Soil water erosion feature extraction method based on image segmentation
By preprocessing remote sensing images and dividing them into topographic functional units, and combining the energy minimization theory and the narrowband level set method, the problems of accuracy and computational efficiency in soil water erosion feature extraction in traditional methods are solved, and high-precision, smooth water erosion area extraction is achieved.
Patent Information
- Application Number
- CN202610120434.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-28
- Publication Date
- 2026-05-05
AI Technical Summary
Traditional methods for extracting soil water erosion features based on image segmentation suffer from mixed pixels, noise, and blurred boundaries of land features in remote sensing images. This results in significant deviations in the spatial location, extent, and geometric accuracy of the extracted water erosion features, making it difficult to meet the needs of high-precision quantitative analysis and remediation engineering planning.
High-resolution remote sensing images were acquired and preprocessed to extract soil spectral, texture, and exposure features. Topographic functional units were divided using a digital elevation model, a hierarchical segmentation rule base was constructed, and a water erosion contour extraction model based on energy minimization theory was iteratively optimized. The narrowband level set method was then used for calculation.
It significantly improves the accuracy and computational efficiency of soil water erosion feature extraction, achieves sub-pixel-level positioning accuracy and contour smoothness, meets the needs of high-precision quantitative analysis and remediation projects, and reduces computational costs.
Smart Images
Figure CN121982541A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of water erosion feature extraction technology, specifically a method for extracting soil water erosion features based on image segmentation. Background Technology
[0002] Soil water erosion is a key factor leading to land degradation and ecological deterioration. Utilizing remote sensing technology to automatically extract large-scale, high-precision soil water erosion features is of great significance for soil and water conservation planning and effectiveness evaluation. With the widespread availability of high-resolution remote sensing imagery, accurately identifying and delineating the spatial extent and fine contours of soil water erosion areas from complex land cover backgrounds has become a research hotspot and technical challenge in this field.
[0003] Traditional methods for extracting soil water erosion features based on image segmentation first preprocess the acquired remote sensing images, then preliminarily identify water erosion areas by calculating spectral indices and combining them with texture feature analysis. Then, image segmentation methods based on fixed thresholds or edge detection operators are used to segment the water erosion areas and directly generate the boundaries of the water erosion areas as the final soil water erosion feature output.
[0004] However, after the above-mentioned traditional methods are completed, due to the presence of mixed pixels, noise and blurred boundaries of ground features in remote sensing images, the contours obtained by direct segmentation often appear jagged, broken or irregular in shape, which seriously deviates from the smoothness and continuity that the real water erosion boundary should have. As a result, the extracted water erosion features have significant deviations in spatial location, range and contour geometric accuracy, making it difficult to meet the needs of high-precision quantitative analysis and remediation engineering planning. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides a method for extracting soil water erosion features based on image segmentation, thus solving the problems existing in the background technology.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for extracting soil water erosion features based on image segmentation, comprising the following steps: Step S1: Acquire the original soil remote sensing image of the target area, and preprocess the original soil remote sensing image to obtain the preprocessed soil remote sensing image. Step S2: Extract features from the preprocessed soil remote sensing image to obtain soil spectral features, soil texture features and soil exposure features, and fuse them to generate a water erosion sensitivity layer; Step S3: Collect digital elevation model data of the target area, calculate the terrain factor based on the digital elevation model data, divide the target area into different terrain functional units according to the terrain factor, and form a hierarchical segmentation rule base based on the different terrain functional units and preset differentiated decision thresholds. Step S4: Based on the hierarchical segmentation rule base, construct a terrain segmentation mechanism, and segment the water erosion sensitivity layer through the terrain segmentation mechanism to obtain a water erosion area segmentation map; Step S5: Construct a water erosion contour extraction model based on the energy minimization theory, input the water erosion region segmentation map into the water erosion contour extraction model, and output the soil water erosion features.
[0007] Preferably, acquiring raw soil remote sensing images of the target area and preprocessing the raw soil remote sensing images to obtain preprocessed soil remote sensing images includes: Data collection was conducted, and a satellite imaging sensor was selected to acquire high-resolution soil remote sensing images of the target area at a single time phase, resulting in the raw soil remote sensing images. ; Remote sensing images of the original soil A standardized preprocessing process is implemented; the first step in preprocessing is the radiation correction step, which converts the raw digital quantization values into surface reflectance using an atmospheric correction model. Geometric fine correction is performed using a coordinate transformation function model:
[0008] in, The coordinates of the original soil remote sensing image; These are the corrected geographic coordinates. It is a rational function model; Represents the set of ground control points; After the above preprocessing steps, the original soil remote sensing image The soil remote sensing image is converted into a preprocessed soil remote sensing image. .
[0009] Preferably, feature extraction is performed on the preprocessed soil remote sensing image to obtain soil spectral features, soil texture features, and soil exposure features, and these are then fused to generate a water erosion sensitivity layer, including: Multi-dimensional feature extraction is performed on preprocessed soil remote sensing images. Three physical characteristic components that are indicative of water erosion phenomena were identified in the analysis: soil spectral characteristics, soil texture characteristics, and soil exposure characteristics. Soil spectral characteristics It is a linear combination of vegetation index and soil brightness index to comprehensively characterize spectral anomalies; The entropy value is calculated using the gray-level co-occurrence matrix to measure the complexity and irregularity of the texture, thus obtaining soil texture features. ; The characteristics of soil exposure were obtained using the Normalized Differential Exposure Index (NDBI). ; After obtaining the aforementioned soil spectral features, soil texture features, and soil exposure features, feature normalization processing is required. The normalized features are then weighted and fused pixel by pixel to generate a unified water erosion sensitivity layer. .
