A wetland environment degradation identification processing method based on remote sensing image monitoring

Abstract

The application discloses a kind of wetland environmental degradation identification processing methods based on remote sensing image monitoring, method includes: obtaining original remote sensing image including multi-temporal data and pre-processing, through different time nodes different wave band remote sensing image, capture dynamic wetland environmental change, truly reflect wetland condition;Based on texture feature and spectral feature combination image between each region texture boundary transition clear value to the remote sensing image after pre-processing is implemented regional growth segmentation, obtain local area image, feature extraction result is combined by texture feature and spectral feature, so that subsequent image segmentation processing can take into account details and overall structure;Finally, based on local area image by pre-trained wetland degradation detection model is analyzed, obtains the type of each local area image, provides accurate data basis for the detection and remedy of wetland degradation.

Description

Technical Field

[0001] This invention relates to the field of remote sensing monitoring, and in particular to a method for identifying and processing wetland environmental degradation based on remote sensing image monitoring. Background Technology

[0002] With the increasing severity of global environmental problems, wetland degradation has attracted widespread attention. As an important ecosystem, wetlands not only provide rich biodiversity but also play a vital role in water resource management and climate regulation. However, with rapid urbanization and industrialization, wetland environments have suffered severe pressure, and degradation has become increasingly apparent.

[0003] Existing wetland degradation monitoring technologies primarily rely on the analysis of remote sensing imagery. However, wetland degradation is typically a long-term and complex dynamic process, involving changes in water level and vegetation cover. These changes often exhibit phased characteristics over time, and existing technologies struggle to accurately capture these subtle shifts, resulting in inaccurate identification of wetland degradation and an inability to provide effective scientific evidence for decision-making.

[0004] Furthermore, only highly accurate segmented images can predict and calculate wetland degradation during post-processing of remote sensing monitoring images; therefore, highly accurate wetland degradation segmentation images are a crucial technological foundation. Remote sensing monitoring images of wetland degradation often contain transitional zones between vegetation and water. The complexity of these transitional zones poses a challenge to traditional image processing algorithms during identification, as the texture features of these areas may be blurred and difficult to distinguish effectively. In addition, significant texture differences between different types of land features (such as vegetation, shallow water areas, and bare land) further increase the difficulty of extracting texture information between pixels, thus affecting the final monitoring results. Summary of the Invention

[0005] The purpose of this invention is to provide a method and storage medium for identifying and processing wetland environmental degradation based on remote sensing image monitoring, which solves the above-mentioned technical problems pointed out in the prior art.

[0006] This invention provides a method for identifying and processing wetland environmental degradation based on remote sensing image monitoring, comprising the following steps:

[0007] Acquire raw remote sensing images that include multi-temporal data, the raw remote sensing images including raw remote sensing images of multiple bands; preprocess the raw remote sensing images to obtain preprocessed remote sensing images;

[0008] Texture and spectral features are extracted from the pixels of the preprocessed remote sensing image; based on the texture and spectral features and the clarity value of the texture boundary transition between regions in the image, region growing segmentation is performed on the preprocessed remote sensing image to obtain local region images;

[0009] Based on the images of the local areas, a pre-trained wetland degradation detection model is used for identification and analysis to obtain the type of each local area image.

[0010] In another aspect, the present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described wetland environment degradation identification and processing method based on remote sensing image monitoring.

[0011] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages: Analysis of the above-mentioned wetland environment degradation identification and processing method and storage medium based on remote sensing image monitoring provided by the present invention shows that, in specific applications, firstly, original remote sensing images including multi-temporal data are acquired, the original remote sensing images including original remote sensing images of multiple bands; the original remote sensing images are preprocessed to obtain preprocessed remote sensing images, and dynamic wetland environment changes are captured through remote sensing images of different time points and different bands, truly reflecting the wetland condition; furthermore, by improving the pixel points of the preprocessed remote sensing images... Texture and spectral features are extracted. Based on the texture and spectral features, combined with the clarity values ​​of texture boundary transitions between regions in the image, region growing segmentation is performed on the preprocessed remote sensing image to obtain local region images. This provides quantitative indicators for subsequent accurate region segmentation and classification. The feature extraction results, through the combination of texture and spectral features, enable subsequent image segmentation processing to take into account both details and overall structure. Finally, based on the local region images, a pre-trained wetland degradation detection model is used for identification and analysis to obtain the type of each local region image, providing an accurate data foundation for the detection and remediation of wetland degradation. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the overall operation process of a wetland environment degradation identification and processing method based on remote sensing image monitoring;

[0013] Figure 2 This is a schematic diagram simulating wetland degradation types, based on a method for identifying and processing wetland environmental degradation using remote sensing image monitoring.

[0014] Figure 3 This is a schematic diagram simulating non-wetland degradation types, representing a wetland environment degradation identification and processing method based on remote sensing image monitoring.

[0015] Figure 4This is a schematic diagram of an oversegmentation simulation of a wetland environment degradation identification and processing method based on remote sensing image monitoring;

[0016] Figure 5 This is a schematic diagram of the merging simulation of small water bodies in wetlands, which is a method for identifying and processing wetland environmental degradation based on remote sensing image monitoring.

[0017] Figure 6 This is a schematic diagram of vegetation cover merging simulation for a wetland environmental degradation identification and processing method based on remote sensing image monitoring. Detailed Implementation

[0018] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0020] Example 1

[0021] like Figure 1 As shown, Embodiment 1 of the present invention provides a method for wetland environmental degradation identification and processing based on remote sensing image monitoring, including the following operation steps:

[0022] Step S10: Acquire raw remote sensing images including multi-temporal data, the raw remote sensing images including raw remote sensing images of multiple bands; preprocess the raw remote sensing images to obtain preprocessed remote sensing images.

[0023] The preprocessing includes radiometric correction, geometric correction, and noise filtering;

[0024] It should be noted that the above-described embodiments of this application first utilize remote sensing technology to acquire original remote sensing images of a large area. These original remote sensing images include original images containing multiple temporal phases (images at different time points) and multiple bands (such as visible light, infrared, near-infrared, etc.), which can reflect the characteristics of wetlands at different times and under different spectra. The multi-temporal data ensures that the dynamic changes of the wetland environment over time can be captured, providing a basis for subsequent time-series analysis. At the same time, the multi-band characteristics of the original remote sensing images provide rich information for subsequent spectral feature extraction.

[0025] Furthermore, the raw remote sensing images undergo preprocessing. First, radiometric correction is performed to correct sensor response inconsistencies and enhance resistance to changes in atmospheric and lighting conditions, ensuring that the spectral information of the images accurately reflects the ground features. Next, geometric correction is performed to calibrate the raw remote sensing images to a unified geographic coordinate system, eliminating spatial distortions caused by sensor angles and platform jitter, and ensuring accurate registration of images from different time phases. Finally, noise filtering is performed by using spatial filtering or other noise reduction algorithms to reduce sensor noise and random interference in the images, improving the reliability of subsequent analysis.

