A dynamic assessment system and method for the ecological health of riparian vegetation

Abstract

This invention discloses a dynamic assessment system and method for the ecological health of riparian vegetation, relating to the field of riparian vegetation ecological health technology. Specifically, it includes the following steps: determining whether each image cell grid has shrub colors that are close to the background color based on color and texture features, and marking such image cell grids as areas to be identified; extracting boundary recognition feature information for each area to be identified, and analyzing the extracted features to assess the degree of boundary ambiguity when shrub colors are close to the background color, and classifying them according to the assessment results; based on the classification results, selecting corresponding shrub spatial distribution area extraction methods for different categories of areas to be identified, in order to extract the shrub spatial distribution areas in each area to be identified. This invention solves the problem of difficult identification of shrub boundaries due to ambiguity, and achieves accurate extraction of shrub spatial distribution and accurate assessment of ecological status against complex backgrounds.

Description

Technical Field

[0001] This invention relates to the field of riparian vegetation ecological health technology, specifically to a dynamic assessment system and method for riparian vegetation ecological health. Background Technology

[0002] Riverbank vegetation ecological health refers to the overall ecological state of the vegetation community in a riverbank area in terms of structural integrity, functional stability, and ecosystem service capacity. It encompasses not only static indicators such as vegetation species composition, biodiversity, and cover, but also its responsiveness to environmental disturbances, self-repair capabilities, and long-term support for ecological functions such as soil and water conservation, microclimate regulation, and habitat maintenance. As riverbank areas are sensitive zones where land and water meet, their vegetation ecological status directly impacts flood control, pollution control, water purification, and regional ecological security. However, under the combined effects of natural changes (such as climate change and hydrological fluctuations) and human disturbances (such as engineering construction and agricultural development), the ecological health of riverbank vegetation exhibits a high degree of dynamism and vulnerability. Therefore, static monitoring at a single point in time is insufficient to comprehensively reflect its ecological evolution process and potential risks. Thus, there is an urgent need to construct a multi-time-period, multi-dimensional, and quantifiable dynamic assessment method. Through continuous data collection and indicator analysis, this method can accurately grasp the health trends and fluctuation characteristics of the vegetation system, thereby providing a scientific and forward-looking decision-making basis for ecological protection and restoration management.

[0003] Existing dynamic assessment technologies for riparian vegetation ecological health typically rely on multi-source data acquisition methods such as remote sensing imagery, drone aerial photography, or ground sensors to periodically or continuously collect vegetation information of riparian areas. These data are then quantified using image recognition, spectral analysis, and Geographic Information System (GIS) technologies to measure indicators such as vegetation cover, greenness index (e.g., NDVI), species diversity, and biomass changes. After data acquisition, the assessment system generally includes four main stages: data preprocessing (e.g., image denoising, registration, and correction), indicator extraction (selecting appropriate ecological health indicators based on the ecological model), time-series analysis (comparing data change trends at different time points), and comprehensive evaluation. The comprehensive evaluation often involves setting health level classification standards and combining expert knowledge or multi-indicator weighted methods to score or classify the ecological status of the current and historical periods. Furthermore, some advanced technologies have introduced machine learning algorithms to train models to identify abnormal changes or predict future ecological trends, thereby improving the automation and intelligence of the assessment.

[0004] The existing technology has the following shortcomings:

[0005] In areas with complex topography at the foot of riverbanks, shrubs are often interspersed with herbs and trees. Shrubs typically have low canopies, dense foliage, and limited color variation. Especially near bare sand or rocky backgrounds, the color of shrubs in images closely matches the surrounding background, resulting in indistinct edges. Due to the lack of significant color transitions and structural contrasts in the images, the boundaries of shrub areas are difficult to accurately segment. Current dynamic assessment techniques for the ecological health of riverbank vegetation typically rely on edge detection and color segmentation methods in image analysis. These methods are highly dependent on boundary sharpness and color difference, and cannot adapt to the slow transition between shrub boundaries and the background in images. Therefore, they cannot accurately extract the spatial distribution area of ​​shrub vegetation based on the degree of boundary blurring when shrub colors are similar to the background colors. This identification bias leads to a continuous underestimation of shrub cover area, resulting in inaccurate extraction of subsequent structural integrity indicators. Consequently, it incorrectly reflects the ecological health status of the area and may ultimately lead to unnecessary or even destructive interventions in shrub-dominated areas that should remain stable, seriously interfering with the scientific and precise nature of ecological governance.

[0006] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a dynamic assessment system and method for the ecological health of riverbank vegetation to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for dynamic assessment of the ecological health of riparian vegetation, specifically comprising the following steps:

[0009] Image data containing the riverbank area is acquired, and the image data is preprocessed, including image denoising, registration and multi-channel alignment. The riverbank area image is extracted in combination with the terrain information and divided into several image unit grids.

[0010] Based on color and texture features, determine whether there are shrub colors that are similar to the background color in each image cell grid, and mark the image cell grid with such a situation as the region to be identified;

[0011] Extract the boundary recognition feature information of each region to be identified, and analyze it after extraction. Evaluate the degree of boundary recognition ambiguity of each region to be identified when the shrub color is close to the background color, and classify them according to the evaluation results.

[0012] Based on the classification results, select the corresponding shrub vegetation spatial distribution area extraction method for different categories of areas to be identified, so as to extract the shrub spatial distribution area in each area to be identified;

[0013] The image range of the extracted shrub spatial distribution area is statistically analyzed, and the ecological health status of the riverbank area is displayed based on the statistical results.

[0014] Preferably, based on color and texture features, it is determined whether each image unit grid has a shrub color that is close to the background color, and the image unit grid with such a situation is marked as a region to be identified, including:

[0015] Analyze the color distribution differences between shrub vegetation areas and adjacent non-vegetation areas in each image cell grid;

[0016] Analyze the differences in texture arrangement direction and density between shrub vegetation areas and adjacent non-vegetation areas in each image unit grid;

[0017] When the color distribution difference between the shrub vegetation area and the adjacent non-vegetation area is lower than the color judgment standard set based on the pixel color value clustering difference threshold, and the distribution difference of the texture arrangement direction and density between the shrub vegetation area and the adjacent non-vegetation area is lower than the texture judgment standard set based on the texture direction histogram overlap ratio, it is determined that the image unit grid has a situation where the shrub color and the background color are similar.

[0018] Image cell grids where the shrub color is similar to the background color are marked as regions to be identified.

[0019] Preferably, boundary recognition feature information of each region to be identified is extracted and analyzed after extraction. The degree of boundary recognition ambiguity of each region to be identified is evaluated when the shrub color is close to the background color, and the regions are classified according to the evaluation results. Specifically:

[0020] Extract the boundary recognition feature information of each region to be identified, and perform preprocessing operations after extraction;

[0021] Boundary structure contrast information and image texture color fusion information are extracted from the preprocessed boundary recognition feature information, and normalization is performed after extraction;

[0022] Based on the normalized boundary structure contrast information and image texture color fusion information, the boundary transition perception coefficient and boundary recognition obstacle index of each region to be identified are generated respectively.

[0023] Based on the generated boundary transition perception coefficient and boundary recognition obstacle index of each region to be identified, a boundary recognition fuzziness index of each region to be identified is generated.

[0024] A pre-defined threshold range for boundary recognition fuzziness index is determined and compared with the generated boundary recognition fuzziness index of each region to be recognized. The degree of boundary recognition fuzziness of each region to be recognized is evaluated based on the comparison results when the shrub color is close to the background color, and the regions are classified based on the evaluation results.

[0025] Preferably, the logic for obtaining the boundary transition sensing coefficients of each region to be identified is as follows:

[0026] Boundary structure comparison information is extracted from the preprocessed boundary recognition feature information, specifically including three types of data: the average gray-level gradient magnitude, the gray-level transition length, and the standard deviation of the gradient direction of each region to be identified. These three types of data are divided by their respective preset maximum values ​​to obtain the normalized average gray-level gradient magnitude, gray-level transition length, and standard deviation of the gradient direction of each region to be identified, which are then labeled as follows: , and , The normalized first The average gray-level gradient magnitude of the edges of the regions to be identified. The normalized first The grayscale transition length at the edge of the region to be identified The normalized first The standard deviation of the gradient direction at the edge of each region to be identified. , It is a positive integer;

[0027] Calculate the boundary transition sensing coefficient of each region to be identified. The specific calculation method is as follows: the gray-scale transition length of the edges of each normalized region to be identified is... The average gray-level gradient magnitude of the edges of each region to be identified is raised to the power of the normalized value. The natural logarithm of the ratio obtained by adding one and the normalized standard deviation of the gradient direction of each region to be identified. The square roots are summed to obtain the boundary transition sensing coefficients of each region to be identified. .

