Wetland area state monitoring method and system based on image enhancement

Abstract

The invention belongs to the technical field of image recognition, and particularly relates to a wetland area state monitoring method and system based on image enhancement, and the method comprises the steps: carrying out the preprocessing of an obtained remote sensing image, and obtaining a water body image of a target wetland area; for each pixel point in the water body region, constructing a spatial spectrum joint feature vector capable of representing spectral characteristics and spatial aggregation characteristics based on the local gradient entropy and the neighborhood spectral dispersion of the neighborhood of the pixel point; independent component analysis is carried out on the spatial spectrum joint feature vectors of all the pixel points, and algae component images are identified and separated according to statistical distribution features; and post-processing the separated algae component image to monitor the wetland water body state. According to the method, the feature vector fusing the spatial context information is constructed, so that the extraction capability and the anti-noise performance of the early-stage weak water bloom signal are remarkably enhanced, and the spatial continuity and the accuracy of a monitoring result are ensured.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology. More specifically, this invention relates to a method and system for monitoring the state of wetland areas based on image enhancement. Background Technology

[0002] Wetlands are extremely important ecosystems, and their health has a decisive impact on biodiversity, water resource security, and regional ecological balance. Large-scale wetland monitoring using remote sensing images acquired from platforms such as satellites has become a key technical means in modern environmental science research. Eutrophication, as a core indicator of wetland ecosystem degradation, often initially manifests as trace accumulations of phytoplankton such as cyanobacteria, forming weak algal blooms on the water surface. Timely detection and early warning at this stage are crucial for subsequent ecological restoration.

[0003] Currently, Independent Component Analysis (ICA) is often used as a signal processing technique for interpreting remote sensing images. Its basic principle is that the mixed spectral signal received by a single pixel in a remote sensing image is a linear superposition of multiple statistically independent and pure ground object source signals (such as pure water, chlorophyll, suspended sediment, etc.). The ICA algorithm separates these independent source signal components from the mixed signal through mathematical transformation.

[0004] However, traditional ICA algorithms treat each pixel in an image as an independent and unrelated spectral sample in their mathematical models, completely ignoring the close spatial relationships and contextual information between pixels and their neighbors. This pixel-level processing method, purely based on spectral statistical characteristics, fundamentally contradicts the essential characteristic of algal blooms as a physical phenomenon that necessarily exists in a spatially continuous and clustered manner. This leads to extreme sensitivity to noise, easily misjudging random noise as independent signals, generating a large amount of false information, and resulting in fragmented separation results. Even when some real algal signals are separated, they appear as discrete salt-and-pepper noise-like spots, unable to form continuous patches, rendering subsequent quantitative analysis baseless. Therefore, overcoming the monitoring accuracy problem caused by the neglect of spatial correlation in traditional ICA methods is a pressing technical challenge that needs to be addressed in this field. Summary of the Invention

[0005] To address the technical problem of inaccurate monitoring results caused by the neglect of spatial correlation in existing ICA methods, this invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a wetland area status monitoring method based on image enhancement, comprising: preprocessing an acquired remote sensing image to obtain a water body image of a target wetland area; constructing a spatial spectral joint feature vector for each pixel in the water body image, wherein the spatial spectral joint feature vector can simultaneously characterize the spectral characteristics of the pixel and the spatial aggregation characteristics related to its neighboring pixels; constructing a feature data matrix from the spatial spectral joint feature vectors of all pixels in the water body image, and performing independent component analysis on the feature data matrix to separate algal component images characterizing the spatial distribution of algae; and postprocessing the algal component images to generate wetland area status monitoring results, thereby realizing the monitoring of the wetland area status.

[0007] This invention modifies the input features of the independent component analysis algorithm, no longer using the isolated original spectral vector of a single pixel as input. Instead, it constructs a spatial spectral joint feature vector for each pixel that can simultaneously characterize its spectral uniqueness and spatial clustering. By performing independent component analysis on this enhanced feature vector that incorporates neighborhood information, the separation process can adaptively perceive and utilize the spatial continuity features of algal blooms, thereby suppressing noise at the source and ensuring the integrity and accuracy of the separation results.

[0008] Preferably, the calculation of the neighborhood spatial texture features of the pixel point yields the local gradient entropy; the calculation of the neighborhood spectral clustering features of the pixel point yields the neighborhood spectral dispersion; spatial adaptive weights are generated based on the local gradient entropy and the neighborhood spectral dispersion; and the original spectral vector of the pixel point and the average spectral vector of its neighborhood are weighted and fused using the spatial adaptive weights to generate a joint spatial spectral feature vector.