[0026] Step S20: Extract texture features and spectral features from the pixels of the preprocessed remote sensing image; perform region growing segmentation on the preprocessed remote sensing image based on the texture features and spectral features combined with the texture boundary transition clarity value between regions in the image to obtain local region images;

[0027] It should be noted that the texture features in the above embodiments of this application are calculated by using multi-scale windows and statistical measures (such as mean, variance, entropy, etc.) to reveal the surface structure, roughness, and spatial heterogeneity of ground objects; the spectral features are based on multi-band reflectance data to extract key features such as the ratio between each band (such as NDVI, NDWI) and absolute reflectance value, reflecting the physical and chemical properties of ground objects; by extracting texture and spectral features from the preprocessed image, quantitative indicators are provided for subsequent accurate region segmentation and classification. The feature extraction results, through the combination of texture features and spectral features, enable subsequent image segmentation processing to take into account both details and overall structure;

[0028] Step S30: Based on the local area images, perform identification and analysis using a pre-trained wetland degradation detection model to obtain the type of each local area image;

[0029] The types of the aforementioned local area images include wetland degradation types and non-wetland degradation types (non-wetland degradation types include seasonal transient low water level areas, slightly disturbed but self-healing transitional areas, and ecological restoration transitional areas).

[0030] It should be noted that, in the above embodiments of this application, local area images of various time phases are input into the wetland degradation detection model, and the type of local area corresponding to each local area image is identified through comprehensive analysis of multiple time nodes.

[0031] In step S30 of the above-described embodiment of this application, the wetland degradation detection model analyzes four aspects of the local area image: spectral features, texture features, spatiotemporal dynamic analysis, and texture transition boundary sharpness value analysis, thereby outputting the type of the local area image, such as... Figure 2 and Figure 3As shown, specifically, in terms of spectral characteristics, wetland degradation is characterized by: a significant decrease in NDVI (Normalized Difference Vegetation Index), indicating changes in reflectance due to reduced vegetation cover (such as withering or salinization); abnormal NDWI (Normalized Difference Water Index), indicating that degraded wetlands may experience water drying or salinization, with NDWI values ​​significantly different from healthy water bodies; and fluctuating spectral reflectance, with increased dispersion of multi-band reflectance data in degraded wetland areas due to exposed land or pollutants.

[0032] Non-wetland degradation types are characterized as follows: in seasonally low water level areas, NDWI values ​​change cyclically with the season, but the overall spectral characteristics are consistent with those of natural water bodies; the spectral characteristics of slightly disturbed transitional areas are close to those of healthy wetlands, with small fluctuations in NDVI and NDWI values; and the spectra of transitional areas during ecological restoration show a gradual recovery trend (e.g., NDVI slowly increases).

[0033] In terms of texture characteristics, wetland degradation types exhibit high roughness and heterogeneity, with exposed land or fragmented vegetation leading to a high standard deviation in texture; abrupt spatial gradient changes, with human interference (such as excavation) or salinization boundaries potentially producing sharp gradients; non-wetland degradation types exhibit texture uniformity, with continuous vegetation or stable water areas having a lower standard deviation in texture and a gentler spatial gradient; natural gradual transitions, such as the boundary between reed beds and water bodies, where the texture trend vectors are aligned (with a small angle), conforming to the laws of natural gradation.

[0034] In terms of spatiotemporal (i.e., comprehensive analysis of texture and spectral data from multiple temporal phases) characteristics, wetland degradation types are characterized by: continuous changes, with multi-temporal data showing a continuous decline in NDVI or no recovery from NDWI anomalies; increased spatial heterogeneity, with degraded areas exhibiting gradually fragmented texture features in the time series; and non-degraded types are characterized by: periodic changes, with the spectral characteristics of seasonally low water levels changing in accordance with hydrological cycles; and a self-healing trend, with the spectral and texture features of slightly disturbed areas gradually recovering to a healthy state in subsequent temporal phases.

[0035] In terms of the manifestation of texture transition boundary clarity values, wetland degradation types exhibit high texture transition boundary clarity values. Due to internal heterogeneity or spectral abrupt changes, the texture transition boundary clarity values ​​exceed the threshold, suppressing merging and preserving them as independent degradation areas. Non-wetland degradation types exhibit low texture transition boundary clarity values. Natural transition areas (such as vegetation-water gradient zones) have texture trend vectors that are consistent (small angle) and spectral differences that are controllable, resulting in texture transition boundary clarity values ​​below the threshold, allowing them to merge into the same non-degraded area.

[0036] By analyzing the spectral features, texture features, spatiotemporal dynamics, and texture transition boundary clarity values ​​of local area images using the aforementioned wetland degradation detection model, if the spectral features are abnormal, the texture features are highly heterogeneous, and the texture transition boundary clarity value is high, then the local area image is determined to be of the wetland degradation type, and further monitoring and wetland degradation remediation measures are required for this area; conversely, if the spectral features conform to the natural cycle, the texture features are uniform, and the texture transition boundary clarity value is low, then the local area image is determined to be of the non-wetland degradation type, and only continuous observation is required.

[0037] This application first acquires raw remote sensing images including multi-temporal data, comprising raw remote sensing images of multiple bands. The raw remote sensing images are then preprocessed to obtain preprocessed remote sensing images. By capturing dynamic changes in the wetland environment through remote sensing images of different time points and different bands, the dynamic wetland environment is accurately reflected. Further, texture and spectral features are extracted from the pixels of the preprocessed remote sensing images. Based on the texture and spectral features, combined with the clarity values ​​of texture boundary transitions between different regions in the image, region growing segmentation is performed on the preprocessed remote sensing images to obtain local region images. This provides quantitative indicators for subsequent accurate region segmentation and classification. The feature extraction results, through the combination of texture and spectral features, enable subsequent image segmentation processing to consider both details and overall structure. Finally, based on the local region images, a pre-trained wetland degradation detection model is used for identification and analysis to obtain the type of each local region image, providing an accurate data foundation for the detection and remediation of wetland degradation.

[0038] Specifically, in step S20, based on the texture features and spectral features combined with the clarity values ​​of texture boundary transitions between regions in the image, region growing segmentation is performed on the preprocessed remote sensing image to obtain a local region image, including the following steps:

[0039] Step S21: Concatenate the texture features and spectral features of each pixel to obtain a high-dimensional feature vector (i.e., the high-dimensional feature vector of each pixel); perform principal component analysis on the high-dimensional feature vector to obtain a dimensionality-reduced feature vector (i.e., the dimensionality-reduced feature vector of each pixel).

[0040] It should be noted that, after concatenating the texture features and spectral features in the above embodiments of this application, the high-dimensional feature vector needs to be standardized by Z-score to obtain a standardized high-dimensional feature vector, thereby eliminating the difference in dimensions. Furthermore, principal component analysis is performed on the standardized high-dimensional feature vector to retain principal components with a contribution rate greater than 95%, thus solving the curse of dimensionality and retaining core information.

[0041] Step S22: Randomly select multiple seed points from the reduced feature vectors; based on the similarity d between the seed points and the reduced feature vectors of each neighboring pixel of the seed point;

[0042] Step S23: Merge the seed point and the neighboring pixels based on the similarity d to obtain an initial segmentation region; remove the seed point from all the dimensionality reduction feature vectors and repeat the above operation until all dimensionality reduction feature vectors have been traversed (i.e. all dimensionality reduction feature vectors have been merged) to obtain multiple initial segmentation regions.

[0043] Step S24: Based on all the initial segmented regions, the texture boundary transition clarity value of the initial segmented regions is analyzed through reasonable texture transition and then merged to obtain multiple local region images.

[0044] It should be noted that, in the above embodiments of this application, the texture features and spectral features of each pixel are first concatenated to form a high-dimensional feature vector. By combining texture and spectral information, the features of the image are comprehensively described. The high-dimensional feature vector is standardized by Z-score to eliminate the influence of different units and ensure that different features can be effectively compared. Principal component analysis (PCA) is performed on the standardized high-dimensional feature vector to reduce the dimensionality of the data while retaining the most important information, thereby improving the efficiency and accuracy of subsequent image segmentation processing.

