized category label data, a conditional random field model is used to integrate spatial context information and perform semantic consistency smoothing on the boundaries of ground features to obtain refined ground feature classification data.

[0160] Furthermore, the terrain recognition module 205 is also used for:

[0161] Based on detailed land cover classification data, the area proportion of each land cover type is calculated using spatial statistical methods to obtain land cover structure data;

[0162] Based on topographic feature data, geomorphic units are divided through cluster analysis to obtain geomorphic unit division data;

[0163] Based on land cover structure data and geomorphic unit division data, terrain identification results are generated through association rule mining.

[0164] Furthermore, the terrain recognition module 205 is also used for:

[0165] Cluster analysis was performed on the terrain feature data to obtain the initial clustering results;

[0166] The initial clustering results are evaluated to assess the clustering effectiveness, and clustering evaluation data is obtained.

[0167] Based on the cluster evaluation data, the cluster centers are iteratively optimized and adjusted to obtain the geomorphic unit division data.

[0168] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the terrain feature recognition method based on remote sensing technology as described above.

[0169] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0170] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0171] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for terrain feature recognition based on remote sensing technology, characterized in that, The method includes: Remote sensing image data is acquired and preprocessed to obtain surface reflectance data; the surface reflectance data is used to characterize the reflectance characteristics of ground objects in different spectral bands. Based on the surface reflectance data, preliminary land cover classification is performed on each pixel in the remote sensing image data to obtain initial land cover classification data. Based on the initial land feature classification data and the digital elevation model data, slope and aspect analysis is performed on the elevation information to obtain terrain feature data; the terrain feature data is used to characterize the surface undulation and orientation distribution within the region. Based on the initial land cover classification data and the terrain feature data, the boundaries of each land cover type are optimized and classified to obtain refined land cover classification data. Based on the detailed land feature classification data and the terrain feature data, a terrain identification result is generated; the terrain identification result includes landform unit division and land feature cover structure.

2. The method according to claim 1, characterized in that, The preprocessing of the remote sensing image data to obtain surface reflectance data includes: The remote sensing image data is subjected to radiometric calibration processing to convert the original digital quantization values ​​into radiance values, thereby obtaining radiometric calibration data; Based on the aforementioned radiometric calibration data, atmospheric scattering and absorption effects are eliminated by using a correction model to obtain atmospheric correction data. The atmospheric correction data is subjected to geometric correction processing to obtain geometric correction data; Based on the geometric correction data, the surface reflectance is calculated using the following formula to obtain the surface reflectance data: in, For surface reflectance, For band The radiance value, This is the Earth-Sun distance correction factor. For band Solar irradiance at the top of the atmosphere, This is the solar zenith angle.

3. The method according to claim 1, characterized in that, Based on the surface reflectance data, preliminary land cover classification is performed on each pixel in the remote sensing image data to obtain initial land cover classification data, including: Based on the surface reflectance data, vegetation coverage is calculated using the normalized vegetation index to obtain vegetation characteristic data. Based on the surface reflectance data, water body information is extracted through the normalized differential water body index to obtain water body characteristic data; Based on the surface reflectance data, texture features are calculated using the gray-level co-occurrence matrix to obtain texture feature data; Based on the vegetation feature data, water feature data, and texture feature data, a random forest classifier is used to perform preliminary classification of land cover categories, resulting in the initial land cover classification data.

4. The method according to claim 1, characterized in that, The step involves analyzing the elevation information based on the initial land cover classification data and the digital elevation model data, performing slope and aspect analysis to obtain terrain feature data, including: Based on the digital elevation model data, the elevation change rate of the pixels in the east-west and north-south directions in the elevation matrix is ​​calculated by using the variable window search method that takes into account the terrain undulation, and the elevation gradient component data is obtained. Based on the elevation gradient component data, slope data and aspect data are obtained; the slope data is used to describe the degree of inclination of the ground surface; the aspect data is used to describe the azimuth angle of the slope. Based on the elevation gradient component data, the profile curvature and plane curvature are obtained using the following formulas: in, The curvature of the profile is used to characterize the surface erosion and deposition capacity. Planar curvature is used to characterize the surface runoff capacity. Indicates elevation In the east-west direction rate of change on Indicates elevation In the north-south direction rate of change on East-west direction Elevation curvature, For mixed directional curvature, North-South Direction Elevation curvature; Based on the cross-sectional curvature and the planar curvature, curvature characteristic data is obtained; The slope data, aspect data, and curvature feature data are fused at multiple scales to generate the terrain feature data.

5. The method according to claim 1, characterized in that, Based on the initial land cover classification data and the terrain feature data, the boundaries of each land cover type are optimized and classified to obtain refined land cover classification data, including: Based on the initial land cover classification data and terrain feature data, a spectral-terrain correlation map is obtained; Based on the spectral-topography correlation map, the initial land cover category labels are globally optimized to obtain optimized category label data; Based on the optimized category label data, a conditional random field model is used to fuse spatial context information and perform semantic consistency smoothing on the boundaries of ground features to obtain the refined ground feature classification data.

6. The method according to claim 1, characterized in that, The step of generating terrain recognition results based on the detailed land cover classification data and the terrain feature data includes: Based on the aforementioned detailed land cover classification data, the area proportion of each land cover type is calculated using spatial statistical methods to obtain land cover structure data; Based on the terrain feature data, geomorphic units are divided through cluster analysis to obtain geomorphic unit division data; Based on the land cover structure data and landform unit division data, terrain identification results are generated through association rule mining to obtain the terrain identification results.

7. The method according to claim 6, characterized in that, The process of dividing landform units based on the terrain feature data through cluster analysis to obtain landform unit division data includes: Cluster analysis was performed on the terrain feature data to obtain initial clustering results; The initial clustering results are evaluated to assess the clustering effectiveness, resulting in clustering evaluation data. Based on the clustering evaluation data, the cluster centers are iteratively optimized and adjusted to obtain the geomorphic unit division data.

8. A terrain feature recognition device based on remote sensing technology, characterized in that, The device includes: The data acquisition and preprocessing module is used to acquire remote sensing image data and preprocess the remote sensing image data to obtain surface reflectance data; the surface reflectance data is used to characterize the reflectance characteristics of ground objects in different spectral bands. The initial land cover classification module is used to perform preliminary land cover classification on each pixel in the remote sensing image data based on the surface reflectance data, so as to obtain initial land cover classification data. The terrain feature module is used to perform slope and aspect analysis on the elevation information based on the initial land feature classification data and the digital elevation model data to obtain terrain feature data; the terrain feature data is used to characterize the surface undulation and orientation distribution within the region; The land feature classification optimization module is used to optimize and classify the boundaries of various land feature types based on the initial land feature classification data and the terrain feature data, so as to obtain refined land feature classification data. The terrain recognition module is used to generate terrain recognition results based on the detailed land feature classification data and the terrain feature data; the terrain recognition results include landform unit division and land feature cover structure.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images