[0009] By calculating the texture and spectral clustering features of the pixel neighborhood step by step and designing spatially adaptive weights for intelligent fusion, it is possible to accurately generate enhanced features containing rich spatial context information for each pixel, laying a solid foundation for subsequent high-precision separation.

[0010] Preferably, the local gradient entropy satisfies the following relationship: ;in, For pixels The local gradient entropy, where g is the pixel. The gradient magnitude level is defined within the neighborhood window. This represents the probability that the average gradient magnitude of the pixels within the neighborhood window falls within level g.

[0011] Local gradient entropy can effectively assess the complexity of spectral changes within a pixel's neighborhood. Pure water surfaces have a simple texture and low entropy values; however, early algal blooms introduce subtle textures, leading to increased entropy values.

[0012] Preferably, the neighborhood spectral dispersion satisfies the following relationship: ;in, For pixels The neighborhood spectral dispersion, The size of the neighborhood window of a pixel. The number of spectral bands in the remote sensing image. and Neighboring pixels and center pixel The reflectance value in the b-th band, It represents the neighborhood of a pixel.

[0013] Neighborhood spectral dispersion can assess the degree of cohesion between the central pixel and its surrounding environment from the perspective of spectral similarity. A pixel belonging to a homogeneous region has low dispersion, while an isolated noise point has high dispersion.

[0014] Preferably, the spatial adaptive weights satisfy the following relationship: ;in, For pixels Spatial adaptive weights, () is a linear normalization function. For local gradient entropy, The spectral dispersion of the neighborhood, and It is a positive adjustment coefficient.

[0015] Spatial adaptive weights can intelligently adjust the fusion ratio of the original spectral information and the neighborhood spatial information. The pixel weights of typical algal bloom features are increased to suppress single-point noise by using neighborhood averaging; the pixel weights of typical isolated noise point features are decreased to preserve the uniqueness of their original spectra, thereby achieving adaptive filtering.

[0016] Preferably, the joint spatial spectral eigenvectors satisfy the following relationship: ;in, For pixels The joint eigenvector of spatial spectrum For spatial adaptive weights, For pixels The original spectral vector, The size of the neighborhood window of a pixel. It represents the neighborhood of a pixel.

[0017] Preferably, after performing independent component analysis on the feature data matrix, the method further includes: calculating the skewness of the pixel value distribution of each separated independent component image, and selecting the independent component image with the largest skewness as the algae component image.

[0018] Preferably, the post-processing of the algal component image includes: applying an adaptive threshold segmentation method to the algal component image to extract a spatial binary map of the algal bloom; and calculating the area, perimeter, and aggregation index of the algal bloom based on the spatial binary map to generate monitoring results.

[0019] Preferably, the method for preprocessing the acquired remote sensing images is to process the remote sensing images using the normalized difference water index to extract water body images.

[0020] Secondly, the present invention provides a wetland area status monitoring system based on image enhancement, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned wetland area status monitoring method based on image enhancement is implemented.

[0021] By adopting the above technical solution, a computer program for the image enhancement-based wetland area status monitoring method is generated and stored in a memory for loading and execution by a processor. This allows for the creation of a terminal device based on the memory and processor, facilitating its use.

[0022] This invention employs adaptive local smoothing of the input independent component analysis data by constructing a joint spatial-spectral feature vector, effectively suppressing the interference of isolated random noise points in remote sensing images on the separation results. This results in the final separated algal component image no longer exhibiting the fragmented, discrete forms produced by traditional methods, but rather spatially continuous and morphologically more complete patches. This highly aligns with the actual spatial distribution patterns of algal blooms as physical entities, significantly enhancing the algorithm's noise resistance and the spatial coherence of the results.

[0023] Furthermore, spatial adaptive weighting can intelligently distinguish between homogeneous and heterogeneous regions. Within algal blooms, by strengthening the fusion with neighboring spectra, the common spectral features of the region are amplified, allowing weak overall signals to accumulate and be enhanced. At noise points, the effect of neighborhood averaging is suppressed, preventing noise diffusion and thus significantly improving the signal-to-noise ratio of the target signal. This enables more accurate separation of weak but spatially clustered algal signals from complex backgrounds, effectively improving the detection rate of early weak algal blooms. Attached Figure Description

[0024] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1This is a flowchart illustrating an image enhancement-based wetland area status monitoring method according to the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, 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.