[0045] Furthermore, multiple seed points are randomly selected from the dimensionality-reduced feature vectors to provide initial reference points for subsequent region segmentation. The similarity d between the seed points and their neighboring pixels' dimensionality-reduced feature vectors is calculated to prepare for region merging, provide a basis for segmentation, and help identify similar regions. Based on the calculated similarity d, the seed points and neighboring pixels are merged to form initial segmented regions, grouping pixels with similar features together to lay the foundation for generating larger regions. The processed seed points are removed, and the above operation is repeated from the remaining dimensionality-reduced feature vectors until all feature vectors have been processed, ensuring that every pixel is considered, thus forming a complete set of segmented regions. Furthermore, based on all initial segmented regions, a reasonable texture transition analysis is used for further merging to optimize the segmentation results, making the transition between adjacent regions more natural and reasonable, thereby improving the final image segmentation effect.

[0046] Region growing requires maintaining spatial connectivity of pixels or small regions. However, when faced with complex wetland boundaries, small water bodies, and discontinuous vegetation, connectivity breaks can easily occur, resulting in multiple parts of an image that originally belonged to the same local region being divided into multiple local region images (e.g., Figure 4(As shown in the figure). At the same time, in the process of region growth and boundary refinement, the boundaries of wetland environments are usually relatively continuous but have gradual changes, which can also cause over-segmentation of images, resulting in too many local area images. Many of these local area images actually belong to the same region. This processing will cause a waste of computing power or type recognition error in the subsequent wetland degradation detection model to identify and analyze the types of various local area images during the S30 implementation process.

[0047] Specifically, in step S24, based on the texture boundary transition clarity values ​​of all the initial segmented regions through texture reasonable transition analysis, they are merged again to obtain multiple local region images, including the following operation steps:

[0048] Step S241: Extract a set of adjacent pixel pairs based on the boundary lines of every two adjacent initial segmented regions;

[0049] It should be noted that the above-described embodiment of this application extracts a set of adjacent pixel pairs by extracting the boundary lines of two adjacent initial segmentation regions. Specifically, for two adjacent initial segmentation regions, initial segmentation region a and initial segmentation region b, pixels in initial segmentation region a that are in direct contact with initial segmentation region b are selected from the boundary lines of the two adjacent initial segmentation regions, and pixels in initial segmentation region b that are in direct contact with initial segmentation region a are selected. Adjacent pixel pairs are obtained by combining the positions of the pixels selected in initial segmentation region a and the positions of the pixels selected in initial segmentation region b. A set of adjacent pixel pairs is obtained based on all adjacent pixel pairs.

[0050] Step S242: Obtain the texture difference evaluation value of each adjacent pixel pair in the set of adjacent pixel pairs by calculating the Euclidean distance; obtain the texture difference evaluation value of the adjacent region boundary by averaging all the texture difference evaluation values ​​of adjacent pixel pairs.

[0051] Step S243: Analyze the texture features of each pixel in the initial segmentation region to obtain the texture trend vector of the initial segmentation region; calculate the texture transition coefficient based on the texture trend vector;

[0052] It should be noted that the texture trend vector obtained in the above embodiments of this application is obtained by statistically analyzing the local texture information of pixels in the initial segmentation region through a multi-scale window. Based on the statistical analysis of the local texture information, the texture roughness and local orientation information of the pixels can be obtained. Furthermore, the gradient direction of the neighborhood of each pixel in the region is calculated (for example, by using a differential operator to obtain the local principal direction), and the orientation angle of each pixel is recorded. These orientation data are used to construct an orientation histogram to determine the dominant texture direction in the region. Then, calculate the deviation of the orientation angle of all pixels from the dominant texture direction. (The deviation is calculated using circular standard deviation or variance; the lower the deviation, the more concentrated the texture direction.) Therefore, the texture trend vector is calculated based on the dominant texture direction and the deviation value. The above texture trend vector can be calculated as follows:

[0053] ; where (1–V) plays a weighting role, ensuring that when the internal consistency is high (V is small), the trend vector is "stronger".

[0054] Step S244: Calculate the texture boundary transition clarity value based on the texture difference evaluation value of the adjacent region boundary and the texture transition coefficient;

[0055] The calculation method for the texture boundary transition clarity value is as follows: ;

[0056] in, This represents the texture boundary transition clarity value between the initial segmented region a and the initial segmented region b. Let be the texture transition coefficient vector of the initial segmented region 'a'. Let be the texture transition coefficient vector of the initial segmented region b. The normalized spectral characteristics of the initial segmented region a, The normalized spectral characteristics of the initial segmented region b. Let a be the average spatial gradient of the initial segmented region a. Let b be the average spatial gradient of the initial segmented region. Let a be the standard deviation of the texture intensity of the initial segmented region a. Let b be the standard deviation of the texture intensity of the initial segmented region. This is an evaluation value for the difference in texture at the boundary of adjacent regions. These are the weighting coefficients. This is the value for determining texture trend consistency. This represents the normalized spectral feature similarity value. This represents the average spatial gradient similarity value. This is the value for determining the uniformity of texture.

[0057] In the above embodiments of this application, the texture trend consistency determination value (i.e.) is first used. Calculate the cosine angle of the texture transition coefficients of the initial segmented regions to quantify the spatial consistency of texture changes between the two regions. In wetlands, the boundary between vegetation cover (such as reed beds) and water usually exhibits a gradual transition, and its texture features (such as roughness and directionality) show a continuous spatial variation trend. If the texture transition vector directions of adjacent regions are close (small angle), the cosine value approaches 1, indicating that they belong to the same gradual process and should be merged preferentially. By measuring the vector angle rather than absolute difference, the influence of local texture fluctuations (such as shading by individual plants) on the overall trend judgment can be ignored, avoiding over-segmentation due to minor noise. Furthermore, through... (Normalized spectral feature similarity value) and The average spatial gradient similarity value supplements the shortcomings of texture analysis from the perspectives of physical and morphological attributes, forming a multi-dimensional feature constraint. Wetland degradation is often accompanied by vegetation withering (decreased NDVI) or water salinization (abnormal NDWI). Spectral differences can effectively distinguish between degraded areas and areas with seasonal changes (such as transient low water levels). The average spatial gradient difference reflects the sharpness of the boundary. Natural wetland boundaries (such as the water-land interface) usually have gentle gradients, while human disturbances (such as ditch digging) may produce abrupt gradients. When the gradient difference is too large, merging is suppressed to avoid confusing natural and human boundaries.

[0058] Furthermore, utilizing Introducing the transition effect of texture differences (T) at the boundaries of adjacent regions, and dynamically adjusting the merging threshold, small water bodies in wetlands often experience increased local texture differences (high T values) due to shadows or phytoplankton. However, if they are spatially adjacent and spectrally consistent (e.g., both are water bodies), this mechanism reduces the transition effect by decreasing the texture difference between adjacent regions. Value merging is allowed to avoid breakage. For the reed-mudflat gradient zone, although the overall texture difference is small, the boundary pixels may have a higher T value due to the mixed pixel effect. Texture transition compensation can avoid over-segmentation caused by local texture fluctuations; furthermore, (Texture uniformity consistency judgment value) By comparing the standard deviation of texture intensity, the uniformity difference of texture within the region is measured. The water area of ​​healthy wetland has a low standard deviation of texture (small s), while degraded wetland may show a high standard deviation (large s) due to siltation. If the standard deviation of adjacent areas is significantly different, it indicates that their internal structural heterogeneity is different and merging should be suppressed. The transitional area of ​​ecological restoration may contain patchy vegetation (medium s), which is significantly different from stable vegetation area (low s) or bare area (high s). This mechanism can avoid false merging and preserve the detailed features of the restoration process.