[0028] Preferably, the logic for obtaining the boundary recognition obstacle index of each region to be identified is as follows:

[0029] Image texture and color fusion information is extracted from the preprocessed boundary recognition feature information. Specifically, this includes three types of data: the mean Euclidean distance of each region to be recognized, the texture orientation histogram cross ratio, and the pixel color variance ratio. These three types of data are divided by their respective preset maximum values ​​to obtain the normalized mean Euclidean distance of each region to be recognized, the texture orientation histogram cross ratio, and the pixel color variance ratio, which are then labeled as follows: , and , The normalized first The Euclidean distance between the color mean of each region to be identified The normalized first The cross ratio of the texture orientation histogram of each region to be identified. The normalized first The pixel color variance ratio of the region to be identified , It is a positive integer;

[0030] Calculate the boundary recognition obstacle index for each region to be identified. The specific calculation method is as follows: the cross ratio of the texture orientation histograms of each normalized region to be identified is calculated. The first term is obtained by squaring the pixel color variance ratio of each normalized region to be identified. Taking its square root yields the second term, which is the normalized color mean Euclidean distance of each region to be identified. Add one and take the reciprocal, then add another one and take the natural logarithm as the third term. Add the first, second, and third terms to obtain the boundary recognition obstacle index for each region to be identified. .

[0031] Preferably, based on the boundary transition sensing coefficients of each generated region to be identified. and boundary recognition barrier index The boundary recognition fuzziness index of each region to be identified is generated by weighted summation, specifically by determining the boundary transition perception coefficient of each region to be identified. and boundary recognition barrier index Each region has a pre-set first non-zero weight coefficient and a second non-zero weight coefficient, the sum of which is one; the first non-zero weight coefficient is multiplied by the boundary transition sensing coefficient of each region to be identified. The result is multiplied by the second non-zero weighting coefficient and the boundary recognition obstacle index of each region to be identified. The results are summed to obtain the boundary recognition fuzziness index of each region to be identified. .

[0032] Preferably, a pre-defined threshold range for boundary recognition fuzziness index is determined. And after determination, it identifies the fuzzy index of the boundaries of each region to be identified. A comparison was performed, and the degree of ambiguity in the boundary recognition of each region to be identified was evaluated based on the comparison results when the shrub color and the background color were similar. The regions were then classified according to the evaluation results. The specific comparison analysis and classification are as follows:

[0033] like If the shrub color is close to the background color, the boundary recognition blur level of the area to be identified is low blur level, and then the area to be identified is classified as the first category area;

[0034] like If the shrub color is similar to the background color, the boundary recognition blur of the area to be identified is of medium blur, and the area to be identified is classified as the second category area.

[0035] like If the boundary of the area to be identified is highly blurred when the color of the shrub is similar to the background color, then the area to be identified is classified as a third category area.

[0036] Preferably, based on the classification results, the corresponding shrub vegetation spatial distribution area extraction method is selected for different categories of areas to be identified, specifically as follows:

[0037] For the region to be identified that is classified as the first category region, a fixed threshold segmentation method based on the distribution of pixel values ​​in the green band among the three bands of red, green and blue is used to process and extract the spatial distribution region of shrubs;

[0038] For the region to be identified that is classified as the second category region, the Laplacian edge convolution operation is first performed on the image to enhance the boundary information. Then, the red, green and blue band images are stitched together to form a fused image. Based on the joint threshold judgment of edge response intensity and color difference in the fused image, the spatial distribution region of shrubs is extracted.

[0039] For the region to be identified that is classified as the third category region, the current image and the image of the same region to be identified at a historical time point are obtained, and a difference map of the change of green band pixels in the two temporal images is generated. Then, the region with enhanced green pixel value and edge gradient magnitude change below the preset boundary change threshold is identified in the difference map and is taken as the shrub spatial distribution region.

[0040] Preferably, a dynamic assessment system for the ecological health of riparian vegetation includes an image preprocessing module, a heterochromatic interference identification module, a boundary fuzziness assessment module, a hierarchical extraction decision module, and an ecological display generation module.

[0041] The image preprocessing module acquires image data containing the riverbank area, preprocesses the image data including image denoising, registration and multi-channel alignment, and extracts the riverbank area image by combining terrain information, and divides the riverbank area image into several image unit grids.

[0042] The heterochromatic interference recognition module determines whether there is a situation where the color of the shrub is similar to the background color in each image unit grid based on color and texture features, and marks the image unit grid with such a situation as the area to be recognized.

[0043] The boundary blur assessment module extracts the boundary recognition feature information of each region to be identified, analyzes it after extraction, evaluates the degree of boundary recognition blur of each region to be identified when the shrub color is close to the background color, and classifies it according to the evaluation results.

[0044] The hierarchical extraction decision module selects the corresponding shrub vegetation spatial distribution area extraction method for different categories of areas to be identified based on the classification results, so as to extract the shrub spatial distribution area in each area to be identified.

[0045] The ecological display generation module performs image range statistics on the extracted spatial distribution areas of shrubs and outputs a display of the ecological health status of the riverbank area based on the statistical results.

[0046] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0047] 1. This invention achieves quantitative analysis of the blurring degree of shrub boundaries by constructing two types of high-dimensional parameter models: "boundary transition perception coefficient" and "boundary recognition obstacle index," which fuse image gradient structure and texture color fusion features. This effectively overcomes the high dependence of traditional image segmentation methods on color transitions and edge sharpness. Based on this, by classifying the blurring degree and implementing a hierarchical recognition strategy, it can select the most adaptive spatial distribution region extraction method for complex scenes where shrubs and background colors are extremely similar and boundary contours are unclear. This significantly improves the accuracy and stability of shrub extraction and avoids spatial deviations caused by misidentification.

[0048] 2. This invention is based on the technical concept of "hierarchical extraction of the region to be identified," which matches regions with different levels of fuzziness with three extraction paths: fixed threshold segmentation, fusion channel edge enhancement, and multi-temporal image change detection, thus constructing a flexible region adaptation mechanism. This multi-strategy fusion extraction framework not only improves the method's adaptability to different image sharpness, background complexity, and temporal dynamics, but also significantly enhances the system's generalization ability under multiple source images, different observation conditions, and different vegetation habitats, exhibiting a wider range of engineering applications and robust performance.

[0049] 3. This invention uses image unit grids as the smallest unit of analysis and maps the extracted results to the entire riverbank area for statistical output. The proposed technical solution enables structural-level assessment of shrub cover distribution, providing fundamental data support with higher spatial resolution and clearer structural expression for ecological health status. This closed-loop analysis process, from pixel-level identification to regional-level statistics, not only improves the accuracy of ecological assessment results but also avoids erroneous interventions caused by shrub identification biases. It significantly enhances the scientific rigor, precision, and controllability of ecological governance, demonstrating good practical application value and promising prospects for wider application. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0051] Figure 1 This is a flowchart illustrating a dynamic assessment system and method for the ecological health of riparian vegetation according to the present invention.

[0052] Figure 2 This is a schematic diagram of the modules of a dynamic assessment system and method for the ecological health of riverbank vegetation according to the present invention. Detailed Implementation

[0053] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0054] This invention provides, for example Figure 1 The method for dynamic assessment of the ecological health of riparian vegetation, as shown, specifically includes the following steps:

[0055] Image data containing the riverbank area is acquired, and the image data is preprocessed, including image denoising, registration and multi-channel alignment. The riverbank area image is extracted in combination with the terrain information and divided into several image unit grids.