[0010] Preferably, digital elevation model data of the target area is collected, and based on the digital elevation model data, terrain factors are calculated, including: Collect digital elevation model data covering the target area, denoted as The data source originates from publicly available databases of satellite-borne radar or space shuttle radar terrain mapping missions; digital elevation model data... Preprocessing is performed to fill data gaps and smooth noise, resulting in reliable preprocessed digital elevation model data, denoted as . ; Based on preprocessed digital elevation model data Perform terrain factor calculations to provide quantitative indicators for unit division, such as slope. The calculation is as follows:
[0011] in, and These represent the rates of elevation change of the DEM in the east-west and north-south directions, respectively; arctan( ) is the arctangent function; These are partial differential symbols; x and y represent preprocessed digital elevation model data, respectively. The east-west and north-south coordinate axes; slope It is a core factor controlling surface runoff energy and erosion potential.
[0012] Preferably, dividing the target area into different terrain functional units based on the terrain factors includes: Based on geomorphological and water erosion principles, and using calculated topographic factors, the target area is divided into three basic topographic functional units, which are then set as follows: These represent the top, middle, and bottom areas of the slope, respectively.
[0013] Preferably, based on different terrain functional units, a hierarchical segmentation rule base is formed by pre-setting differentiated decision thresholds, including: Set a key differentiated decision threshold for each unit. The thresholds of all units and their corresponding relationships constitute a hierarchical segmentation rule base. The core mapping relationship of this rule base is simplified to: That is, for any unit The threshold is the criterion for determining whether a pixel is classified as a water erosion zone. The rule base is built based on the following core principles: In the slope crest area where erosion signals are weak Set a relatively low decision threshold ; Mid-slope area dominated by linear erosion Set a moderate decision threshold In the foothills of slopes where sedimentation is significant and boundaries are indistinct. Set a relatively high decision threshold .
[0014] Preferably, the terrain segmentation mechanism, based on a hierarchical segmentation rule base, includes: A terrain segmentation mechanism, M, is constructed to transform geographical knowledge into image segmentation decisions. Its core is establishing a mapping relationship from spatial location to a dynamic decision threshold for any pixel in the image. The mechanism operates as follows: First, perform terrain unit localization, based on pixels. The geographic coordinates are used to determine the topographic functional unit to which it belongs. ; Query the rule base again Obtain the terrain unit Preset decision threshold and assign it to the pixel. This process can be formalized using the following mapping function:
[0015] in, It is a pixel The minimum sensitivity threshold required to be classified as a water erosion area is determined by the topographic unit in which it is located. It is a query function that takes a terrain unit type as input and then returns the best empirical threshold preset for that type of terrain from a hierarchical segmentation rule base.
[0016] Preferably, the water erosion sensitivity layer is segmented using the terrain segmentation mechanism to obtain a water erosion area segmentation map, including: The water erosion sensitivity layer is segmented using a terrain segmentation mechanism to obtain the dynamic decision threshold for each pixel. Then, the water erosion sensitivity layer The segmentation decision is performed according to the following rule: if the sensitivity value of a pixel is not lower than the threshold required by the terrain unit where it is located, then it is determined to be a water-eroded area. The segmentation decision formula is as follows:
[0017] in, It is a water erosion area segmentation map in pixels The value at the location; 1 represents a water erosion area, 0 represents a non-water erosion area; continuously displaying water erosion sensitivity information. Based on a reasonable critical criterion that varies with topographical space. Binarization is performed to generate a preliminary water erosion area segmentation map. .
[0018] Preferably, the water erosion profile extraction model constructed based on the energy minimization theory includes: The model for extracting water erosion profiles is constructed. This model uses the level set method as a framework to transform the boundary evolution problem into an energy functional problem. The minimization problem of the level set function It is an implicitly expressed scalar function with a zero level set. The outline of the water-eroded region is shown; the total energy functional is as follows:
[0019] in, For image data items, it drives the contour to fit the gradient edges in the image; For structural constraints, prior knowledge of the texture and spectral continuity of the water erosion region is introduced; As a regularization term, it constrains the smoothness of the contour and avoids unreasonable bends; These are weighting coefficients used to balance the contributions of each item; Model optimization employs a narrow-band level set method for iterative solution to efficiently solve the aforementioned energy functional; after the soil water erosion feature extraction model converges, the final level set function is used. Extract its zero level set :
[0020] in, It is the domain; An element in a set is represented by a point. This refers to the set of all points that form a curve.
[0021] Preferably, the narrowband level set method includes: Initialize the binary segmentation image. Transform it into a signed distance function, and use it as a level set function. The initial value is determined, and a structure is constructed centered at the zero level set with a width of [value missing]. The initial narrow band; In iterative evolution, that is, in the th... In this iteration, the following steps are performed: First, calculate the evolution velocity. For all pixels within the current narrow band, calculate the evolution velocity field of the level set function. The velocity field is determined by the gradient descent flow of the energy functional, which includes driving forces from image gradient, texture structure constraints and curvature smoothing. Using the explicit Euler scheme, update the level set function only within a narrow band:
[0022] in, It is a level set function; superscript It is an iteration count index; Indicates after the first The boundary state after the next iteration of optimization; For time step; Then perform narrowband updates and reinitialization, checking if the zero-level set has moved close to the boundary of the current narrowband; if so, reconstruct the narrowband with the new zero-level set as the center, and process the new narrowband... Perform a quick reinitialization; Repeat the above process iteratively until the change in the position of the zero level set is less than a predetermined threshold. or reaching the maximum number of iterations. At this point, the water erosion profile extraction model is considered to have converged, yielding the final level set function. .
[0023] This invention provides a method for extracting soil water erosion features based on image segmentation, involving machine learning and deep learning techniques, which has the following beneficial effects: (1) This image segmentation-based method for extracting soil water erosion features scientifically divides the target area into topographic functional units with different water erosion development patterns, such as slope top, middle slope, and slope foot, based on quantitative analysis of topographic relief data (DEM) and topographic factors such as slope and aspect. On this basis, differentiated decision thresholds are formulated for different units to form a hierarchical segmentation rule base. This design transforms the image segmentation process from blind pixel processing to a guided geographical cognition process, significantly improving the geographical rationality and mechanistic consistency of the segmentation results, and laying a reliable spatial framework foundation for subsequent high-precision extraction.