[0059] Step S245: If the texture boundary transition sharpness value is less than or equal to a preset texture boundary transition sharpness value threshold and the texture trend consistency judgment value is met... Less than or equal to the first threshold, normalized spectral feature similarity value Less than or equal to the second threshold, average spatial gradient similarity value Less than or equal to the third threshold, texture uniformity consistency judgment value When the difference in texture between adjacent regions is less than or equal to the fourth threshold and the difference in texture between adjacent regions is less than or equal to the fifth threshold, the two adjacent initial segmented regions are merged to obtain a local region image; the above operation is repeated until all initial segmented regions have been traversed to obtain multiple local region images.

[0060] It should be noted that the aforementioned texture boundary transition clarity value is a composite measurement method for quantifying the texture similarity between adjacent regions. Its core idea is that in remote sensing image segmentation, the textures of adjacent regions gradually transition with increasing spatial distance (for example, the boundary between wetlands and riparian vegetation gradually becomes clear as the spatial distance increases, rather than suddenly becoming clear). Based on this, when the texture boundary transition clarity value is less than or equal to the preset texture boundary transition clarity value threshold, it proves that the boundary between two adjacent initial segmented regions is not obvious and belongs to the form of gradual transition. They are more likely to be the same region because the region growing algorithm over-segmentes wetland boundaries, small water bodies, and discontinuous vegetation. Therefore, the two adjacent initial segmented regions should be merged into the same region. Conversely, the boundary between the two adjacent initial segmented regions is obvious and belongs to local region images of two different region types (such as the heterogeneous boundary between a wetland degradation type region and a non-wetland degradation type region, such as the boundary between salinization and vegetation areas). The two initial segmented regions should be identified as two local region images.

[0061] Based on the technical solutions adopted in the above embodiments of this application, by combining texture feature differences and spatial proximity, the region merging is dynamically adjusted, which is particularly suitable for handling complex scenarios with gradually changing boundaries of wetlands, water bodies, and other land features.

[0062] Based on the technical solution adopted in the above embodiments of this application, for the connection of small water bodies, the texture difference at the boundary of the broken water body caused by shadow (T=0.8, the fifth threshold is 1, 0.8 is less than 1, satisfying the above evaluation value of texture difference between adjacent regions less than or equal to the fifth threshold), and spectral similarity (normalized spectral feature similarity value). =0.1, the second threshold is 0.3, 0.1 is less than 0.3, satisfying the above-mentioned normalized spectral feature similarity value (less than or equal to the second threshold) and the gradient is gentle (mean spatial gradient similarity value). =0.3, the third threshold is 0.5, 0.3 is less than 0.5, satisfying the above requirement that the average spatial gradient similarity value is less than or equal to the third threshold). ≈0.17 (threshold 0.2, 0.17 is less than 0.2, satisfying the above texture boundary transition clarity value is less than or equal to the preset texture boundary transition clarity value threshold), texture uniformity consistency judgment value is 0.6, the fourth threshold is 0.9, 0.6 is less than 0.9, satisfying the above texture uniformity consistency judgment value. If the value is less than or equal to the fourth threshold, the merging is successful;

[0063] In summary, the texture boundary transition clarity value adopted in the embodiments of this application achieves adaptive merging control of complex wetland landform boundaries by fusing texture trend vectors, spectra, gradients and regional uniformity features, and introducing a dynamic transition merging mechanism for boundary differences. This closely combines the needs of the universality of gradual transition, landform fragmentation and multi-temporal analysis in wetland degradation monitoring, reduces over-segmentation while retaining key degradation features, and provides a high-precision segmentation basis for subsequent model recognition.

[0064] For example: the connection of small water bodies in wetlands: such as Figure 5 As shown, small, broken bodies of water may exhibit texture differences due to shadows or vegetation obstruction, but when they are spatially adjacent, the clarity value of the texture boundary transition will lower its merging threshold, thus preventing breakage.

[0065] Gradual transition of vegetation cover: such as Figure 6 As shown, the boundary between reed beds and open water may have small texture differences, but the space is continuous. By calculating and analyzing the clarity value of the texture boundary transition, over-segmentation caused by texture fluctuations can be avoided.

[0066] The texture boundary transition clarity value in the above embodiments of this application dynamically corrects the "absoluteness" of the traditional texture difference measurement by introducing the texture transition effect of spatial distance, making it more in line with the spatial distribution pattern of ground objects in natural scenes. In complex environments such as wetlands, it can avoid over-segmentation caused by texture fluctuations and suppress the erroneous merging of spatially disconnected areas.

[0067] Specifically, in step S243, the texture transition coefficient is calculated based on the texture trend vector, including the following steps:

[0068] Step S2431: Construct a texture trend vector matrix P based on the texture trend vector in the initial segmentation region;

[0069] It should be noted that in remote sensing wetland monitoring, different pixels may exhibit local differences due to vegetation, shallow water areas, or bare land. Constructing a texture trend vector matrix P aggregates the local texture information of each pixel, providing a data foundation for subsequent statistical analysis of the overall dominant texture direction within the region.

[0070] Step S2432: Calculate the covariance matrix between the texture trend vector of the i-th pixel and the texture trend vector of the j-th pixel in the initial segmentation region based on the texture trend vector matrix P. ;

[0071] The covariance matrix The calculation method is as follows: ;

[0072] In the formula, This represents the total number of pixels in the initial segmented region a; This is the transpose of the texture trend vector matrix P; The covariance between the texture trend vector of the i-th pixel and the texture trend vector of the j-th pixel in the initial segmentation region (where, in the covariance matrix...) In this diagram, the diagonal elements represent the variance of each dimension, while the off-diagonal elements describe the correlation between them.

[0073] It should be noted that in complex wetlands, different areas may exhibit continuous texture changes along a certain dominant direction. By capturing the statistical distribution of texture trends within a region through the covariance matrix, a mathematical basis is provided for extracting the overall texture direction of the region (i.e., texture transition coefficient), enabling the system to suppress local noise or irregular distribution interference.

[0074] Step S2433: Initially obtain the first texture transition coefficient based on the texture trend vector. ;

[0075] The texture transition coefficient The calculation method is as follows: ;

[0076] It should be noted that in wetland environments, different small areas may exhibit texture fluctuations. Taking the mean and normalizing can be seen as a preliminary "averaging" of multiple uncertain factors, providing a reasonable starting point for subsequent iterative feedback to determine a unified texture orientation within the region.

[0077] Step S2434: Based on the first texture transition coefficient Through the covariance matrix After iterative updates and correction using the spectral features, the texture transition coefficient is output. .