[0026] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] This invention discloses a method for monitoring the state of wetland areas based on image enhancement, referring to... Figure 1 This includes steps S1-S4: S1. Preprocess the acquired remote sensing images to obtain water images of the target wetland area.

[0028] In an optional embodiment, multispectral remote sensing images covering the target wetland area are first acquired. For example, L1C level image data acquired by the European Space Agency's Sentinel-2 satellite can be used. This data contains multiple spectral bands. To effectively identify algae containing chlorophyll, four key bands are selected: blue (Band 2, ~492nm), green (Band 3, ~560nm), red (Band 4, ~665nm), and near-infrared (NIR, Band 8, ~833nm).

[0029] In this optional embodiment, after acquiring remote sensing images, standardized preprocessing is required to ensure the accuracy of the analysis. The first step is radiometric calibration, which converts the dimensionless raw data recorded by the sensor into atmospheric top reflectance with clear physical meaning. The second step is atmospheric correction, which uses a professional radiative transfer model, such as the FLAASH module, to eliminate the influence of scattering and absorption by atmospheric molecules and aerosols on the light signal, and obtains image data that can reflect the true surface reflectance. This preprocessing step is crucial for the accurate detection of subsequent weak signals, ensuring that the analysis reflects the true spectral characteristics of the wetland itself.

[0030] Furthermore, to focus the analysis on the water body and eliminate interference from irrelevant features such as land and buildings, an accurate binary mask image of the water body is needed. Specifically, this can be extracted using the Normalized Difference Water Index (NDWI), which satisfies the following relationship:

[0031] Green and NIR represent the true surface reflectance in the green light band and near-infrared band, respectively. Since water has strong absorption characteristics in the near-infrared band and high reflectance in the green light band, the NDWI value of water bodies is usually positive, while that of non-water bodies is negative or close to zero.

[0032] In this optional embodiment, by setting an appropriate threshold, for example, such as a threshold of 0.1, the pixels in the image can be divided into two categories: water bodies and non-water bodies, thereby generating a water body mask. All subsequent steps are performed only within the area covered by the water body mask.

[0033] In this way, through a standardized preprocessing procedure, high-quality remote sensing images containing only water areas can be obtained, providing a reliable data foundation for subsequent accurate feature extraction and analysis.

[0034] S2. Construct a spatial spectral joint feature vector for each pixel in the water image. The spatial spectral joint feature vector can simultaneously characterize the spectral characteristics of the pixel and the spatial clustering characteristics related to its neighboring pixels.

[0035] In an optional embodiment, pixel points are obtained. The true reflectance values ​​in the four bands constitute its original spectral vector. This vector represents only the spectral information of that pixel.

[0036] Furthermore, the neighborhood spatial texture features of a pixel can be calculated to obtain the local gradient entropy. This is used to assess the complexity of spectral variations within a pixel's neighborhood. A calm and pristine water surface appears as a smooth and homogeneous region in an image, with minimal gradient changes and low information entropy; however, the formation of early, weak algal blooms introduces subtle irregular textures, leading to increased local gradient changes and higher information entropy.

[0037] Specifically, in terms of pixels Define an exemplary size centered at [center]. A neighborhood window is defined, and the Sobel gradient operator is used to calculate the gradient magnitude of each pixel within the window across four bands. The average gradient magnitude is then calculated to obtain an average gradient magnitude map. The average gradient magnitude within the window is divided into 16 exemplary bins, and the number of pixels falling within each bin is counted to construct a probability histogram of the local gradient distribution. The local gradient entropy then satisfies the following relationship:

[0038] Where g is the pixel point The gradient magnitude level is defined within the neighborhood window. This represents the probability that the average gradient magnitude of the pixels within the neighborhood window falls within level g.

[0039] Furthermore, the neighborhood spectral clustering characteristics of a pixel can be calculated to obtain the neighborhood spectral dispersion. This is used to evaluate the spectral similarity between the center pixel and its surrounding environment. A pixel belonging to a homogeneous region, such as the interior of a large algal bloom, should have a spectrum highly similar to its surrounding pixels, exhibiting low dispersion. Conversely, an isolated noise point will have a spectrum significantly different from its neighboring pixels, resulting in high dispersion. The dispersion of the neighborhood spectrum satisfies the following relationship:

[0040] in, The size of the neighborhood window of a pixel. The number of spectral bands in the remote sensing image. and Neighboring pixels and center pixel The reflectance value in the b-th band, It represents the neighborhood of a pixel.