[0056] Image data of riverbank areas can be acquired through various remote sensing platforms, including high-resolution satellite imagery (such as Sentinel-2 and Landsat-8), low-altitude aerial images from UAVs, or near-ground photographic images acquired by ground-based mobile devices. Upon receiving the raw image data, the software system first performs image denoising, typically employing median filtering, Gaussian filtering, or wavelet denoising algorithms to reduce pixel-level anomalies caused by sensor noise, weather interference, and other factors. Subsequently, spatial correction is performed on multi-temporal or multi-platform images using feature-point-based image registration methods (such as SIFT / SURF feature extraction and affine transformation matching) or template matching methods based on grayscale information, ensuring spatial alignment between images acquired at different times or angles. After image registration, multi-channel alignment is performed, synchronously overlaying data from multiple bands or sensor channels, including visible light, near-infrared, and elevation, with pixel-level precision, to form a structurally consistent composite image with multispectral features, used for subsequent shrub identification and health assessment.

[0057] Topographic information can be imported into the system via DEM (Digital Elevation Model), DOM (Digital Orthophoto), or geographic vector boundary files (such as river channel vector lines and shoreline area layers) for fusion processing. In the software, preliminary analysis blocks are first generated based on the river centerline and the buffer zones on the left and right banks. Then, combined with elevation data, the location of the slope toe, the slope top boundary, and the water body edge characteristics are determined. Pixels on non-riverbank highlands or water surfaces are eliminated, accurately extracting riverbank image areas with geomorphic continuity and vegetation growth potential. Next, based on image coordinates or geographic coordinates, the extracted riverbank areas are divided into several image unit grids according to rules. The division method can be a fixed-size raster (e.g., 50m × 50m) or adaptive partitioning (e.g., adaptive grids based on image texture changes). Each image unit grid serves as the smallest analysis unit for subsequent evaluation, facilitating image recognition, index calculation, and health classification processing by block, supporting dynamic analysis and spatial comparison.

[0058] The reason for preprocessing and gridding image data is that raw images often suffer from noise interference, temporal misalignment, and spectral inconsistencies. Without registration, filtering, and multi-channel alignment, shrub information at the same location in the image may experience spatial drift, blurring, or even loss, severely impacting the accuracy of subsequent feature extraction and recognition algorithms. Furthermore, riverbank areas are spatially irregular, with blurred boundaries and complex terrain variations. Without incorporating terrain information to eliminate non-analyzable areas, non-target regions will interfere with the evaluation results, leading to the mis-extraction of non-vegetation objects such as farmland, bare land, water bodies, or man-made structures. Image gridding provides stable spatial indexing units for subsequent analysis, allowing the model to independently analyze the degree of boundary blurring and health status of shrubs on a grid-by-grid basis. This supports regional index aggregation, time-series comparison, and visualization, effectively improving the overall system's stability, scalability, and interpretability.

[0059] Based on color and texture features, determine whether there are shrub colors that are similar to the background color in each image cell grid, and mark the image cell grid with such a situation as the region to be identified;

[0060] In this embodiment, based on color and texture features, it is determined whether each image unit grid has a shrub color that is close to the background color, and the image unit grid with this condition is marked as a region to be identified, including:

[0061] Analyze the color distribution differences between shrub vegetation areas and adjacent non-vegetation areas in each image cell grid;

[0062] To analyze the color distribution differences between shrubland areas and adjacent non-vegetation areas in each image cell grid, this can be achieved using software methods such as image partitioning, color statistics, and distribution modeling. First, based on image preprocessing and segmentation, the suspected shrubland areas in each image cell grid are initially partitioned with their surrounding non-vegetation areas. This can be done using existing normalized vegetation indices (such as NDVI) or color-based K-means clustering to identify candidate shrubland and background areas within each image cell grid. Then, the distribution of pixels in the RGB, HSV, or Lab color spaces is statistically analyzed in each region, and histograms or Gaussian mixture models are constructed to model the color distribution. After modeling, the statistical differences between the color distribution of shrubland areas and adjacent non-vegetation areas are calculated, for example, using Bhattacharyya distance, KL divergence, or comparison of mean and variance, to determine the degree of color differentiation. If the color distributions of the two areas overlap significantly, i.e., the difference is small, it indicates that the shrubland area color is similar to the background area, providing a basis for subsequent identification of areas with blurred boundaries. The entire process can be achieved through image processing software or a deep learning image feature extraction toolchain, supporting batch processing of image grid data.

[0063] Analyze the differences in texture arrangement direction and density between shrub vegetation areas and adjacent non-vegetation areas in each image unit grid;

[0064] To analyze the differences in texture arrangement direction and density between shrubland areas and adjacent non-vegetated areas in each image cell grid, image processing methods involving texture feature extraction and distribution analysis can be used. First, within each image cell grid, shrubland areas and adjacent non-vegetated areas are initially divided using NDVI or color clustering results. Then, direction-sensitive texture analysis methods, such as Gray-Level Co-occurrence Matrix (GLCM), Gabor filter banks, or Local Binary Pattern (LBP), are used to extract texture direction distribution and density features from the two areas respectively. In direction extraction, the dominant direction and its concentration of texture arrangement in each area can be determined by comparing the energy responses of multiple directions (e.g., 0°, 45°, 90°, 135°), reflecting the orderliness and directionality of the texture structure. In density analysis, the frequency of change of texture patterns or texture energy per unit area can be calculated to measure the richness of detail within the area. After extraction, the texture features of shrub and non-vegetated areas are statistically modeled (e.g., texture orientation histograms, density distribution curves, etc.), and the differences in orientation distribution and density statistics between the two are calculated, quantified by methods such as cross-entropy, histogram overlap ratio, or mean difference. A low difference indicates similar texture structures between the two regions, posing a risk of blurred boundary recognition. This analysis process can be implemented using image processing and pattern recognition software tools, such as OpenCV, Matlab image toolboxes, or Python image processing libraries, suitable for batch, automated region texture recognition and comparison tasks.

[0065] When the color distribution difference between the shrub vegetation area and the adjacent non-vegetation area is lower than the color judgment standard set based on the pixel color value clustering difference threshold, and the distribution difference of the texture arrangement direction and density between the shrub vegetation area and the adjacent non-vegetation area is lower than the texture judgment standard set based on the texture direction histogram overlap ratio, it is determined that the image unit grid has a situation where the shrub color and the background color are similar.

[0066] To determine if an image cell grid has a shrub color similar to the background color when the color distribution difference between the shrub vegetation area and the adjacent non-vegetation area is lower than the color judgment standard set based on the pixel color value clustering difference threshold, and the distribution difference in texture arrangement direction and density between the shrub vegetation area and the adjacent non-vegetation area is lower than the texture judgment standard set based on the overlap ratio of texture direction histograms, two sets of difference judgment processes can be set up and automated for analysis. In the color dimension, the pixels of the shrub area and the adjacent non-vegetation area are first clustered (using methods such as K-means or MeanShift) to form two types of principal color feature vectors. Then, based on these feature vectors, the distance between their cluster centers or the distribution overlap in a certain color space (such as Lab space) is calculated. If the distance is lower than a preset threshold, it indicates a small color difference. In the texture dimension, the texture direction histogram of each region is extracted, and the overlap area ratio between the histograms of the two regions is calculated. When this ratio is higher than a set limit, it indicates similar texture distribution. When both conditions are met, the system automatically marks the image grid as a "color-similar region" for subsequent fine boundary processing. The entire judgment process can be completed automatically with the help of image recognition and statistical analysis software without human intervention, and supports parallel execution in multiple images.

[0067] The "color judgment standard based on pixel color value clustering difference threshold" refers to analyzing the typical distribution patterns of color value clustering distances in areas with clear and blurred boundaries after statistically analyzing the main color features of a large number of shrub and non-vegetation areas in multiple image units. An empirical threshold is then selected as the standard for distinguishing whether colors are similar; for example, when the cluster center distance is less than the statistical average at a certain percentile, the colors are considered similar. The "texture judgment standard based on texture direction histogram overlap ratio" extracts texture direction distribution curves of shrub and non-vegetation areas from a large number of training samples, calculates the proportion of histogram intersection areas in samples with clear and blurred boundaries, and sets an overlap threshold as the judgment criterion. For example, when the overlap ratio of the texture direction histograms of two areas exceeds a certain set ratio (e.g., 85%), their textures are considered similar. Both standards can be pre-trained offline using data statistics and sample annotation and embedded in the evaluation program to ensure consistency in subsequent processing and objectivity of the judgment results.