[0024] (2) This image segmentation-based method for extracting soil water erosion features refines the rough boundaries of the initial segmentation using a water erosion contour extraction model based on energy minimization theory. The energy function defined by this model integrates multiple constraints such as image gradient, consistency of local texture and spectral structure, and contour smoothness. By iteratively optimizing and driving boundary evolution, it can effectively overcome the jaggedness, breakage, and positioning deviation problems existing in the initial segmentation boundary. This enables the output water erosion area vector boundary to achieve sub-pixel-level positioning accuracy, with smooth and continuous contours, while perfectly preserving the morphological details of real water erosion gullies. As a result, the extracted soil water erosion features achieve high fidelity in both spatial range and geometric morphology, meeting the engineering requirements for quantitative analysis and precise management.
[0025] (3) The soil water erosion feature extraction method based on image segmentation innovatively introduces a narrow-band level set method to replace the traditional global update algorithm when iteratively solving the energy minimization model. This method calculates and updates the level set function only within a narrow band-shaped region around the current active boundary, thereby reducing the computational complexity of the algorithm from being proportional to the image area to being proportional to the boundary perimeter. This improvement brings a significant increase in computational efficiency, enabling the optimization process of complex boundaries for large-scale, high-resolution remote sensing images to be completed within an acceptable time. It successfully solves the problem of high computational cost that is usually associated with high-precision models, greatly enhancing the feasibility and timeliness of this method in large-scale practical applications. Attached Figure Description
[0026] Figure 1 This is a flowchart of a method for extracting soil water erosion features based on image segmentation proposed in this invention.
[0027] Figure 2 This invention proposes a method for extracting soil water erosion features based on image segmentation to obtain a hierarchical map of the terrain segmentation mechanism.
[0028] Figure 3 This is a hierarchical map of soil water erosion features obtained in a soil water erosion feature extraction method based on image segmentation proposed in this invention. Detailed Implementation
[0029] 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.
[0030] Please see Figures 1-3This invention provides a technical solution: a method for extracting soil water erosion features based on image segmentation. Specifically, the method for extracting soil water erosion features based on image segmentation is provided below. Figure 1 The method includes the following steps: Step S1: Acquire the original soil remote sensing image of the target area, and preprocess the original soil remote sensing image to obtain the preprocessed soil remote sensing image.
[0031] This step is the data preparation stage for the soil water erosion feature extraction method. Its core objective is to acquire a single-temporal, high-quality raw remote sensing image of the target area, and through a series of standardization processes, obtain a preprocessed image whose radiometric and geometric accuracy meets the requirements of subsequent analysis. The output of this step is the sole data basis for subsequent feature extraction operations. The specific execution flow is as follows: First, data collection is performed by selecting a satellite imaging sensor to acquire high-resolution soil remote sensing images of the target area at a single time phase, thus obtaining the raw soil remote sensing images. Remote sensing images of original soil The quality requirements should be met, including a spatial resolution better than 2 meters and a cloud cover of less than 10%, to ensure that the textures and morphological features related to soil water erosion are clearly distinguishable.
[0032] Subsequently, the original soil remote sensing images A standardized preprocessing workflow is implemented. The first step in preprocessing is radiometric correction, which converts the raw digital quantization (DN) values into surface reflectance using an atmospheric correction model. This is an effective technique for achieving radiometric consistency between images, and the process can be represented as follows:
[0033] in, λ is the surface reflectance, λ refers to the multispectral band of the remote sensing satellite, and its wavelength range and center wavelength information can be extracted from the metadata file of the remote sensing image; DN is the raw digital value recorded by the satellite imaging sensor, which has no physical dimension. The solar elevation angle is extracted from image metadata. Multispectral band Atmospheric optical thickness is extracted from image metadata; ( The ) represents the physical model for radiometric calibration and atmospheric correction; here, the 6S model (an atmospheric correction model based on radiative transfer theory) is used. This ensures that subsequent feature extraction is based on a stable and comparable radiometric reference.
[0034] Next, geometric fine correction is performed. This step aims to eliminate geometric distortions caused by satellite imaging sensor errors and terrain undulations, and is achieved through a coordinate transformation function model:
[0035] in, The coordinates of the original soil remote sensing image; These are the corrected geographic coordinates. It is a rational function model (RFM); This represents the set of ground control points; its purpose is to ensure the absolute accuracy of the spatial location of the image, providing geometric assurance for the subsequent accurate extraction of water erosion boundaries. This formula determines the coordinates from the original soil remote sensing image. To corrected geographic coordinates The mapping relationship.
[0036] Finally, after the above preprocessing steps, the original soil remote sensing image... The soil remote sensing image is converted into a preprocessed soil remote sensing image. .
[0037] This step is the data preparation and preprocessing stage. Its core task is to acquire a single-temporal high-resolution raw remote sensing image of the target area, and through a series of standardization processes such as radiometric calibration, atmospheric correction, and geometric fine correction, eliminate sensor errors, atmospheric interference, and geometric distortions, ultimately generating a preprocessed soil remote sensing image with accurate radiometric characteristics and precise geometric location. This ensures the accuracy and reliability of the spectral and texture information upon which feature extraction in step S2 depends, providing a fundamental guarantee for constructing a high-quality water erosion sensitivity layer. Without rigorous image preprocessing, all subsequent advanced analyses will lose their accuracy prerequisite.
[0038] Step S2: Extract features from the preprocessed soil remote sensing image to obtain soil spectral features, soil texture features and soil exposure features, and fuse them to generate a water erosion sensitivity layer.