[0078] It should be noted that the above-described embodiments of this application first construct a texture trend vector matrix P based on the texture trend vector in the initial segmented region, integrating the local texture information of each pixel to lay the foundation for subsequent analysis of the main direction of the overall texture in the region. In remote sensing wetland monitoring, due to the presence of different land features (such as vegetation, shallow water areas, and bare land), the texture information between pixels may have significant differences. Constructing the texture trend vector matrix P can effectively capture and characterize these local differences. Through the texture trend vector matrix P, the covariance matrix of the texture trend vectors of any two points (the i-th and j-th pixels) in the initial segmented region is calculated, thereby reflecting the statistical relationship of texture trends between pixels and capturing the texture change patterns within the region. The diagonal elements in the covariance matrix show the variance of each dimension, while the off-diagonal elements reveal... The correlation between them is understood. Therefore, the covariance matrix provides the necessary mathematical basis for extracting the overall texture direction of the region (the ultimate goal is the texture transition coefficient), suppressing errors caused by local noise or irregular distribution. Based on the texture trend vector, the first texture transition coefficient is initially calculated to "average" the texture fluctuations in the region, providing a reasonable starting point so that subsequent steps can iteratively feedback and accurately determine the unified texture direction of the region. Based on the first texture transition coefficient, iterative updates are performed in combination with the information of the covariance matrix, and corrections are made through spectral features. The final output is the texture transition coefficient. By introducing more features and information, the accuracy of the texture transition coefficient is significantly improved, making the final result more robust. By effectively integrating the features of texture and spectrum, reliable data support is provided for remote sensing monitoring of wetlands.

[0079] During the execution of step S2434 in the above embodiments of this application, due to the first texture transition coefficient The calculation may be affected by a mismatch between texture orientation and spectral features (such as NDVI) (e.g., withered vegetation and healthy vegetation have similar textures but different spectra), thus affecting the first texture transition coefficient. The calculation results deviated, therefore, for the first texture transition coefficient Further corrections are needed using spectral features to obtain a texture transition coefficient that accurately reflects the regional characteristics and is constrained by the spectrum. For details, please refer to the execution process of steps S24341-S24345.

[0080] Specifically, in step S2434, based on the first texture transition coefficient Through the covariance matrix After iterative updates and correction using the spectral features, the texture transition coefficient is output. The operation includes the following steps:

[0081] Step S24341: Initialize iteration parameters; the iteration parameters include an iteration counter and a maximum iteration threshold, and a texture transition coefficient convergence judgment threshold; the iteration count of the iteration counter is initially 0;

[0082] Step S24342: Increment the iteration count of the iteration counter by 1 to obtain the current iteration count t; based on the first texture transition coefficient With the covariance matrix The second texture transition coefficient was calculated. ;

[0083] Second texture transition coefficient The calculation method is as follows: = × ;

[0084] Step S24343: Based on the first texture transition coefficient With the second texture transition coefficient The convergence judgment value of the texture transition coefficient was calculated. ;

[0085] The convergence judgment value of the texture transition coefficient The calculation method is as follows: ;

[0086] Step S24344: Determine the convergence value of the texture transition coefficient. If the texture transition coefficient is less than the convergence threshold, then output the second texture transition coefficient. For the texture transition coefficients to be determined If not, then further determine whether the number of iterations has reached the maximum number of iterations threshold; if yes, then output the second texture transition coefficient. For the texture transition coefficients to be determined If not, then adjust the second texture transition coefficient. Returning to the above operation (i.e., adjusting the second texture transition coefficient) Return as the first texture transition coefficient mentioned above. Then repeat steps S24342-S24344 above), and iterate again until the output yields the texture transition coefficients to be determined. ;

[0087] Step S24345: Based on the texture transition coefficient to be determined The target texture transition coefficient is obtained by combining the spectral features.

[0088] It should be noted that, based on the first texture transition coefficient With the covariance matrix The second texture transition coefficient was calculated. During execution, the method is used to gradually suppress local noise and strengthen the main texture change trend within the region. For example, for the gradual boundary between reed beds and water in wetlands, the iterative results can accurately obtain the dominant change direction, thus making subsequent region merging and degradation detection more accurate. Specifically, the dominant texture direction of the region is first strengthened by iteratively using the covariance matrix to suppress the influence of local noise (such as occlusion by individual plants). For example, in the reed bed region, the interference direction of a small number of withered plants is corrected by the principal components of the covariance matrix, and the overall direction is closer to the trend of healthy vegetation.

[0089] Furthermore, when step S24345 is executed again, spectral correction is used to ensure that the texture transition coefficient is consistent with physical properties (such as vegetation health and water distribution). For example, in salinized areas, due to spectral abnormalities (low NDVI), even if the texture direction is similar to that of healthy soil, the transition coefficient will be lowered to prevent erroneous merging.

[0090] Specifically, in step S24345, based on the texture transition coefficient to be determined... The target texture transition coefficient is obtained by modifying the spectral features, including the following steps:

[0091] Step S243451: Based on the normalized spectral characteristics of the initial segmented region a The spectral consistency factor is calculated.

[0092] The spectral consistency factor is calculated as follows: ;

[0093] In the formula, The average global spectral features of the initial segmented region a;

[0094] Step S243452: Based on the spectral consistency factor, determine the texture transition coefficient to be determined. Corrections are made to obtain the target texture transition coefficients. ;

[0095] The target texture transition coefficient The calculation method is as follows: = × ;

[0096] It should be noted that wetland areas typically exhibit relatively uniform reflectance characteristics in the spectrum, while degraded areas may show inconsistent spectral variations. Using spectral information to correct the texture transition coefficient ensures that the transition coefficient fully reflects regional continuity only when the texture variation is consistent with the spectral physical properties.

[0097] In other words, if the spectrum within the region matches the overall situation (such as healthy water bodies or continuous reed distribution), then ≈1, preserving the original texture transition direction;

[0098] If abnormalities occur (such as spectral abrupt changes caused by salinization or vegetation withering). This will reduce the texture transition effect in that area, thus preventing incorrect area merging;

[0099] Further explanation: In the first texture transition coefficient The inconsistency between spectrum and texture leads to errors in the calculation of texture transition coefficient. This application's embodiments introduce a spectral consistency factor and incorporate it into the calculation of texture transition coefficient to ensure that spectrum and texture jointly guide the merging decision.

[0100] Furthermore, through the corrected target texture transition coefficient The result of adjusting the texture boundary transition clarity value is to suppress erroneous merging operations (i.e., when the texture directions of salinized areas and healthy soils are similar, the spectral consistency factor is used to correct for erroneous merging operations).

[0101] In the specific implementation process of the above embodiments of this application, those skilled in the art discovered that step S244 in the above embodiments of this application... Normalized spectral similarity values ​​are typically obtained by analyzing spectral features extracted from raw remote sensing images acquired during clear weather. However, in foggy weather or when there is perspective interference such as clouds, fog, or rain in the atmosphere, the uneven distribution of this perspective interference can lead to significant errors in the extraction of spectral features. Specifically, fog diffuses the distribution of spectral reflectance of ground objects, significantly increases the gray variance of the boundary neighborhood, and causes fog scattering to increase overall brightness. The mean gray value of the neighborhood deviates from the true value in the fog-free state. As a result, areas that are originally significantly different (such as degraded wetlands and healthy vegetation) may have their spectral similarity underestimated due to fog obscuring them, affecting the accuracy of the normalized spectral similarity values ​​and thus influencing the subsequent decision on whether to merge or not.

[0102] Specifically, in step S244, the analysis and acquisition of the normalized spectral feature similarity value includes the following steps:

[0103] Step S2441: Analyze the spectral characteristics of the current adjacent initial segmentation regions a and b to obtain the normalized spectral feature similarity value to be corrected. )';According to the historical remote sensing image database, obtain the boundary neighborhood gray values ​​of similar land features under the L-neighborhood window size in a fog-free state;

[0104] It should be noted that the above-described embodiments of this application extract the boundary neighborhood gray values ​​of various similar land features (such as wetland vegetation, water bodies, and saline-alkali soil) from remote sensing images in a fog-free state in the historical remote sensing image database, which serve as the data basis for subsequent fog-affected offset adjustment of normalized spectral feature similarity values.