[0068] Image cell grids where the shrub color is similar to the background color are marked as regions to be identified.

[0069] To mark image cell grids where shrub colors are similar to background colors as regions to be identified, a logical judgment and spatial mapping mechanism can be introduced into the image processing workflow via software. First, after performing color and texture difference analysis on each image cell grid, the system assigns a state flag variable to each grid. This variable records whether the grid meets both conditions: "color difference is below the color judgment standard" and "texture difference is below the texture judgment standard." If both conditions are met, the state variable is set to "1" or marked as "to be identified"; otherwise, it is set to "0" or marked as "no identification required." Subsequently, based on this state variable, the system logically marks the corresponding grid positions in a spatial mapping data structure (such as a two-dimensional array or raster index table). The marking results can be visualized as a mask image, visually presenting the location and extent of all regions to be identified on the image. Furthermore, for ease of subsequent processing and classification, each image cell grid marked as to be identified can also be accompanied by an index number or coordinate index information, and stored uniformly in a task queue or identification list for use in subsequent spatial distribution extraction steps. The entire labeling process does not rely on manual operation and can be completed automatically in real time by the algorithm in the image analysis process, and is suitable for the batch processing needs of large-scale image data.

[0070] Extract the boundary recognition feature information of each region to be identified, and analyze it after extraction. Evaluate the degree of boundary recognition ambiguity of each region to be identified when the shrub color is close to the background color, and classify them according to the evaluation results.

[0071] In this embodiment, boundary recognition feature information of each region to be identified is extracted and analyzed. The degree of boundary recognition ambiguity of each region to be identified is evaluated when the shrub color is close to the background color. The regions are then classified according to the evaluation results. Specifically:

[0072] Extract the boundary recognition feature information of each region to be identified, and perform preprocessing operations after extraction;

[0073] Extracting boundary recognition features from each region to be identified typically uses a mask of the region to be identified generated after image segmentation as the basis for localization, combined with spatial analysis of the image's brightness distribution, pixel gradient, and texture response values. The specific extraction methods include: first, applying Sobel or Canny edge detection operators in the image's grayscale space to extract boundary pixels for each region to be identified, constructing a boundary pixel sequence; then, extracting features such as horizontal and vertical brightness gradients, edge magnitude, gradient direction, edge thickness, and texture response distribution from this boundary pixel band, forming a multi-dimensional boundary feature vector set; furthermore, using texture analysis methods such as gray-level co-occurrence matrix, Gabor filter, or Local Binary Pattern (LBP) to further extract texture direction and density information within the boundary band region. This process is generated pixel-by-pixel or block-by-block through sliding windows or region masks, with all extracted information bound to a region index to form a complete set of boundary recognition feature information for subsequent processing.

[0074] The main purpose of preprocessing is to improve the comparability and stability of boundary recognition feature information and reduce interference caused by factors such as noise, brightness drift, and regional scale differences. First, the extracted feature information is standardized, for example, by normalizing the range of brightness gradient values ​​and orientation angle values, ensuring that each dimension of the feature has a uniform dimension and avoiding weight shifts caused by scale inconsistencies in subsequent analysis. Second, to eliminate the interference of local outliers, median filtering or Gaussian smoothing can be applied to the feature data to enhance the continuity and interpretability of the overall boundary features. Furthermore, principal component analysis (PCA) or feature selection algorithms are introduced into the texture features to compress dimensions and eliminate redundant indicators, ensuring that the input data is both compact and rich in discriminative power. The entire preprocessing process is automatically executed through image processing libraries and mathematical statistical tools, requiring no manual intervention. It is widely adaptable to multi-resolution and multi-source image data, providing a solid foundation for subsequent boundary blur assessment.

[0075] Boundary structure contrast information and image texture color fusion information are extracted from the preprocessed boundary recognition feature information, and normalization is performed after extraction;

[0076] To extract boundary structure contrast information and image texture and color fusion information from preprocessed boundary recognition feature information, multi-dimensional statistical analysis and feature coupling processing of the boundary neighborhood of each region to be identified is usually required. First, when extracting boundary structure contrast information, standardized gradient magnitude and gradient direction maps can be used to calculate the gray-level gradient change amplitude, edge direction consistency, and edge transition distance for each boundary pixel band of the region to be identified. The gray-level gradient change amplitude can be obtained from the average gray-level difference of 3×3 pixel blocks on both sides of the boundary line; direction consistency is obtained by calculating the concentration of gradient directions; and the transition distance is quantified based on the distance curve of edge intensity decaying from the center outwards. These three together constitute the boundary structure contrast information of the region. Second, when extracting image texture and color fusion information, the mean and standard deviation of color channels, the color variance ratio, and the degree of overlap of texture histograms (such as LBP or Gabor textures) can be calculated inside and outside the boundary band. By weighted combination of color differences and texture differences, a comprehensive data expression describing the clarity of region boundaries is formed. These extraction processes are all executed automatically through image analysis and matrix operations, without relying on manual intervention, and can efficiently extract boundary blur-related information from high-resolution images.

[0077] Based on the normalized boundary structure contrast information and image texture color fusion information, the boundary transition perception coefficient and boundary recognition obstacle index of each region to be identified are generated respectively.

[0078] Based on the generated boundary transition perception coefficient and boundary recognition obstacle index of each region to be identified, a boundary recognition fuzziness index of each region to be identified is generated.

[0079] A pre-defined threshold range for boundary recognition fuzziness index is determined and compared with the generated boundary recognition fuzziness index of each region to be recognized. The degree of boundary recognition fuzziness of each region to be recognized is evaluated based on the comparison results when the shrub color is close to the background color, and the regions are classified based on the evaluation results.

[0080] To determine the threshold range of the fuzziness index for boundary recognition using software, offline statistics and distribution modeling can be performed based on labeled regions with known shrub boundary recognition quality from a historical image sample database. Specifically, a representative image dataset is first selected, and the corresponding boundary recognition fuzziness index values ​​are calculated for multiple shrub regions with varying degrees of difficulty in manually confirmed boundary recognition. These values ​​are then categorized into three classes based on recognition performance: "easy to recognize," "medium," and "difficult to recognize." Next, based on the fuzziness index distribution of these three classes of samples, statistical methods such as K-means clustering, Jenks' method (natural discontinuity method), or Gaussian mixture model (GMM) are used to segment the index values, automatically dividing them into three non-overlapping numerical intervals corresponding to low, medium, and high fuzziness, respectively. After segmentation, the upper and lower bounds of each interval are used as preset threshold ranges for the boundary recognition fuzziness index and fixed as callable configuration parameters for subsequent real-time comparison and classification operations. The entire threshold range setting process can be integrated into the model training or system initialization phase for automatic completion, eliminating the need for manual adjustment and ensuring the consistency and objectivity of the evaluation criteria.

[0081] In this embodiment, the logic for obtaining the boundary transition sensing coefficient of each region to be identified is as follows:

[0082] Boundary structure comparison information is extracted from the preprocessed boundary recognition feature information, specifically including three types of data: the average gray-level gradient magnitude, the gray-level transition length, and the standard deviation of the gradient direction of each region to be identified. These three types of data are divided by their respective preset maximum values ​​to obtain the normalized average gray-level gradient magnitude, gray-level transition length, and standard deviation of the gradient direction of each region to be identified, which are then labeled as follows: , and , The normalized first The average gray-level gradient magnitude of the edges of the regions to be identified. The normalized first The grayscale transition length at the edge of the region to be identified The normalized first The standard deviation of the gradient direction at the edge of each region to be identified. , It is a positive integer;

[0083] To obtain three types of data for each region to be identified—average gray-level gradient magnitude, gray-level transition length, and gradient direction standard deviation—image processing algorithms can be used in software to scan the image edge region pixel by pixel. Specifically, firstly, the Sobel or Prewitt operator is used to identify the edge positions in each region, extracting the set of edge pixels. Based on this, the average gray-level gradient magnitude of the region is obtained by calculating the gradient magnitude of each edge pixel and averaging the gradient magnitudes of all edge pixels in that region. Secondly, in the gradient distribution map, using the center point of each edge as a reference point, the continuous transition range of gray-level values ​​from high to low or from low to high along the axis perpendicular to the edge direction is statistically analyzed; that is, the distance where the slope of the gray-level change curve tends to be stable, thus calculating the edge gray-level transition length. Finally, based on the distribution of the gradient direction angles (usually from 0° to 180°) of all edge pixels, their standard deviation is calculated to represent the dispersion of the boundary direction, thus obtaining the gradient direction standard deviation for that region. All of the above operations can be automatically performed in batches within the software using an image analysis toolkit.