[0039] This step is the core feature enhancement stage of the soil water erosion feature extraction method. Its purpose is to quantify and fuse the spectral, texture, and land cover information related to soil water erosion in the preprocessed remote sensing image, thereby generating a composite layer highly sensitive to water erosion features. This layer will serve as the main basis for subsequent image segmentation algorithms, significantly improving the separability of water erosion areas from the background environment. Specifically, Feature extraction and fusion are then performed. All feature calculations are directly based on this preprocessed image data. Radiometric correction in step S1 aims to provide a foundation of good radiometric consistency for this step, ensuring the comparability of features from different regions; its execution flow includes three core steps: feature extraction, normalization, and fusion, ultimately outputting a water erosion sensitivity layer. .
[0040] First, multi-dimensional feature extraction is performed from the preprocessed soil remote sensing images. Three physical characteristic components that are indicative of water erosion phenomena were identified in the analysis: soil spectral characteristics, soil texture characteristics, and soil exposure characteristics. For soil spectral characteristics Extraction reveals that water erosion leads to vegetation degradation and soil exposure, which in turn causes changes in surface reflectance. This change is quantified by calculating the Normalized Difference Vegetation Index (NDVI) and Soil Brightness Index. The NDVI is calculated as follows:
[0041] NIR represents the near-infrared reflectance value of the image; Red represents the red reflectance value of the image. This index effectively quantifies vegetation cover by combining the reflectance values of the near-infrared and red bands. Water erosion areas typically exhibit decreased vegetation cover, hence their lower NDVI values.
[0042] Soil brightness index (BI) is calculated as follows:
[0043] Here, Red represents the reflectance value in the red band of the image, and Green represents the reflectance value in the green band. This index comprehensively reflects the overall reflectance of the soil in the visible light band. Increased soil exposure or changes in moisture often lead to an increase in the BI value. Soil spectral characteristics The vegetation index and soil brightness index are linear combinations to comprehensively characterize spectral anomalies.
[0044] This combination makes High values were obtained in water areas with degraded vegetation and high soil brightness, thereby enhancing the ability to distinguish soil spectral features.
[0045] For soil texture features Extraction reveals that surface roughening and gully fragmentation caused by water erosion lead to significant changes in image texture. The gray-level co-occurrence matrix (GLCM) is used to calculate the entropy value to measure the complexity and irregularity of the texture, thus obtaining soil texture features. The calculation is as follows:
[0046] in, The number of gray levels; It is the first normalized GLCM. line, number The probability value of the column indicates that a pair of pixels has different gray levels. and The probability of entropy. The higher the entropy value, the more complex and disordered the texture. Water-eroded areas, due to their fragmented and rough characteristics, typically have high entropy values.
[0047] Characteristics of soil exposure This directly quantifies the degree of soil exposure. The Normalized Difference Built-up Area Index (NDBI) is used to obtain soil exposure characteristics. The calculation is as follows:
[0048] in, This index represents the shortwave infrared reflectance value of an image. It effectively highlights exposed surfaces by utilizing the high reflectance of bare soil in the shortwave infrared band and vegetation in the near-infrared band. Water-eroded areas typically have higher NDBI values.
[0049] After obtaining the above-mentioned soil spectral characteristics, soil texture characteristics, and soil exposure characteristics, feature normalization processing is required. This is because... The three features have different dimensions and numerical ranges. Direct fusion may result in some features becoming dominant due to differences in magnitude. Therefore, the feature components are normalized to... For example:
[0050] in, It is a feature The mean value across the entire graph; It is a feature The standard deviation across the entire map. This operation (Z-score standardization) transforms each feature into a distribution with a mean of 0 and a standard deviation of 1, eliminating the influence of dimensions and laying the foundation for subsequent weighted fusion. For soil texture features... and nudity characteristics Perform the same normalization operation to obtain the normalized soil texture features. and nudity characteristics .
[0051] Finally, the normalized features are weighted and fused pixel by pixel to generate a unified water erosion sensitivity layer. For any pixel location in the image The fusion calculation is as follows:
[0052] in, The normalized soil spectral characteristics are shown below. Soil texture characteristics and nudity characteristics The weight coefficients, and satisfying The weighting coefficients are adjusted based on prior knowledge of water erosion characteristics in different regions. Several known and typical severely water-eroded and non-water-eroded areas are selected on the image, and the differences in the performance of each feature component in these areas are visually compared to determine the weighting coefficients. The formula integrates information from the three dimensions of spectrum, texture, and exposure at each pixel through weighted summation, ultimately generating a water erosion sensitivity layer. It is a raster layer with the same spatial resolution as the original image, where each pixel value... All of these studies comprehensively quantified the sensitivity of the surface where the pixel is located to water erosion.
[0053] The water erosion sensitivity layer generated in this step serves as the direct data foundation and core basis for all subsequent advanced processing operations in the entire scheme. Through multi-source information fusion, it aggregates and enhances the weak and scattered traces of water erosion in the image into a high-contrast, highly separable single-band feature image. This significantly improves the difference between the water-eroded area and the background features in the feature space, providing accurate and reliable input for the intelligent image segmentation in step S4, and is a key prerequisite for the entire method to achieve high-precision extraction.
[0054] Step S3: Collect digital elevation model data of the target area, calculate the terrain factor based on the digital elevation model data, divide the target area into different terrain functional units according to the terrain factor, and form a hierarchical segmentation rule base based on the different terrain functional units and preset differentiated decision thresholds.
[0055] This step is a crucial one in introducing geographical mechanism constraints into the soil water erosion feature extraction method. Its core purpose is to transform digital elevation model data (DEM) into differentiated strategies that can guide image segmentation, enabling the segmentation process to conform to the natural development patterns of water erosion phenomena in different terrain locations, thereby significantly improving the rationality and accuracy of the segmentation results. Following the data preparation completed in step S1, this step will independently process the terrain data, establishing a rule-based foundation for subsequent intelligent segmentation.