[0105] Step S2442: Calculate the variance and mean of the gray values ​​of the boundary neighborhood based on the gray values ​​of the boundary neighborhood. ;

[0106] It should be noted that in the above embodiments of this application, by calculating the variance of the gray value of the boundary neighborhood and the mean of the gray value of the boundary neighborhood for each of the same type of land cover, a fog-free benchmark is established. In this way, when fog occurs, the normalized spectral feature similarity value is effectively adjusted, thereby improving the accuracy of the decision on whether to merge the subsequent initial segmented areas.

[0107] Step S2443: Based on the average gray value of the target boundary neighborhood of similar land features in the currently obtained remote sensing image. The average gray value of the boundary neighborhood Calculate the mean offset Based on the maximum value of the boundary neighborhood grayscale value variance. and the average gray value of the boundary neighborhood The fog diffusion factor was calculated.

[0108] The fog diffusion factor is calculated as follows: ;

[0109] In the formula, This represents the variance of grayscale values ​​in the neighborhood of similar land features within the current remote sensing image. The maximum variance of gray values ​​in the boundary neighborhood (obtained from historical remote sensing images); Mean offset (the deviation between the mean gray value of the boundary neighborhood obtained from historical remote sensing image analysis and the mean gray value of the target boundary neighborhood in the current remote sensing image). The average gray value of the boundary neighborhood; These are the weighting coefficients;

[0110] Step S2444: Based on the fog diffusion factor, adjust the normalized spectral feature similarity value to be corrected. After correction, the target normalized spectral feature similarity value is obtained. ;

[0111] The target normalized spectral feature similarity value The calculation method is as follows: =( )'×(1+ );

[0112] In the formula, The size of the neighborhood window;

[0113] It should be noted that the above embodiments of this application obtain the normalized spectral feature similarity value to be corrected by first obtaining the spectral features of the initial segmented region a and the initial segmented region b under the current remote sensing image and then analyzing them. Because it is currently unknown whether there is any perspective interference such as fog (analyzing perspective interference for every remote sensing image is computationally too intensive; therefore, the normalized spectral feature similarity value to be corrected is obtained later). Afterwards, a second correction is performed to ensure accurate results while significantly reducing the computational burden of perspective interference analysis. Based on historical remote sensing imagery, the mean and maximum variance of the gray values ​​of the boundary neighborhoods under the same neighborhood window size for the same ground feature are obtained. This provides the necessary data foundation for subsequent analysis of the fog diffusion factor. Then, by analyzing the mean and variance of the gray values ​​of the target boundary neighborhoods under the same neighborhood window size in the current remote sensing image, the deviation between the mean gray values ​​of the boundary neighborhoods obtained from historical remote sensing imagery analysis and the mean gray values ​​of the target boundary neighborhoods in the current remote sensing image is analyzed. This further calculates the fog diffusion factor. Based on the deviation between the baseline value and the current value, the fog diffusion factor is analyzed, and the fog diffusion factor obtained from this deviation is used to correct the current normalized spectral feature similarity value to be corrected. This yields the target normalized spectral feature similarity value. This avoids spectral interference in the current remote sensing image caused by perspective interference such as clouds, rain, and fog, thereby improving the accuracy of subsequent decisions on whether to merge the initial segmented region a and the initial segmented region b.

[0114] Example 2

[0115] Embodiment 2 of the present invention provides a method for wetland environment degradation identification and processing based on remote sensing image monitoring, including the following operation steps:

[0116] Step S10: Acquire raw remote sensing images including multi-temporal data, the raw remote sensing images including raw remote sensing images of multiple bands; preprocess the raw remote sensing images to obtain preprocessed remote sensing images.

[0117] Step S20: Extract texture features and spectral features from the pixels of the preprocessed remote sensing image; based on the texture features and spectral features, combined with the clarity value of the texture boundary transition between regions in the image, segment the preprocessed remote sensing image to obtain local region images;

[0118] Step S30: Based on the local area images, perform identification and analysis using a pre-trained wetland degradation detection model to obtain the type of each local area image;

[0119] The aforementioned local area images include wetland degradation types and non-wetland degradation types;

[0120] Meanwhile, the implementation content of other steps S21-S24 in Embodiment 2 of this application is the same as that in Embodiment 1 above. The difference is that Embodiment 2 of this application continues to propose another scheme for the over-segmentation problem caused by the region growing algorithm in step S24, which is different from the above merging processing method. It is based on the texture boundary transition clarity value of all the initial segmented regions through texture reasonable transition analysis of the initial segmented regions to obtain multiple local region images.

[0121] Specifically, regarding the oversegmentation problem caused by the region growing algorithm in the above embodiment 1, since the initial segmented regions obtained through the above segmentation process may be regions with a very large amount of data, the above analysis and merging process based on the texture boundary transition clarity value of adjacent regions is not applicable when there are a large number of initial segmented regions. It is necessary to calculate the texture boundary transition clarity value for the neighborhood (four-neighborhood or eight-neighborhood) of each initial segmented region, which will bring a huge system computing power pressure.

[0122] Specifically, in step S24, based on the texture boundary transition clarity values ​​of all the initial segmented regions through texture reasonable transition analysis, they are merged again to obtain multiple local region images, including the following operation steps:

[0123] Step S241': Establish a grid index based on the position of each initial segmented region; calculate the necessary merging parameters for each initial segmented region based on the grid index;

[0124] The necessary parameters for merging include the texture transition coefficient vector, normalized spectral features, average spatial gradient, and texture intensity standard deviation.

[0125] It should be noted that in the embodiments of this application described above, firstly, a grid index for each initial segmented region is established based on its location, and index information is marked for it. Then, based on each index information, necessary parameters for merging each initial segmented region are calculated. These necessary parameters are crucial calculation parameters in the subsequent merging analysis process, as the texture transition coefficient vector, normalized spectral features, average spatial gradient, and texture intensity standard deviation are all very important. They play a vital role in whether the initial segmented regions are merged. Among them, spectral features are the direct basis for distinguishing land cover categories (e.g., NDVI is used for vegetation detection, and NDWI is used for water body identification). Normalization processing can eliminate the influence of differences in illumination and sensor response on spectral values, ensuring the spectral comparability between different regions. During the merging process, spectral similarity is the basis for determining whether regions belong to the same category. For the average spatial gradient, natural wetland boundaries (such as vegetation-water transition zones) typically have gentle gradients, while areas affected by human disturbance or degradation (such as ditches and salinization boundaries) have abrupt gradients. Gradient differences can distinguish between natural transitions and human / degraded boundaries, avoiding the erroneous merging of sharp boundaries. Texture intensity standard deviation can measure the uniformity of texture within a region. Healthy wetlands (such as continuous vegetation or stable water bodies) have uniform textures, while degraded areas (such as fragmented vegetation and salinization patches) have high texture heterogeneity. Texture intensity standard deviation identifies regions with inconsistent internal structures, preventing the erroneous merging of heterogeneous regions. Texture transition coefficient vectors provide directional information about texture trends, supplementing the deficiencies of spectral and texture uniformity analysis, and ensuring the correct merging of natural gradient boundaries.