[0041] Furthermore, spatially adaptive weights can be generated based on local gradient entropy and neighborhood spectral dispersion to intelligently adjust the fusion ratio of original spectral information and neighborhood spatial information. The logic is as follows: when a pixel is in a textured region with concentrated spectra, it corresponds to high local gradient entropy and low neighborhood spectral dispersion, which is a typical characteristic of algal blooms. In this case, the weights should be increased to fuse more neighborhood information and suppress noise. When a pixel is in a smooth region but with abnormal spectra, it corresponds to low local gradient entropy and high neighborhood spectral dispersion, which is a typical characteristic of noise. Therefore, the weights should be reduced to preserve the uniqueness of its original spectrum and prevent noise from contaminating the neighborhood. The spatially adaptive weights satisfy the following relationship:

[0042] in, For pixels Spatial adaptive weights, () is a linear normalization function used to normalize the globally calculated values. and The matrices are linearly normalized to... Intervals are used to eliminate differences in dimensions; and The adjustment coefficient is a positive number; for example, it can take values ​​of 1 and 0.5 respectively.

[0043] In an optional embodiment, spatial adaptive weights can be used to weight and fuse the original spectral vector of a pixel and the average spectral vector of its neighborhood to generate a joint spatial spectral feature vector, which satisfies the following relationship:

[0044] in, For pixels Spatial spectral joint eigenvectors, when weight When the weight is close to 1, Primarily determined by the neighborhood average spectrum, it effectively filters out single-point noise; when When approaching 0, It is primarily determined by the original spectrum, thus preserving its original uniqueness.

[0045] Thus, by constructing a joint feature vector that incorporates spatial context information for each pixel, the independent component analysis algorithm is provided with spatial awareness from the data source, enabling it to better distinguish between real spatial clustered signals and random noise.

[0046] S3. Construct a feature data matrix by combining the spatial spectral feature vectors of all pixels in the water image, and perform independent component analysis on the feature data matrix to separate algal component images that characterize the spatial distribution of algae.

[0047] In an optional embodiment, the joint spatial spectral feature vectors of all pixels within the water body region can be used to construct a new feature data matrix. If the water body region has M pixels and the image has N bands (N is 4 in this embodiment), then the size of this matrix is... .

[0048] Furthermore, for this new data matrix that already contains spatial context information, a standard ICA algorithm procedure is executed, such as the FastICA algorithm. This procedure typically includes data centralization and whitening, followed by iterative optimization of an unmixing matrix to maximize the non-Gaussianity of the output signal. Because the input data has been preprocessed, the ICA separation process not only searches for statistically independent spectral sources but also indirectly utilizes spatial clustering to guide the separation, thus more effectively resisting noise interference.

[0049] After the algorithm completes, it will produce independent component images. To automatically identify the components representing algae, the statistical distribution characteristics of each component image can be calculated. Typically, component images representing sparse, anomalous targets such as early algal blooms will exhibit a significant positive skewness in their pixel value distribution, meaning that a few high-value points correspond to algae, while a large number of low-value points correspond to background water. Therefore, by calculating the skewness of each component image and selecting the component image with the largest skewness, it can be determined as the enhanced algal spatial distribution map. In this embodiment, this enhanced algal spatial distribution map is used as the algal component image characterizing the spatial distribution of algae.

[0050] Thus, by performing independent component analysis on enhanced features containing spatial information, weak algal signals can be separated with a higher signal-to-noise ratio and stronger robustness.

[0051] S4. Post-process the algal composition image to generate wetland area status monitoring results, thereby realizing the monitoring of wetland area status.

[0052] In an optional embodiment, an adaptive thresholding method, such as the Otsu algorithm, can be applied to the obtained spatially continuous and high signal-to-noise ratio algal component image. This algorithm can automatically find an optimal threshold to segment the image into the foreground algal bloom and the background pure water body, thereby generating a spatial binary map of the algal bloom.

[0053] Furthermore, the area of ​​algal blooms can be obtained by counting the total number of foreground pixels in the binary image and multiplying it by the actual geographical area represented by a single pixel; the perimeter of the bloom patches can be obtained by calculating the boundary length of all foreground patches; and the aggregation index can be obtained by calculating and evaluating the aggregation degree of the blooms using landscape pattern analysis methods.