[0056] Specifically, this step begins with the collection and preprocessing of digital elevation model (DEM) data. The collected DEM data covers the target area and is denoted as... The data source can be an open database such as spaceborne radar (e.g., ALOSPALSAR) or the Space Shuttle Radar Topography Mission (SRTM), and its spatial resolution should be better than or equal to the resolution of the soil remote sensing image in step S1 to ensure spatial consistency between the terrain analysis and image features; subsequently, the digital elevation model data... Preprocessing steps such as filling data gaps and smoothing noise are performed to obtain reliable preprocessed digital elevation model data, denoted as . This serves as the sole input for all analyses in this step.
[0057] Next, based on the preprocessed digital elevation model data Perform terrain factor calculations to provide quantitative indicators for unit division, such as slope. The calculation is as follows:
[0058] in, and These represent the rates of elevation change of the DEM in the east-west and north-south directions, respectively; arctan( ) is the arctangent function; These are partial differential symbols; x and y represent preprocessed digital elevation model data, respectively. The east-west and north-south coordinate axes; slope It is the core factor controlling surface runoff energy and erosion potential, and the primary basis for dividing geomorphic units.
[0059] Subsequently, based on geomorphological and water erosion principles, topographic functional units were divided according to the calculated topographic factors. The target area was divided into three basic topographic functional units, and the set of topographic functional units is as follows: These represent the top, middle, and bottom of the slope, respectively. The following is the principle behind this division: For the hilltop area These areas are typically the origin of water erosion, primarily manifesting as surface erosion and raindrop splash erosion. They are defined as regions with gentle slopes and forming topographic watersheds, for example: ; For the middle area of the slope This area is a runoff convergence and energy concentration zone, and is the main development area for linear erosion such as gullies and creeks. Its delineation rule is based on areas with steeper slopes, for example: ; For the foot of the slope area This is a sedimentary deposition area, mainly characterized by sedimentary and accumulation patterns. Its classification is based on gently sloping catchment areas, for example: And it is located at the end of the water flow confluence; Finally, based on the water erosion development mechanism of each topographic functional unit, a key differentiated decision threshold is set for each unit. The threshold values of all units and their corresponding relationships constitute a hierarchical segmentation rule base. The core mapping relationship of this rule base is simplified to: That is, for any unit The threshold is the criterion for determining whether a pixel is classified as a water erosion zone. The rule base is built based on the following core principles: In the slope crest area where erosion signals are weak To prevent missed detections, a relatively low decision threshold is set. (For example, the lower quantile range corresponding to the sensitivity layer value range), to make it more sensitive to small surface erosion traces; in the mid-slope area dominated by linear erosion. To accurately extract coherent channels, a moderate decision threshold is set. It can effectively highlight linear features while suppressing noise; in slope toe areas with significant sedimentation and blurred boundaries. To avoid misclassifying scattered deposits into multiple independent patches, a relatively high decision threshold is set. (For example, the high quantile range of the corresponding sensitivity layer value range), and combined with the region merging logic, to depict the overall stacking pattern and ensure the geographical rationality of the segmentation results; This step is crucial for introducing constraints based on geographical mechanisms. Its core task is to calculate basic topographic factors such as slope and aspect based on Digital Elevation Model (DEM) data, and, according to the principles of water erosion development in geomorphology, divide the target area into topographic functional units with different erosion characteristics, such as slope top, slope middle, and slope toe. For each unit's water erosion behavior (e.g., surface erosion at the slope top, gully erosion in the slope middle), a key differentiated decision threshold is set. The thresholds of all units and their corresponding relationships constitute a hierarchical segmentation rule base.
[0060] Step S4: Based on the hierarchical segmentation rule base, construct a terrain segmentation mechanism, and segment the water erosion sensitivity layer through the terrain segmentation mechanism to obtain a water erosion area segmentation map; This step is the core decision-making step in the soil water erosion feature extraction method. Its core purpose is to instantiate the hierarchical segmentation rule base established in step S3 into an executable terrain segmentation mechanism, and apply this mechanism to intelligently segment the water erosion sensitivity layer obtained in step S2, and output a water erosion area segmentation map through clear decision criteria.
[0061] Specifically, the input for this step includes: a water erosion sensitivity layer from step S2, denoted as... , where each pixel value Characterizes the continuous probability of water erosion occurring at this location; derived from the hierarchical segmentation rule base of step S3, denoted as... Different terrain functional units are defined therein. (like The corresponding optimal decision threshold .
[0062] A terrain segmentation mechanism, M, is constructed to transform geographical knowledge into image segmentation decisions. Its core is establishing a mapping relationship from spatial location to dynamic decision thresholds. For any pixel in the image... The mechanism operates as follows: First, perform terrain unit localization, based on pixels. The geographic coordinates are used to determine the topographic functional unit to which it belongs. .
[0063] Query the rule base again Obtain the terrain unit Preset decision threshold and assign it to the pixel. This process can be formalized using the following mapping function:
[0064] in, It is a pixel The minimum sensitivity threshold required to be classified as a water erosion area is determined by the topographic unit in which it is located. It is a query function that takes a terrain unit type as input and then returns the optimal empirical threshold preset for that terrain type from a hierarchical segmentation rule base. The terrain segmentation mechanism M implements spatial differentiation of the segmentation criteria. It ensures that segmentation is performed in areas with weak erosion (such as the top of the slope). Use a lower threshold To improve detection sensitivity, in the slope foot area where spurious changes are prone to occur ( Use a higher threshold This ensures the specificity of the results, thereby making the segmentation behavior consistent with natural laws.