[0084] Dividing these three types of data by their respective preset maximum values ​​is a normalization process. This eliminates differences in numerical scale and physical units, preventing one type of data from dominating the overall calculation result due to excessively large absolute values ​​during the subsequent calculation of the boundary transition sensing coefficient. This ensures equal weighting of various structural information in the index. The preset maximum values ​​can be obtained through sample statistics. Specifically, during the model building phase, a large number of edge sample data are extracted from multiple typical riverbank area images, and the maximum observed values ​​of these three types of data in all samples are recorded. Alternatively, the upper quantile (e.g., the 95th percentile) after removing extreme values ​​can be set as the maximum value, thereby improving the model's stability and anti-interference ability. These statistical threshold settings can be learned and solidified as baseline configurations during the software training phase, and then invoked as needed during runtime, ensuring the scientific rigor and universality of the normalization process.

[0085] Calculate the boundary transition sensing coefficient of each region to be identified. The specific calculation method is as follows: the gray-scale transition length of the edges of each normalized region to be identified is... The average gray-level gradient magnitude of the edges of each region to be identified is raised to the power of the normalized value. The natural logarithm of the ratio obtained by adding one and the normalized standard deviation of the gradient direction of each region to be identified. The square roots are summed to obtain the boundary transition sensing coefficients of each region to be identified. ;

[0086] The specific calculation formula is as follows: In the formula, For the first The boundary transition sensing coefficient of the region to be identified.

[0087] This formula is used to calculate the boundary transition perception coefficient of each region to be identified. Its core purpose is to comprehensively evaluate the blurriness of the shrub boundary region in the image by combining and modeling the multi-dimensional features of the image edge structure. The three types of parameters in the formula reflect the clarity of the boundary from three aspects: boundary direction consistency, gray-scale transition width, and edge sharpness. These parameters are fused through a nonlinear function to enhance the distinguishability of the perception of boundary blur.

[0088] Specifically: the first item The square root of the standard deviation of the normalized edge gradient direction reflects the degree of perturbation in the edge direction. A larger standard deviation indicates more inconsistent and irregular edge contour directions, resulting in greater ambiguity. Using the square root function weakens the dominance of maxima, allowing the influence of boundary direction distribution to be more smoothly integrated into the final exponent. (The second term...) The exponential operation representing the normalized grayscale transition length reflects the continuous blurring effect caused by the slow change in grayscale values ​​in the boundary region. A longer grayscale transition indicates a smoother pixel grayscale change from the shrubs to the background in the boundary region. The exponential function amplifies the weight of such blurred boundary regions in the exponent, making them more prominent in the sense of boundary blur. The denominator... Adding 1 to the normalized average gradient magnitude of edge pixels represents an adjustment factor for edge sharpness. A larger gradient indicates a sharper boundary, thus "suppressing" the blur exponent in the denominator; adding 1 avoids a denominator of 0 and controls the numerical scale. Dividing the exponent term by this gradient term and taking the natural logarithm helps compress the numerical range of the result and enhances the model's ability to distinguish between "blurred" and "sharp" levels under different boundary conditions. The entire formula structure embodies a balancing mechanism: directional confusion and slow transition enhance the sense of blur, while edge sharpness suppresses it. Through the weighted combination of these factors, this exponent can effectively characterize the "difficulty in distinguishing" between shrub boundaries and the background when colors are similar, providing a quantifiable decision-making basis for downstream classification and region extraction strategies.

[0089] No. Boundary transition sensing coefficient of the region to be identified This coefficient reflects the degree of blurring of the boundary features between the shrubs and the background in the image. A higher value indicates that the boundary of the region is more difficult to segment clearly using conventional image analysis methods. Specifically, the coefficient is composed of factors including the dispersion of edge direction, the transition width of grayscale changes, and the sharpness of the edges. More dispersed edge direction, slower grayscale transitions, and weaker edge gradients all lead to a higher coefficient. Since these factors are precisely the key conditions affecting the accuracy of boundary recognition, the value of the boundary transition perception coefficient can be directly used as a basis for assessing the degree of boundary blurring: a smaller coefficient indicates a clear boundary structure, short transitions, and concentrated directions, making the boundary easy to identify; a larger coefficient indicates that the boundary exhibits weak contrast, weak structure, and weak directionality, making it difficult to accurately segment the true boundary when the shrubs and background colors are similar. Therefore, this coefficient has high interpretability and operability in assessing the degree of boundary blurring and is an indispensable key parameter for quantifying "difficult-to-identify boundaries."

[0090] In this embodiment, the logic for obtaining the boundary recognition obstacle index of each region to be identified is as follows:

[0091] Image texture and color fusion information is extracted from the preprocessed boundary recognition feature information. Specifically, this includes three types of data: the mean Euclidean distance of each region to be recognized, the texture orientation histogram cross ratio, and the pixel color variance ratio. These three types of data are divided by their respective preset maximum values ​​to obtain the normalized mean Euclidean distance of each region to be recognized, the texture orientation histogram cross ratio, and the pixel color variance ratio, which are then labeled as follows: , and , The normalized first The Euclidean distance between the color mean of each region to be identified The normalized first The cross ratio of the texture orientation histogram of each region to be identified. The normalized first The pixel color variance ratio of the region to be identified , It is a positive integer;

[0092] The acquisition of three types of data for each region to be identified—the mean color Euclidean distance, the texture orientation histogram cross ratio, and the pixel color variance ratio—can be achieved through software based on image processing and feature extraction algorithms. The mean color Euclidean distance is calculated by taking the mean vectors of pixel colors in the RGB or Lab color space for the shrubland area and its adjacent non-vegetation area, and then using the Euclidean distance formula to measure the difference in mean color between them, reflecting their overall color distinguishability. The texture orientation histogram cross ratio is calculated by extracting the gradient direction distributions of two regions (e.g., using oriented gradient histograms or Gabor filter results to construct histograms) and then calculating their intersection ratio, measuring the similarity of texture directions between regions. The pixel color variance ratio is calculated by taking the ratio of the standard variances of pixel colors in two regions; this value describes the consistency and degree of variation of colors within the region. These three types of data can be automatically obtained by analyzing the pixel attributes of different regions within each image unit grid, and have clear numerical representations.

[0093] Dividing the three types of data by their respective preset maximum values ​​is a normalization process. This standardizes the numerical range, eliminates biases caused by different units and scales in subsequent calculations, and improves the comparability between parameters and the precision of weight control. The preset maximum values ​​can be derived by statistically analyzing the historical maximum observed value or the upper limit of the 95% confidence interval for each type of data from a large number of representative sample images. Alternatively, an upper limit range that does not exceed the maximum value that might occur in the system can be set based on human experience, ensuring that the normalized values ​​remain stable between 0 and 1. This normalization method facilitates the subsequent integration of various indicators into the index calculation model, enhancing the stability and universality of the overall evaluation.

[0094] Calculate the boundary recognition obstacle index for each region to be identified. The specific calculation method is as follows: the cross ratio of the texture orientation histograms of each normalized region to be identified is calculated. The first term is obtained by squaring the pixel color variance ratio of each normalized region to be identified. Taking its square root yields the second term, which is the normalized color mean Euclidean distance of each region to be identified. Add one and take the reciprocal, then add another one and take the natural logarithm as the third term. Add the first, second, and third terms to obtain the boundary recognition obstacle index for each region to be identified. ;

[0095] The specific calculation formula is as follows: In the formula, For the first Boundary recognition obstacle index for each region to be identified.