[0054] In this optional embodiment, by performing the same analysis on remote sensing images acquired at different times and comparing the changing trends of these quantitative indicators over time, dynamic monitoring and effective early warning of wetland algal blooms can be achieved.

[0055] Thus, by post-processing high-quality algal composition maps, accurate and reliable quantitative monitoring results can be produced, providing a scientific basis for wetland ecological management.

[0056] This invention also discloses an image-enhanced wetland area status monitoring system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an image-enhanced wetland area status monitoring method according to the present invention.

[0057] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0058] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, unless otherwise expressly and specifically defined.

[0059] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A method for monitoring the state of a wet area based on image enhancement, characterized by, The method comprises the following steps: preprocessing the obtained remote sensing image to obtain a water body image of a target wetland area; constructing a spatial-spectral joint feature vector for each pixel point in the water body image, the spatial-spectral joint feature vector being capable of representing the spectral characteristics of the pixel point and the spatial aggregation characteristics related to the neighborhood pixels of the pixel point; constructing the spatial-spectral joint feature vectors of all pixel points in the water body image into a feature data matrix, and performing independent component analysis on the feature data matrix to separate an algae component image representing the spatial distribution of algae; post-processing the algae component image to generate a state monitoring result of the wetland area, thereby realizing the monitoring of the state of the wetland area.

2. The image enhancement based wet area condition monitoring method according to claim 1, characterized in that, The construction process of the spatial-spectral joint feature vector is as follows: calculating the neighborhood spatial texture feature of the pixel point to obtain a local gradient entropy; calculating the neighborhood spectral aggregation feature of the pixel point to obtain a neighborhood spectral dispersion degree; generating a spatial adaptive weight based on the local gradient entropy and the neighborhood spectral dispersion degree; performing weighted fusion on the original spectral vector of the pixel point and the average spectral vector of the neighborhood of the pixel point by using the spatial adaptive weight to generate a spatial-spectral joint feature vector.

3. The image enhancement based wet area condition monitoring method according to claim 2, wherein, The local gradient entropy satisfies the relationship: wherein, is the local gradient entropy of the pixel point , g is the gradient magnitude level defined within the neighborhood window of the pixel point , is the probability that the average gradient magnitude of the pixels within the neighborhood window falls within the level g.

4. The image enhancement based wet area condition monitoring method according to claim 2, wherein, The neighborhood spectral dispersion degree satisfies the relationship: wherein, is the neighborhood spectral dispersion of the pixel point , is the size of the pixel point neighborhood window, is the number of spectral bands of the remote sensing image, and are the reflectance values of the neighborhood pixel points and the central pixel point at the bth band, is the pixel point neighborhood.

5. The image enhancement based wet area condition monitoring method according to claim 2, wherein, The spatial adaptive weight satisfies the relationship: wherein, is a spatial adaptive weight for the pixel point , is a linear normalization function, is a local gradient entropy, is a neighborhood spectral dispersion, and are positive adjustment coefficients.

6. The image enhancement based wet area condition monitoring method according to claim 2, wherein, The spatial-spectral joint feature vector satisfies the relationship: wherein, is a spatial spectral joint feature vector of the pixel point , is a spatial adaptive weight, is an original spectral vector of the pixel point , is a size of a pixel point neighborhood window, is a pixel point neighborhood.

7. The image enhancement based wet area condition monitoring method according to claim 1, wherein, After performing the independent component analysis on the feature data matrix, the method further comprises the following steps:

8. The image enhancement based wet area condition monitoring method according to claim 6, characterized in that, calculating the skewness of the pixel value distribution of each separated independent component image, and selecting the independent component image with the maximum skewness as the algae component image. The post-processing of the algae component image comprises the following steps: applying an adaptive threshold segmentation method to the algae component image to extract a spatial binary image of the algae bloom; 9. The image enhancement based wet area condition monitoring method according to claim 1, wherein, calculating the area, perimeter and aggregation degree indexes of the algae bloom based on the spatial binary image to generate the monitoring result.

10. An image enhancement based wet area status monitoring system, characterized in that, The preprocessing of the obtained remote sensing image is performed by using a normalized difference water body index to process the remote sensing image to extract the water body image. The method comprises the following steps: a processor and a memory, the memory storing computer program instructions, when the computer program instructions are executed by the processor, realizing the image enhancement-based wetland area state monitoring method according to any one of claims 1-9.

Images