[0065] The water erosion sensitivity layer is segmented using a terrain segmentation mechanism to obtain the dynamic decision threshold for each pixel. Then, the water erosion sensitivity layer Perform segmentation decision. The decision rule is: if the sensitivity value of a pixel is not lower than the threshold required by the terrain unit where it is located, then it is determined to be a water-eroded area. The segmentation decision formula is as follows:
[0066] in, It is a water erosion area segmentation map in pixels The value at the location; 1 represents a water erosion area, and 0 represents a non-water erosion area. This formula provides a clear definition of the water erosion area and an operational evaluation criterion. It integrates continuous water erosion sensitivity information. Based on a reasonable critical criterion that varies with topographical space. Binarization is performed to generate a preliminary water erosion area segmentation map. .
[0067] The final output of this step is a water erosion area segmentation map. This is the first explicit spatial representation of soil water erosion characteristics determined by the mechanism-driven standard. As the direct input to step S5, although it already possesses basic spatial scope and geographical rationality, its boundaries still have problems such as jaggedness and fractures, which require further refinement and optimization.
[0068] Step S5: Construct a water erosion contour extraction model based on the energy minimization theory. Input the water erosion region segmentation map into the water erosion contour extraction model. The model optimizes the contour based on neighborhood texture and spectral continuity and outputs soil water erosion characteristics.
[0069] This step is the final optimization stage of the soil water erosion feature extraction process. Its core purpose is to refine the boundaries of the preliminary segmentation results output in step S4 using a water erosion contour extraction model based on energy minimization theory. Based on energy minimization theory, this model employs the Narrow Band Level Set Method for efficient and stable numerical solutions in its iterations. This aims to address the jagged edges, breaks, and positioning errors in the initial boundaries, ultimately generating high-precision vector boundaries that conform to the geomorphological structure characteristics.
[0070] Specifically, the input for this step is the water erosion area segmentation map output from step S4, denoted as... This step will construct and solve an energy minimization model, the output of which is the refined soil water erosion characteristics. .
[0071] The model for extracting water erosion profiles is constructed. This model uses the level set method as a framework to transform the boundary evolution problem into an energy functional problem. The minimization problem of the level set function. It is an implicitly expressed scalar function with a zero level set. The outline of the water-eroded region is shown. The total energy functional is as follows:
[0072] in, For image data items, it drives the contour to fit the gradient edges in the image; For structural constraints, prior knowledge of the texture and spectral continuity of the water erosion region is introduced; As a regularization term, it constrains the smoothness of the contour and avoids unreasonable bends; The weighting coefficients are used to balance the contributions of each element. The optimal selection for grid search is determined, and before applying the model, a small, representative region (containing clear water erosion boundaries) is selected as the validation area. Within a preset range (e.g....), A grid search is performed, and the F1 score between the contour extraction results and the manually annotated results is used as the evaluation metric. The set that optimizes the metric is selected. It is applied as a global parameter throughout the entire study area.
[0073] For image data items This drives the contour towards the original image. The gradient edge evolution is mathematically represented by the following formula:
[0074] in, For the image domain; This is the weighting coefficient for that item (usually a preset positive constant); For edge detection functions, in the image gradient Larger areas (edges) have smaller values, thus minimizing the energy at the edges; This is the Dirac Delta function, which restricts the line integral to the vicinity of the zero level set.
[0075] For structural constraint terms This utilizes the texture features extracted in step S2. Based on spectral characteristics, prior knowledge of water-eroded areas (fragmented texture, roughness) is incorporated into the model, and the mathematical formula is as follows:
[0076] in, The structure saliency map is composed of normalized texture features. After smoothing or enhancement processing, the result is ( , (This is the Sigmoid function), and the closer its value is to 1, the more likely the point is to be located within a water erosion area; For the Heaviside function. This penalty applies to the interior of the contour ( ) contains non-significant structures ( The region guides the contour to shrink towards the potential water erosion area with significant texture features.
[0077] For regularization terms This term is used to control the contour length, ensure its smoothness, and suppress the distortion caused by noise. The mathematical formula is as follows:
[0078] in, This is the length penalty weight (preset as a positive constant). Minimizing this term is equivalent to minimizing the length of the contour curve.
[0079] Model optimization employs a narrow-band level set method for iterative solution to efficiently solve the aforementioned energy functional. The core advantage of this method lies in its extremely high computational efficiency, as it narrows the computational domain from the entire image plane to a narrow band-shaped region near the active contour. The narrow-band iterative solution process is as follows: First, initialize the binary segmentation image. Transform it into a signed distance function, and use it as a level set function. The initial value is determined, and a structure is constructed centered at the zero level set with a width of [value missing]. (like The initial narrow band (pixels); In iterative evolution, that is, in the th... In this iteration, the following steps are performed: First, calculate the evolution velocity. For all pixels within the current narrow band, calculate the evolution velocity field of the level set function. The velocity field is determined by the gradient descent flow of the energy functional, which includes driving forces from image gradients, texture structure constraints, and curvature smoothing. Subsequently, the level set function is updated using an explicit Euler scheme, updating the level set function only within the narrow band:
[0080] in, It is a level set function; superscript It is an iteration count index; Indicates after the first The boundary state after the next iteration of optimization; For time step.
[0081] Then perform narrowband updates and reinitialization, checking if the zero-level set has moved close to the boundary of the current narrowband. If so, reconstruct the narrowband centered on the new zero-level set, and then process the data within the new narrowband. Perform a fast reinitialization to keep it as an accurate signed distance function to ensure numerical stability.
[0082] Repeat the above process iteratively until the change in the position of the zero level set is less than a predetermined threshold. or reaching the maximum number of iterations. At this point, the water erosion profile extraction model is considered to have converged, yielding the final level set function. .
[0083] After the final soil water erosion feature extraction model converges, the final level set function is used. Extract its zero level set :
[0084] in, It is the domain; An element in a set is represented by a point. The set of all points forms a curve; this zero-level set This represents the optimized water erosion area outline. The outline is then converted into a vector polygon and topologically checked to obtain the final output of this method—soil water erosion features.