[0096] The boundary recognition obstacle index is calculated by integrating the boundary characteristics of shrub-covered areas and adjacent non-vegetated areas across three dimensions: color, texture direction, and color distribution differences. This index measures the clarity of the boundary between image regions when the shrubs and background colors are similar. The first term of this index... The square of the cross ratio of the normalized texture direction histogram reflects the consistency or overlap of texture directions in regions. Squaring enhances the discriminative power of this index under similar texture conditions; a larger value indicates that the texture directions of the two regions are similar and the boundaries are not obvious. The second term... This is the square root of the pixel color variance ratio, used to measure the uniformity of color distribution differences within a region. A larger value indicates smaller internal color differences and less prominent boundaries; the third term... This is a non-linear mapping of the Euclidean distance between color means, used to smooth out the impact of small differences on the overall result, significantly increasing the contrast index of regions with similar color means. Overall, when all three indices are high, it indicates that the region has small color differences at its boundaries, high overlap in texture directions, uniform color distribution within the region, natural transitions between regions, and blurred boundaries. Consequently, the boundary recognition obstacle index increases, indicating greater difficulty in identifying the boundaries of that region. Conversely, lower index values ​​indicate clear boundary features between regions, making accurate identification easier.

[0097] No. Boundary identification obstacle index of the area to be identified The magnitude of the boundary recognition obstacle index is positively correlated with the degree of boundary recognition ambiguity when the shrub color is similar to the background color. Specifically, a larger index value indicates that the boundary contrast between the area to be identified and adjacent non-vegetated areas is low in multiple dimensions such as color mean, texture direction, and color distribution. This means that the differences between regions are weak, the transition is gentle, and the textures are similar, making it difficult to clearly separate the boundary in the image, thus belonging to a high-ambiguity region. Conversely, a lower index value indicates that the shrub area has significant differences from adjacent areas in terms of color and texture, and the boundary is clearly distinguishable, belonging to a low-ambiguity region. Therefore, the larger the boundary recognition obstacle index value, the higher the degree of boundary recognition ambiguity, and the more complex the extraction strategy needs to be to avoid recognition errors.

[0098] In this embodiment, the boundary transition sensing coefficients of each region to be identified are used as the basis for the results. and boundary recognition barrier index The boundary recognition fuzziness index of each region to be identified is generated by weighted summation, specifically by determining the boundary transition perception coefficient of each region to be identified. and boundary recognition barrier index Each region has a pre-set first non-zero weight coefficient and a second non-zero weight coefficient, the sum of which is one; the first non-zero weight coefficient is multiplied by the boundary transition sensing coefficient of each region to be identified. The result is multiplied by the second non-zero weighting coefficient and the boundary recognition obstacle index of each region to be identified. The results are summed to obtain the boundary recognition fuzziness index of each region to be identified. .

[0099] The specific calculation formula is as follows: In the formula, For the first The boundary recognition fuzziness index of the region to be identified and These are the boundary transition sensing coefficients for each region to be identified. and boundary recognition barrier index Their respective first non-zero weight coefficients and second non-zero weight coefficients, and .

[0100] In practical implementation, the boundary transition sensing coefficient of each region to be identified can be adjusted using software methods. With boundary recognition obstacle index Perform a unified reading and assign a first non-zero weight coefficient to each ( ) and the second non-zero weighting coefficient ( Then, the boundary recognition fuzzy index is calculated using the weighted summation formula. These two weighting coefficients are used to measure... and Relative contribution to fuzziness assessment The larger the value, the more emphasis is placed on boundary transition features. A larger value indicates a greater emphasis on the integration of color and texture features. The preset values ​​for these two coefficients can be statistically fitted using a historical image training dataset: extracting data from known-labeled vegetation area samples... and A measure of consistency with human assessment of ambiguity is determined using the least mean square error method or multiple linear regression to identify an optimal set of metrics. and Alternatively, weights can be allocated using empirical models in different scenarios to ensure adaptability and model stability. All of the above settings can be configured via a graphical interface or loaded and invoked by the software using pre-set model configuration files, requiring no manual intervention.

[0101] In this embodiment, a pre-defined threshold range for boundary recognition fuzziness index is determined. And after determination, it identifies the fuzzy index of the boundaries of each region to be identified. A comparison was performed, and the degree of ambiguity in the boundary recognition of each region to be identified was evaluated based on the comparison results when the shrub color and the background color were similar. The regions were then classified according to the evaluation results. The specific comparison analysis and classification are as follows:

[0102] like If the shrub color is close to the background color, the boundary recognition blur level of the area to be identified is low blur level, and then the area to be identified is classified as the first category area;

[0103] This indicates that the boundary features between the shrub vegetation and the adjacent background in this area are clear, with strong contrast in color, texture, and structure. The boundaries of such areas are easily and accurately identified by current edge detection or segmentation methods in image analysis, exhibiting good recognition conditions. Therefore, in the subsequent stage of shrub spatial distribution region extraction, conventional methods based on color segmentation or edge detection can be directly used to achieve efficient segmentation, avoiding over-processing while ensuring data processing efficiency and discrimination accuracy, thus contributing to the rapid construction of a realistic and reliable shrub cover map.

[0104] like If the shrub color is similar to the background color, the boundary recognition blur of the area to be identified is of medium blur, and the area to be identified is classified as the second category area.

[0105] This indicates that the boundary between the shrub vegetation and the background in this area is partially blurred in the image, meaning there is a certain degree of color fusion or texture similarity, but they are not completely indistinguishable. Relying solely on a single image feature processing method when extracting shrubs from these areas can easily lead to missed extractions or misclassifications. Therefore, it is necessary to combine multi-dimensional processing techniques (such as edge enhancement and multi-band fusion) to improve recognition accuracy. The processing quality of these areas directly affects the accurate calculation of structural integrity and coverage in ecological health indicators, and is a crucial step in ensuring the reliability of the assessment results.

[0106] like If the boundary of the area to be identified is highly blurred when the color of the shrub is similar to the background color, then the area to be identified is classified as a third category area.

[0107] This situation indicates that the boundary contrast between the shrubs and the background in the image is extremely weak, with highly similar colors, textures, and edge transitions, making it difficult for current common segmentation methods to accurately distinguish shrub vegetation. Direct extraction without processing will result in a severe underestimation of shrub cover, affecting the effectiveness of ecological structure indicator assessments. Therefore, when processing such areas, more complex techniques such as multi-temporal image comparison and trend analysis should be employed to enhance discriminative power, ensuring that these shrub areas are reasonably identified and restored, avoiding biased conclusions about the ecological status, and thus affecting the scientific validity of subsequent intervention decisions.

[0108] Based on the classification results, select the corresponding shrub vegetation spatial distribution area extraction method for different categories of areas to be identified, so as to extract the shrub spatial distribution area in each area to be identified;

[0109] In this embodiment, based on the classification results, the corresponding shrub vegetation spatial distribution region extraction method is selected for different categories of regions to be identified. Specifically:

[0110] For the region to be identified that is classified as the first category region, a fixed threshold segmentation method based on the distribution of pixel values ​​in the green band among the three bands of red, green and blue is used to process and extract the spatial distribution region of shrubs;

[0111] For the regions to be identified, which are classified as the first category, the boundary recognition is less ambiguous. The shrub vegetation in the image differs significantly from the surrounding background in color and structure, especially in the pixel values ​​of the green band. Therefore, extraction can be performed using a fixed threshold segmentation method based on the distribution of green band pixel values ​​in the red, green, and blue bands. Specifically, the pixel value matrix of the green band in the image unit is extracted first, and pixel histogram analysis is performed on the entire image. A fixed threshold interval for the green band is determined by combining sample statistics or historical analysis. This interval corresponds to the pixel intensity range formed by reflections from typical shrub leaves. Then, each pixel in the green band image is traversed, and pixels whose values ​​fall within this fixed interval are marked as suspected shrub pixels. Further morphological operations such as opening and closing operations and connected component analysis are used to remove isolated noise points and connect scattered regions, ultimately generating a mask layer for the spatial distribution of shrubs. This approach is suitable for the first category of regions because these regions have clear boundaries and prominent green features. Color-dominated extraction can quickly and efficiently identify the distribution area without introducing complex fusion or temporal analysis operations, thus combining accuracy and computational efficiency.