[0126] Furthermore, during the subsequent merging process, the aforementioned necessary merging parameters will be called multiple times. Real-time calculation would place a greater burden on computing power. By using the "pre-calculation" in this application, the necessary merging parameters are used as a "feature pool" and directly called during subsequent analysis and calculation, thereby improving the overall analysis efficiency.

[0127] Step S242': Based on the necessary merging parameters, calculate the jump difference value between every two grid indices. ;

[0128] The jump difference value The calculation method is as follows: ;

[0129] In the formula, Let Euclidean distance be the normalized spectral features of any two grid indices a and b. This represents the normalized difference value of the standard deviation of texture intensity;

[0130] Step S243': Determine the jump difference value Is it less than or equal to a preset transition difference threshold? If yes (if no, then no further processing is performed, because the initial segmentation regions corresponding to the two mesh indices are clearly not the same local segmentation region, reducing the number of adjacent initial segmentation region pairs for subsequent calculation of texture boundary transition clarity values), then the transition difference value is... The two corresponding grid indices are marked as potential mergeable region pairs;

[0131] It should be noted that, in the above embodiments of this application, the jump difference value A threshold for abrupt changes in the difference between two regions is less than or equal to the preset threshold. This indicates that the two regions are similar in spectral and textural uniformity and may belong to the same category (e.g., different sub-regions of a healthy wetland). Conversely, a threshold that is greater than or equal to the threshold indicates that the two regions are significantly different (e.g., a degraded area and a healthy area) and do not need to be merged. Specifically, spectral features and textural intensity standard deviation are the core characterization parameters of wetland degradation. If the two regions are similar in both aspects, even if there are slight differences in the textural transition coefficient vector or the average spatial gradient, they may belong to the same land cover category (e.g., a gradual boundary caused by seasonal water level changes). Therefore, prioritizing the merging of these regions can preferentially screen out some regions that are extremely unlikely to be merged, thereby reducing the computational burden on subsequent merging processes.

[0132] Step S244': Calculate the angle between the texture trend vectors of the potential mergeable region pairs based on the texture transition coefficient vector. Determine the angle between the texture trend vectors. Whether it is less than or equal to the texture trend vector angle threshold; if so, then the mergeable region pair is determined as a homogeneous region group;

[0133] It should be noted that, in the above embodiments of this application, the included angle of the texture trend vector... When the angle between the texture trend vectors is less than or equal to the threshold, it indicates that the texture directions of the two regions are highly consistent (such as the gradient boundary between reed beds and water), belonging to a natural transition area. In this case, they are directly merged into a homogeneous region group. For example, if the angle between the texture trend vectors of region A (reed beds) and region B (water) is 10°, they are determined to be the same gradient process and merged into a homogeneous group.

[0134] Step S245': Calculate the texture boundary transition sharpness value for the two initial segmented regions corresponding to the homogeneous region group; merge the homogeneous region group based on the texture boundary transition sharpness value to obtain a local region image;

[0135] It should be noted that after merging the homogeneous region group in step S244' above, the texture direction of the two initially segmented regions is consistent, but other features (such as spectrum, gradient, texture uniformity) may still be different. Further verification is required by the texture boundary transition clarity value mentioned above. For example, region A (healthy vegetation) and region B (slightly degraded vegetation) in the homogeneous group may have the same texture direction, but the spectrum or gradient is significantly different. It is necessary to determine whether to merge by judging the texture boundary transition clarity value.

[0136] The above method of calculating the texture boundary transition clarity value of two initial segmented regions corresponding to the homogeneous region group and merging the homogeneous region group according to the texture boundary transition clarity value to obtain the local region image to be determined is completely consistent with the method of Embodiment 1 above, and will not be repeated here. The difference is that in Embodiment 2 of this application, the homogeneous region group is not necessarily two adjacent initial segmented regions. Instead, after the operation of steps S241' to S244' above, even if the two initial segmented regions are not adjacent, they may still be homogeneous region groups of the same local region image. Compared with the method of calculating the texture boundary transition clarity value of each two adjacent initial segmented regions in Embodiment 1 above, it will reduce the amount of texture boundary transition clarity value calculation by a large margin, thereby significantly improving the merging efficiency.

[0137] In summary, the wetland environmental degradation identification and processing method based on remote sensing image monitoring proposed in this invention first acquires raw remote sensing images of a large area using remote sensing technology, covering image data at different time points and under different spectra (such as visible light and infrared). This multi-temporal, multi-band dataset provides a solid foundation for subsequent analysis of wetland dynamic changes. Furthermore, preprocessing operations correct the inhomogeneity of sensor response, enhancing the image's resistance to environmental factors (such as atmospheric and illumination variations), thereby ensuring that spectral information accurately reflects ground features. The images are corrected to a unified geographic coordinate system, eliminating spatial distortion caused by sensor angle changes and platform jitter, ensuring accurate registration of images from different time phases. Additionally, noise reduction algorithms such as spatial filtering reduce noise and random interference in the images, improving data quality and the reliability of subsequent analysis. Furthermore, by utilizing multi-scale windows and statistical measures (such as mean square root), the method further enhances the wetland's environmental degradation identification and processing capabilities. The values, variances, and entropies are calculated to reveal the structure and roughness of the ground surface, reflecting spatial heterogeneity. Based on multi-band reflectance data, key features such as NDVI (Normalized Difference Vegetation Index) and NDWI (Normalized Difference Water Index) are extracted. Then, the extracted texture and spectral features are combined to segment local area images, effectively taking into account both image details and overall structure, providing an accurate foundation for subsequent classification. Furthermore, through the analysis of local area images, a pre-trained wetland degradation detection model is used for identification to obtain the type of each local area, such as seasonal low water level areas and slightly disturbed transition areas, providing an accurate basis for subsequent wetland degradation treatment.

[0138] Furthermore, in the further processing, in response to the oversegmentation of the regional growth process, this application embodiment introduces a texture boundary transition clarity value by fusing texture trend vectors, spectra, gradients and regional uniformity features, and introduces a dynamic transition merging mechanism for boundary differences, to achieve adaptive merging control of complex wetland landform boundaries. This closely combines the universality of gradual transition, landform fragmentation and multi-temporal analysis requirements in wetland degradation monitoring, reducing oversegmentation while retaining key degradation features, and providing a high-precision segmentation basis for subsequent model recognition.