[0085] The output of this step This is the optimized, high-precision vector boundary and spatial distribution map of water erosion areas. It accurately depicts the spatial location, extent, and geometric outline of soil water erosion areas. Its boundaries are smooth and continuous, and highly consistent with the texture and soil spectral characteristics in the image. It can be directly used in Geographic Information System (GIS) software for quantitative analysis and mapping, completing the entire extraction process from remote sensing imagery to calculable water erosion features.
[0086] This technical solution proposes a method for extracting soil water erosion features based on image segmentation, which is particularly suitable for the automated monitoring and assessment of soil water erosion on sloping farmland in the Loess Plateau. This region, characterized by fragmented terrain and well-developed erosion gullies, is a challenging area for achieving precise soil and water conservation. The core objective is to extract the spatial location, extent, and vector contour of soil water erosion areas from soil remote sensing images with high precision and efficiency, using these as soil water erosion features to address the shortcomings of traditional methods in terms of accuracy, geographical applicability, and practical efficiency.
[0087] This method first performs rigorous radiometric calibration, atmospheric correction, and geometric fine correction on the acquired high-resolution soil remote sensing images, producing high-quality preprocessed images that lay an accurate physical and geometric foundation for subsequent analysis. From the preprocessed images, spectral features, texture features, and exposure features are quantitatively extracted. A comprehensive water erosion sensitivity layer is generated through pixel-by-pixel weighted fusion, which greatly enhances the separability of water erosion information and serves as the core basis for subsequent intelligent segmentation. Based on digital elevation model data, topographic factors are calculated, and the target area is divided into topographic functional units such as slope top, slope middle, and slope foot according to geomorphological principles. Differentiated decision thresholds are set for each unit based on its water erosion development patterns, forming a hierarchical segmentation rule base. This step transforms geographical knowledge into computable rules, a key innovation ensuring that the segmentation results conform to natural laws. A topographic segmentation mechanism is constructed, combining the water erosion sensitivity layer with the hierarchical rule base. This mechanism dynamically calls the corresponding decision threshold for segmentation based on the topographic functional unit where each pixel is located, generating a water erosion area segmentation map. This step achieves the first geographically plausible intelligent identification from continuous features to spatial regions. To address the jagged and fractured edges in the initial segmentation boundary, a contour extraction model based on energy minimization theory is constructed. Starting with the initial segmentation image, this model integrates an energy function constrained by image gradient, local structural consistency, and contour smoothness. It then employs a narrow-band level set method for efficient iterative solving, ultimately driving the boundary to converge to the true position with sub-pixel precision. This results in a smooth, continuous, and accurate vector boundary for the water erosion region, representing the final soil water erosion characteristics.
[0088] 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 statement "including a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0089] 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 likenesses.
Claims
1. A method for extracting soil water erosion features based on image segmentation, characterized in that, Includes the following steps: Step S1: Acquire the original soil remote sensing image of the target area, and preprocess the original soil remote sensing image to obtain the preprocessed soil remote sensing image. Step S2: Extract features from the preprocessed soil remote sensing image to obtain soil spectral features, soil texture features and soil exposure features, and fuse them to generate a water erosion sensitivity layer; Step S3: Collect digital elevation model data of the target area, calculate the terrain factor based on the digital elevation model data, divide the target area into different terrain functional units according to the terrain factor, and form a hierarchical segmentation rule base based on the different terrain functional units and preset differentiated decision thresholds. Step S4: Based on the hierarchical segmentation rule base, construct a terrain segmentation mechanism, and segment the water erosion sensitivity layer through the terrain segmentation mechanism to obtain a water erosion area segmentation map; Step S5: Construct a water erosion contour extraction model based on the energy minimization theory, input the water erosion region segmentation map into the water erosion contour extraction model, and output the soil water erosion features.
2. The method for extracting soil water erosion features based on image segmentation according to claim 1, characterized in that, Acquire raw soil remote sensing images of the target area, and preprocess the raw soil remote sensing images to obtain preprocessed soil remote sensing images, including: Data collection was conducted, and a satellite imaging sensor was selected to acquire high-resolution soil remote sensing images of the target area at a single time phase, resulting in the raw soil remote sensing images. ; Remote sensing images of the original soil A standardized preprocessing process is implemented; the first step in preprocessing is the radiation correction step, which converts the raw digital quantization values into surface reflectance using an atmospheric correction model. Geometric fine correction is performed using a coordinate transformation function model: ； in, The coordinates of the original soil remote sensing image; These are the corrected geographic coordinates. It is a rational function model; Represents the set of ground control points; After the above preprocessing steps, the original soil remote sensing image The soil remote sensing image is converted into a preprocessed soil remote sensing image. .
3. The method for extracting soil water erosion features based on image segmentation according to claim 2, characterized in that, Feature extraction is performed on the preprocessed soil remote sensing image to obtain soil spectral features, soil texture features, and soil bareness features, which are then fused to generate a water erosion sensitivity layer, including: Multi-dimensional feature extraction is performed on preprocessed soil remote sensing images. Three physical characteristic components that are indicative of water erosion phenomena were identified in the analysis: soil spectral characteristics, soil texture characteristics, and soil exposure characteristics. Soil spectral characteristics It is a linear combination of vegetation index and soil brightness index to comprehensively characterize spectral anomalies; The entropy value is calculated using the gray-level co-occurrence matrix to measure the complexity and irregularity of the texture, thus obtaining soil texture features. ; The characteristics of soil exposure were obtained using the Normalized Differential Exposure Index (NDBI). ; After obtaining the aforementioned soil spectral characteristics, soil texture characteristics, and soil exposure characteristics, feature normalization processing is required. The normalized features are then weighted and fused pixel by pixel to generate a unified water erosion sensitivity layer. .