[0112] For the region to be identified that is classified as the second category region, the Laplacian edge convolution operation is first performed on the image to enhance the boundary information. Then, the red, green and blue band images are stitched together to form a fused image. Based on the joint threshold judgment of edge response intensity and color difference in the fused image, the spatial distribution region of shrubs is extracted.

[0113] For the region to be identified, which is classified as the second category, its boundary recognition is moderately blurred. While the color difference and structural boundary information between the shrubs and the background in the image are relatively weak, they still possess a certain basis for identification. Therefore, a method combining edge enhancement and color difference judgment is adopted to improve extraction accuracy. Specifically, the image is first subjected to Laplacian edge convolution, which significantly improves the edge response intensity of the shrub boundary by enhancing areas with prominent gradient changes. Then, the red, green, and blue band images are standardized and concatenated to construct a three-channel fused image, which retains both color information and edge features. Subsequently, the edge response intensity (e.g., based on the absolute value of the convolution result) and the color difference with surrounding pixels (e.g., based on Euclidean distance or cosine similarity) are calculated for each pixel in the fused image. A joint threshold is then set to determine whether a pixel belongs to the shrub region: only when a pixel simultaneously satisfies both the edge response intensity exceeding the edge threshold and the color difference exceeding the color threshold is it identified as a pixel in the target region. Finally, connectivity analysis is performed on the pixels that meet the conditions to extract continuous regions as the shrub distribution range. The reason for adopting this approach is that the second category region has a certain degree of ambiguity in both color and edge features. Segmentation by a single dimension is prone to misjudgment or missed detection. By adopting a dual-dimensional joint threshold strategy in the fused image, the remaining feature difference information can be effectively utilized to achieve more robust region recognition.

[0114] For the region to be identified that is classified as the third category region, the current image and the image of the same region to be identified at a historical time point are obtained, and a difference map of the change of green band pixels in the two temporal images is generated. Then, the region with enhanced green pixel value and edge gradient magnitude change below the preset boundary change threshold is identified in the difference map and is taken as the shrub spatial distribution region.

[0115] For regions classified as the third category, the degree of ambiguity in boundary recognition is high, meaning that the shrubs and background in the current image are almost indistinguishable in multiple dimensions such as color, texture, and edges. Conventional image segmentation and edge detection methods cannot accurately distinguish the spatial distribution areas of the shrubs. To extract the spatial distribution areas of shrubs that are still identifiable even when the image boundary recognition is highly ambiguous within the third category, a joint method of two-temporal image difference analysis and boundary gradient evaluation can be used. First, the current image and images acquired at historical time points are selected in the software. After ensuring that the georegistration accuracy of the two images is consistent, the pixel values ​​in the green band of these two images are extracted to construct a green channel change map for each pixel at the two time points. By calculating the pixel difference in the green band of the two images, a green band change difference map is generated. Subsequently, an edge detection algorithm (such as the Sobel, Laplacian, or Canny operator) is applied to this difference map to calculate the edge gradient magnitude of each pixel. The difference between the changed edge gradient map and the edge gradient map of the original image is then calculated to obtain the edge gradient magnitude change map. Based on this, a boundary change evaluation threshold is set. Regions with significantly enhanced green pixel values ​​but edge gradient changes below this preset threshold are identified as areas exhibiting enhanced vegetation over time but with stable structural boundaries. These can be inferred to be shrub regions that were originally difficult to clearly segment from single-phase images. The advantage of this method is that by combining the changing trends of time series analysis with the stability characteristics of spatial structure, it can effectively improve the extraction accuracy of shrub regions in situations where boundary recognition is ambiguous. It is suitable for scenarios where the color of shrubs is close to the height of the background in the image, making traditional boundary extraction difficult to perform effectively. The entire process can be implemented in image analysis software as a streamlined algorithm, supporting automatic recognition and batch processing.

[0116] The image range of the extracted shrub spatial distribution area is statistically analyzed, and the ecological health status of the riverbank area is displayed based on the statistical results.

[0117] To achieve the goal of "statistical analysis of the extracted spatial distribution areas of shrubs in images, and outputting a display of the ecological health status of riverbank areas based on the statistical results," the first step is to perform pixel-level range analysis on the identified spatial distribution areas of shrubs. Specifically, using the pixel labels marked as shrub areas in the image segmentation results, the total number of pixels within each image cell grid and the entire riverbank area is accumulated, and then converted into area units based on the spatial resolution of the image. This statistical operation can be automatically completed using raster image processing methods, and the statistical results represent the spatial range of shrub distribution. To facilitate regional comparison, the riverbank can be divided into zones along its longitudinal or transverse direction, and the shrub coverage area of ​​each zone can be summarized separately to observe its spatial distribution pattern and relative concentration.

[0118] After completing the above statistics, to showcase the ecological health status, the statistically obtained shrub distribution range can be compared with the set regional expected vegetation range, historical average distribution values, or relevant standard parameters to construct a regional evaluation level based on spatial coverage ratio. For example, by setting different area coverage thresholds, each image unit or region can be assigned a low, medium, or high ecological status label. Then, combined with graphic rendering, the evaluation results can be visualized using color grading or level labeling, giving the ecological assessment results an intuitive expression. This display method not only assists researchers or decision-makers in judging the shrub habitat status but also serves as an input interface to provide basic data for subsequent processing stages such as ecological restoration simulation and dynamic evolution monitoring. The entire process can be achieved at the software level through the linkage of image statistical analysis, geographic information system mapping components, and graphic display interfaces, resulting in a seamless and user-friendly workflow.

[0119] like Figure 2 The system shown is a dynamic assessment system for the ecological health of riparian vegetation, including an image preprocessing module, a heterochromatic interference identification module, a boundary fuzziness assessment module, a hierarchical extraction decision module, and an ecological display generation module.

[0120] The image preprocessing module acquires image data containing the riverbank area, preprocesses the image data including image denoising, registration and multi-channel alignment, and extracts the riverbank area image by combining terrain information, and divides the riverbank area image into several image unit grids.

[0121] The heterochromatic interference recognition module determines whether there is a situation where the color of the shrub is similar to the background color in each image unit grid based on color and texture features, and marks the image unit grid with such a situation as the area to be recognized.

[0122] The boundary blur assessment module extracts the boundary recognition feature information of each region to be identified, analyzes it after extraction, evaluates the degree of boundary recognition blur of each region to be identified when the shrub color is close to the background color, and classifies it according to the evaluation results.

[0123] The hierarchical extraction decision module selects the corresponding shrub vegetation spatial distribution area extraction method for different categories of areas to be identified based on the classification results, so as to extract the shrub spatial distribution area in each area to be identified.

[0124] The ecological display generation module performs image range statistics on the extracted spatial distribution areas of shrubs and outputs a display of the ecological health status of the riverbank area based on the statistical results.

[0125] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0126] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0127] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0128] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0129] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0130] The units described as separate components may or may not be physically separate. 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0131] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0132] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for dynamic assessment of the ecological health of riparian vegetation, characterized in that, Specifically, the following steps are included: Image data containing the riverbank area is acquired, and the image data is preprocessed, including image denoising, registration and multi-channel alignment. The riverbank area image is extracted in combination with the terrain information and divided into several image unit grids. Based on color and texture features, determine whether there are shrub colors that are similar to the background color in each image cell grid, and mark the image cell grid with such a situation as the region to be identified; Extract the boundary recognition feature information of each region to be identified, and analyze it after extraction. Evaluate the degree of boundary recognition ambiguity of each region to be identified when the shrub color is close to the background color, and classify them according to the evaluation results. Based on the classification results, select the corresponding shrub vegetation spatial distribution area extraction method for different categories of areas to be identified, so as to extract the shrub spatial distribution area in each area to be identified; The image range of the extracted shrub spatial distribution area is statistically analyzed, and the ecological health status of the riverbank area is displayed based on the statistical results. Boundary recognition feature information of each region to be identified is extracted and analyzed. The degree of ambiguity in boundary recognition of each region is evaluated when the shrub color is close to the background color. Based on the evaluation results, the regions are classified. Specifically: Extract the boundary recognition feature information of each region to be identified, and perform preprocessing operations after extraction; Boundary structure contrast information and image texture color fusion information are extracted from the preprocessed boundary recognition feature information, and normalization is performed after extraction; Based on the normalized boundary structure contrast information and image texture color fusion information, the boundary transition perception coefficient and boundary recognition obstacle index of each region to be identified are generated respectively. Based on the generated boundary transition perception coefficient and boundary recognition obstacle index of each region to be identified, a boundary recognition fuzziness index of each region to be identified is generated. A pre-defined threshold range for boundary recognition fuzziness index is determined and compared with the generated boundary recognition fuzziness index of each region to be recognized. The degree of boundary recognition fuzziness of each region to be recognized is evaluated based on the comparison results when the shrub color is close to the background color, and the regions are classified based on the evaluation results.