[0139] Furthermore, this application embodiment also optimizes the over-segmentation problem in the region growing process by introducing jump difference values, thereby reducing computational pressure and improving merging efficiency.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; those skilled in the art can modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying and processing wetland environmental degradation based on remote sensing image monitoring, characterized in that, The following steps are included: Acquire raw remote sensing images that include multi-temporal data, the raw remote sensing images including raw remote sensing images of multiple bands; preprocess the raw remote sensing images to obtain preprocessed remote sensing images; Texture and spectral features are extracted from the pixels of the preprocessed remote sensing image; Based on the texture features and spectral features, combined with the clarity values ​​of texture boundary transitions between regions in the image, region growing segmentation is performed on the preprocessed remote sensing image to obtain local region images. Based on the local area images, a pre-trained wetland degradation detection model is used for identification and analysis to obtain the type of each local area image; The process of performing region growing segmentation on the preprocessed remote sensing image based on the texture features, the spectral features, and the clarity values ​​of texture boundary transitions between regions in the image to obtain local region images includes the following steps: The texture features and spectral features of each pixel are concatenated to obtain a high-dimensional feature vector; principal component analysis is then performed on the high-dimensional feature vector to obtain a dimensionality-reduced feature vector. Multiple seed points are arbitrarily selected from the dimensionality-reduced feature vectors; the similarity d between the seed points and their neighboring pixels is used to obtain the seed points. Based on the similarity d, the seed point and the neighboring pixels are merged to obtain an initial segmentation region; the seed point is screened out from all the dimensionality reduction feature vectors and the above operation is repeated until all the dimensionality reduction feature vectors are traversed to obtain multiple initial segmentation regions; Based on all the initial segmented regions, the texture boundary transition clarity value between regions is calculated, and then the regions are merged again based on the texture boundary transition clarity value to obtain multiple local region images; The process of calculating the texture boundary transition sharpness value between regions based on all the initial segmented regions, and then merging them based on the texture boundary transition sharpness value to obtain multiple local region images includes the following steps: A set of adjacent pixel pairs is extracted based on the boundary lines of every two adjacent initial segmentation regions; The texture features of each adjacent pixel pair in the set of adjacent pixel pairs are obtained by calculating the Euclidean distance to obtain the texture difference evaluation value of the adjacent pixel pair; the texture difference evaluation value of the adjacent region boundary is obtained by averaging all the texture difference evaluation values ​​of the adjacent pixel pairs. The texture trend vector of the initial segmented region is obtained by analyzing the texture features of each pixel in the initial segmented region; the texture transition coefficient is then calculated based on the texture trend vector. The texture boundary transition clarity value is calculated based on the evaluation value of the texture difference between adjacent regions and the texture transition coefficient. The calculation method for the texture boundary transition clarity value is as follows: ； in, This represents the texture boundary transition clarity value between the initial segmented region a and the initial segmented region b. Let be the texture transition coefficient vector of the initial segmented region 'a'. Let be the texture transition coefficient vector of the initial segmented region b. The normalized spectral characteristics of the initial segmented region a, The normalized spectral characteristics of the initial segmented region b. Let a be the average spatial gradient of the initial segmented region a. Let b be the average spatial gradient of the initial segmented region. Let a be the standard deviation of the texture intensity of the initial segmented region a. Let b be the standard deviation of the texture intensity of the initial segmented region. This is an evaluation value for the difference in texture at the boundary of adjacent regions. These are the weighting coefficients. This is the value for determining texture trend consistency. This represents the normalized spectral feature similarity value. This represents the average spatial gradient similarity value. This is the value for determining the uniformity of texture. If the texture boundary transition sharpness value is less than or equal to a preset texture boundary transition sharpness value threshold and the texture trend consistency judgment value is met... Less than or equal to the first threshold, normalized spectral feature similarity value Less than or equal to the second threshold, average spatial gradient similarity value Less than or equal to the third threshold, texture uniformity consistency judgment value When the evaluation value of the texture difference between adjacent regions is less than or equal to the fourth threshold and the evaluation value of the boundary texture difference between adjacent regions is less than or equal to the fifth threshold, the two adjacent initial segmented regions are merged to obtain a local region image; the above operation is repeated until all initial segmented regions have been traversed to obtain multiple local region images. The process of calculating the texture boundary transition sharpness value between regions based on all the initial segmented regions, and then merging them based on the texture boundary transition sharpness value to obtain multiple local region images, also includes the following steps: A grid index is established based on the location of each initial segmentation region; the necessary parameters for merging each initial segmentation region are calculated based on the grid index. The necessary parameters for merging include the texture transition coefficient vector, normalized spectral features, average spatial gradient, and texture intensity standard deviation. Based on the aforementioned necessary merging parameters, calculate the jump difference value between every two grid indices. ; Determine the jump difference value Is it less than or equal to a preset transition difference threshold? If so, then set the transition difference value... The two corresponding grid indices are marked as potential mergeable region pairs; The texture trend vector angle of the potential mergeable region pair is calculated based on the texture transition coefficient vector. Determine the angle between the texture trend vectors. Whether it is less than or equal to the texture trend vector angle threshold; if so, then the mergeable region pair is determined as a homogeneous region group; The texture boundary transition clarity value is calculated for the two initial segmented regions corresponding to the homogeneous region group; the homogeneous region group is merged based on the texture boundary transition clarity value to obtain a local region image.

2. The wetland environment degradation identification and processing method based on remote sensing image monitoring according to claim 1, characterized in that, The process of analyzing and calculating the texture transition coefficient based on the texture trend vector includes the following steps: A texture trend vector matrix P is constructed based on the texture trend vector in the initial segmentation region; Based on the texture trend vector matrix P, the covariance matrix between the texture trend vector of the i-th pixel and the texture trend vector of the j-th pixel in the initial segmentation region is calculated. ; The first texture transition coefficient is initially obtained based on the texture trend vector. ; Based on the first texture transition coefficient Through the covariance matrix After iterative updates and correction using the spectral features, the texture transition coefficient is output. .

3. The wetland environment degradation identification and processing method based on remote sensing image monitoring according to claim 2, characterized in that, Based on the first texture transition coefficient Through the covariance matrix After iterative updates and correction using the spectral features, the texture transition coefficient is output. The operation includes the following steps: Initialize the iteration parameters; the iteration parameters include an iteration counter and a maximum iteration threshold, and a texture transition coefficient convergence judgment threshold; the iteration count of the iteration counter is initially set to 0; Increment the iteration count of the iteration counter by 1 to obtain the current iteration count t; based on the first texture transition coefficient With the covariance matrix The second texture transition coefficient was calculated. ; Based on the first texture transition coefficient With the second texture transition coefficient The convergence judgment value of the texture transition coefficient was calculated. ; The convergence judgment value of the texture transition coefficient is determined iteratively. The texture transition coefficient to be determined is output when it is less than the convergence threshold or the number of iterations reaches the maximum number of iterations threshold. ; Based on the texture transition coefficient to be determined The target texture transition coefficient is obtained by combining the spectral features.

4. The wetland environment degradation identification and processing method based on remote sensing image monitoring according to claim 3, characterized in that, The texture transition coefficient to be determined The target texture transition coefficient is obtained by modifying the spectral features, including the following steps: Based on the normalized spectral characteristics of the initial segmented region a The spectral consistency factor is calculated. Based on the spectral consistency factor, the texture transition coefficient to be determined is... Corrections are made to obtain the target texture transition coefficients. .

5. The wetland environment degradation identification and processing method based on remote sensing image monitoring according to claim 4, characterized in that, The analysis and acquisition of the normalized spectral feature similarity values ​​includes the following steps: The similarity value of the normalized spectral features to be corrected is obtained by analyzing the spectral characteristics of the current adjacent initial segmentation regions a and b. )';According to the historical remote sensing image database, obtain the boundary neighborhood gray values ​​of similar land features under the L-neighborhood window size in a fog-free state; The variance and mean of the gray values ​​in the boundary neighborhood are calculated based on the gray values ​​of the boundary neighborhood. ; Based on the average gray value of the target boundary neighborhood of similar land features in the currently obtained remote sensing images The average gray value of the boundary neighborhood Calculate the mean offset Based on the maximum value of the boundary neighborhood grayscale value variance. and the average gray value of the boundary neighborhood The fog diffusion factor was calculated. Based on the fog diffusion factor, the normalized spectral feature similarity value to be corrected ( After correction, the target normalized spectral feature similarity value is obtained. .

6. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the wetland environmental degradation identification and processing method based on remote sensing image monitoring as described in any one of claims 1-5.

Images