4. The method for extracting soil water erosion features based on image segmentation according to claim 3, characterized in that, Collect digital elevation model data of the target area, and calculate terrain factors based on the digital elevation model data, including: Collect digital elevation model data covering the target area, denoted as The data source originates from publicly available databases of satellite-borne radar or space shuttle radar topographic mapping missions; digital elevation model data... Preprocessing is performed to fill data gaps and smooth noise, resulting in reliable preprocessed digital elevation model data, denoted as . ; Based on preprocessed digital elevation model data Perform terrain factor calculations to provide quantitative indicators for unit division, such as slope. The calculation is as follows: ； in, and These represent the rates of elevation change of the DEM in the east-west and north-south directions, respectively; arctan( ) is the arctangent function; These are partial differential symbols; x and y represent preprocessed digital elevation model data, respectively. The east-west and north-south coordinate axes; slope It is a core factor controlling surface runoff energy and erosion potential.
5. The method for extracting soil water erosion features based on image segmentation according to claim 4, characterized in that, The target area is divided into different terrain functional units based on the terrain factors, including: Based on geomorphological and water erosion principles, and using calculated topographic factors, the target area is divided into three basic topographic functional units, which are then set as follows: These represent the top, middle, and bottom areas of the slope, respectively.
6. The method for extracting soil water erosion features based on image segmentation according to claim 5, characterized in that, Based on different terrain functional units, preset differentiated decision thresholds are used to form a hierarchical segmentation rule base, including: Set a key differentiated decision threshold for each unit. The thresholds of all units and their corresponding relationships constitute a hierarchical segmentation rule base. The core mapping relationship of this rule base is simplified to: That is, for any unit The threshold is the criterion for determining whether a pixel is classified as a water erosion zone. The rule base is built based on the following core principles: In the slope crest area where erosion signals are weak Set a relatively low decision threshold ; Mid-slope area dominated by linear erosion Set a moderate decision threshold In the foothills of slopes where sedimentation is significant and boundaries are indistinct. Set a relatively high decision threshold .
7. The method for extracting soil water erosion features based on image segmentation according to claim 6, characterized in that, Based on a hierarchical segmentation rule base, a terrain segmentation mechanism is constructed, including: A terrain segmentation mechanism, M, is constructed to transform geographical knowledge into image segmentation decisions. Its core is establishing a mapping relationship from spatial location to a dynamic decision threshold for any pixel in the image. The mechanism operates as follows: First, perform terrain unit localization, based on pixels. The geographic coordinates are used to determine the topographic functional unit to which it belongs. ; Then query the rule base Obtain the terrain unit Preset decision threshold and assign it to the pixel. This process can be formalized using the following mapping function: ； in, It is a pixel The minimum sensitivity threshold required to be classified as a water erosion area is determined by the topographic unit in which it is located. It is a query function that takes a terrain unit type as input and then returns the best empirical threshold preset for that type of terrain from a hierarchical segmentation rule base.
8. The method for extracting soil water erosion features based on image segmentation according to claim 7, characterized in that, The water erosion sensitivity layer is segmented using the aforementioned terrain segmentation mechanism to obtain a water erosion area segmentation map, including: The water erosion sensitivity layer is segmented using a terrain segmentation mechanism to obtain the dynamic decision threshold for each pixel. Then, the water erosion sensitivity layer The segmentation decision is performed according to the following rule: if the sensitivity value of a pixel is not lower than the threshold required by the terrain unit where it is located, then it is determined to be a water-eroded area. The segmentation decision formula is as follows: ； in, It is a water erosion area segmentation map in pixels The value at the location; 1 represents a water erosion area, 0 represents a non-water erosion area; continuously displaying water erosion sensitivity information. Based on a reasonable critical criterion that varies with topographical space. Binarization is performed to generate a preliminary water erosion area segmentation map. .
9. A method for extracting soil water erosion features based on image segmentation according to claim 8, characterized in that, A water erosion profile extraction model is constructed based on the energy minimization theory, including: The model for extracting water erosion profiles is constructed. This model uses the level set method as a framework to transform the boundary evolution problem into an energy functional problem. The minimization problem of the level set function It is an implicitly expressed scalar function with a zero level set. The outline of the water-eroded region is shown; the total energy functional is as follows: ； in, For image data items, it drives the contour to fit the gradient edges in the image; For structural constraints, prior knowledge of the texture and spectral continuity of the water erosion region is introduced; As a regularization term, it constrains the smoothness of the contour and avoids unreasonable bends; These are weighting coefficients used to balance the contributions of each item; Model optimization employs a narrow-band level set method for iterative solution to efficiently solve the aforementioned energy functional; after the soil water erosion feature extraction model converges, the final level set function is used. Extract its zero level set : ； in, It is the domain; An element in a set is represented by a point. This refers to the set of all points that form a curve.
10. A method for extracting soil water erosion features based on image segmentation according to claim 9, characterized in that, The narrowband level set method includes: Initialize the binary segmentation image. Transform it into a signed distance function, and use it as a level set function. The initial value is determined, and a structure is constructed centered at the zero level set with a width of [value missing]. The initial narrow band; In iterative evolution, that is, in the th... In this iteration, the following steps are performed: First, calculate the evolution velocity. For all pixels within the current narrow band, calculate the evolution velocity field of the level set function. The velocity field is determined by the gradient descent flow of the energy functional, which includes driving forces from image gradient, texture structure constraints and curvature smoothing. Using the explicit Euler scheme, update the level set function only within a narrow band: ； in, It is a level set function; superscript It is an iteration count index; Indicates after the first The boundary state after the next iteration of optimization; For time step; Then perform narrowband updates and reinitialization, checking if the zero-level set has moved close to the boundary of the current narrowband; if so, reconstruct the narrowband with the new zero-level set as the center, and process the new narrowband... Perform a quick reinitialization; Repeat the above process iteratively until the change in the position of the zero level set is less than a predetermined threshold. or reaching the maximum number of iterations. At this point, the water erosion profile extraction model is considered to have converged, yielding the final level set function. .