2. The method for dynamic assessment of the ecological health of riparian vegetation according to claim 1, characterized in that, Based on color and texture features, determine whether each image cell grid contains shrubs whose color is close to the background color, and mark the image cell grids containing such shrubs as regions to be identified, including: Analyze the color distribution differences between shrub vegetation areas and adjacent non-vegetation areas in each image cell grid; Analyze the differences in texture arrangement direction and density between shrub vegetation areas and adjacent non-vegetation areas in each image unit grid; When the color distribution difference between the shrub vegetation area and the adjacent non-vegetation area is lower than the color judgment standard set based on the pixel color value clustering difference threshold, and the distribution difference of the texture arrangement direction and density between the shrub vegetation area and the adjacent non-vegetation area is lower than the texture judgment standard set based on the texture direction histogram overlap ratio, it is determined that the image unit grid has a situation where the shrub color and the background color are similar. Image cell grids where the shrub color is similar to the background color are marked as regions to be identified.

3. The method for dynamic assessment of the ecological health of riparian vegetation according to claim 1, characterized in that, The logic for obtaining the boundary transition sensing coefficients of each region to be identified is as follows: Boundary structure comparison information is extracted from the preprocessed boundary recognition feature information, specifically including three types of data: the average gray-level gradient magnitude, the gray-level transition length, and the standard deviation of the gradient direction of each region to be identified. These three types of data are divided by their respective preset maximum values ​​to obtain the normalized average gray-level gradient magnitude, gray-level transition length, and standard deviation of the gradient direction of each region to be identified, which are then labeled as follows: , and , The normalized first The average gray-level gradient magnitude of the edges of the regions to be identified. The normalized first The grayscale transition length at the edge of the region to be identified The normalized first The standard deviation of the gradient direction at the edge of each region to be identified. , It is a positive integer; The specific calculation formula is as follows: In the formula, For the first The boundary transition perception coefficient of the region to be identified.

4. The method for dynamic assessment of the ecological health of riparian vegetation according to claim 3, characterized in that, The logic for obtaining the boundary recognition obstacle index of each region to be identified is as follows: Image texture and color fusion information is extracted from the preprocessed boundary recognition feature information. Specifically, this includes three types of data: the mean Euclidean distance of each region to be recognized, the texture orientation histogram cross ratio, and the pixel color variance ratio. These three types of data are divided by their respective preset maximum values ​​to obtain the normalized mean Euclidean distance of each region to be recognized, the texture orientation histogram cross ratio, and the pixel color variance ratio, which are then labeled as follows: , and , The normalized first The Euclidean distance between the color mean of each region to be identified The normalized first The cross ratio of the texture orientation histogram of each region to be identified. The normalized first The pixel color variance ratio of the region to be identified , It is a positive integer; The specific calculation formula is as follows: In the formula, For the first Boundary recognition obstacle index for each region to be identified.

5. The method for dynamic assessment of the ecological health of riparian vegetation according to claim 4, characterized in that, Based on the generated boundary transition sensing coefficients of each region to be identified and boundary recognition barrier index The boundary recognition ambiguity index of each region to be identified is generated by weighted summation. The specific calculation formula is as follows: In the formula, For the first The boundary recognition fuzziness index of the region to be identified and These are the boundary transition sensing coefficients for each region to be identified. and boundary recognition barrier index Their respective first non-zero weight coefficients and second non-zero weight coefficients, and .

6. The method for dynamic assessment of the ecological health of riparian vegetation according to claim 5, characterized in that, Determine the pre-defined threshold range for boundary recognition fuzziness index. And after determination, it identifies the fuzzy index of the boundaries of each region to be identified. A comparison was performed, and the degree of ambiguity in the boundary recognition of each region to be identified was evaluated based on the comparison results when the shrub color and the background color were similar. The regions were then classified according to the evaluation results. The specific comparison analysis and classification are as follows: like If the shrub color is close to the background color, the boundary recognition blur level of the area to be identified is low blur level, and then the area to be identified is classified as the first category area; like If the shrub color is similar to the background color, the boundary recognition blur of the area to be identified is of medium blur, and the area to be identified is classified as the second category area. like If the boundary of the area to be identified is highly blurred when the color of the shrub is similar to the background color, then the area to be identified is classified as a third category area.

7. The method for dynamic assessment of the ecological health of riparian vegetation according to claim 6, characterized in that, Based on the classification results, the corresponding shrub vegetation spatial distribution region extraction method is selected for different categories of regions to be identified, specifically as follows: For the region to be identified that is classified as the first category region, a fixed threshold segmentation method based on the distribution of pixel values ​​in the green band among the three bands of red, green and blue is used to process and extract the spatial distribution region of shrubs; For the region to be identified that is classified as the second category region, the Laplacian edge convolution operation is first performed on the image to enhance the boundary information. Then, the red, green and blue band images are stitched together to form a fused image. Based on the joint threshold judgment of edge response intensity and color difference in the fused image, the spatial distribution region of shrubs is extracted. For the region to be identified that is classified as the third category region, the current image and the image of the same region to be identified at a historical time point are obtained, and a difference map of the change of green band pixels in the two temporal images is generated. Then, the region with enhanced green pixel value and edge gradient magnitude change below the preset boundary change threshold is identified in the difference map and is taken as the shrub spatial distribution region.

8. A dynamic assessment system for the ecological health of riparian vegetation, used to implement the dynamic assessment method for the ecological health of riparian vegetation as described in any one of claims 1-7, characterized in that, It includes an image preprocessing module, a heterochromatic interference recognition module, a boundary blur assessment module, a hierarchical extraction decision module, and an ecological display generation module; The image preprocessing module acquires image data containing the riverbank area, preprocesses the image data including image denoising, registration and multi-channel alignment, and extracts the riverbank area image by combining terrain information, and divides the riverbank area image into several image unit grids. The heterochromatic interference recognition module determines whether there is a situation where the color of the shrub is similar to the background color in each image unit grid based on color and texture features, and marks the image unit grid with such a situation as the area to be recognized. The boundary blur assessment module extracts the boundary recognition feature information of each region to be identified, analyzes it after extraction, evaluates the degree of boundary recognition blur of each region to be identified when the shrub color is close to the background color, and classifies it according to the evaluation results. The hierarchical extraction decision module selects the corresponding shrub vegetation spatial distribution area extraction method for different categories of areas to be identified based on the classification results, so as to extract the shrub spatial distribution area in each area to be identified. The ecological display generation module performs image range statistics on the extracted spatial distribution areas of shrubs and outputs a display of the ecological health status of the riverbank area based on the statistical results. The boundary blur assessment module extracts boundary recognition feature information for each region to be identified, analyzes it after extraction, evaluates the degree of boundary blurring for each region when the shrub color is close to the background color, and classifies it according to the assessment results. Specifically: Extract the boundary recognition feature information of each region to be identified, and perform preprocessing operations after extraction; Boundary structure contrast information and image texture color fusion information are extracted from the preprocessed boundary recognition feature information, and normalization is performed after extraction; Based on the normalized boundary structure contrast information and image texture color fusion information, the boundary transition perception coefficient and boundary recognition obstacle index of each region to be identified are generated respectively. Based on the generated boundary transition perception coefficient and boundary recognition obstacle index of each region to be identified, a boundary recognition fuzziness index of each region to be identified is generated. A pre-defined threshold range for boundary recognition fuzziness index is determined and compared with the generated boundary recognition fuzziness index of each region to be recognized. The degree of boundary recognition fuzziness of each region to be recognized is evaluated based on the comparison results when the shrub color is close to the background color, and the regions are classified based on the evaluation results.

Images