Method for evaluating ecological environment condition in real time by using remote sensing image

Abstract

The invention provides a method for evaluating an ecological environment condition in real time by using a remote sensing image, and relates to the technical field of ecological environment monitoring, and the method comprises the steps: S1, obtaining remote sensing image data which covers data under different time, ecological regions and environment conditions, S2, employing a noise suppression algorithm based on low-rank matrix decomposition, and S3, carrying out the real-time evaluation of the ecological environment condition through employing a noise suppression algorithm based on low-rank matrix decomposition. According to the noise suppression algorithm, a low-rank part and a noise part of an image are decomposed, and global information and local detail information of the image are effectively separated, and S3, hierarchical processing and information compression are carried out on the image through a compressed sensing technology. According to the method for evaluating the ecological environment condition in real time by utilizing the remote sensing image, through a noise suppression algorithm combining low-rank matrix decomposition and singular value decomposition, noise can be accurately removed from the remote sensing image, and meanwhile, global information and local details of the image are reserved. The quality of the remote sensing image is effectively improved, and a clear and interference-free data basis is provided for subsequent analysis.

Description

Technical Field

[0001] This invention relates to the field of ecological environment monitoring technology, specifically a method for real-time evaluation of the ecological environment using remote sensing images. Background Technology

[0002] Remote sensing technology is increasingly widely used in ecological and environmental monitoring, becoming an important means of assessing changes in the Earth's surface ecology. Remote sensing images, acquired through platforms such as satellites and drones, can quickly cover large areas, providing a convenient approach for environmental monitoring. The application of remote sensing images not only effectively reflects the physical characteristics of the ground environment but also provides rich spatiotemporal data, helping scientists better understand the dynamic changes in the ecological environment. With the development of remote sensing technology, the resolution of remote sensing images is constantly improving, making their application in precise ecological and environmental monitoring even more widespread. Using remote sensing images, researchers can monitor multiple aspects such as land use change, vegetation cover, wetland ecology, and water pollution, forming a comprehensive understanding of ecological and environmental changes. For example, remote sensing images are widely used in forest resource monitoring, grassland degradation, and lake water quality assessment, providing data support for the protection and restoration of the ecological environment. Modern remote sensing image processing technology is constantly improving, and image processing software and algorithms are also gradually being upgraded. Traditional image processing methods have been gradually replaced by advanced methods based on machine learning and artificial intelligence, enabling more efficient extraction and analysis of information from remote sensing images.

[0003] Despite significant progress in ecological and environmental monitoring, remote sensing technology still faces core challenges – namely, the high complexity and high noise levels associated with remote sensing image data processing. Existing remote sensing image processing methods, especially in high-resolution image processing, rely on complex image analysis algorithms that require substantial computational resources and are susceptible to noise interference during processing. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for real-time evaluation of ecological and environmental conditions using remote sensing images. The technical problem this invention aims to solve is: how to combine low-rank matrix factorization and compressed sensing to achieve accurate processing and fusion of remote sensing image data and real-time sensor data, thereby optimizing the extraction of ecological and environmental features and model parameters.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for real-time evaluation of the ecological environment using remote sensing imagery, comprising: S1. Acquire remote sensing image data, which covers data from different times, ecological regions, and environmental conditions; S2. A noise suppression algorithm based on low-rank matrix factorization is adopted. The noise suppression algorithm effectively separates the low-rank part and the noise part of the image by decomposing the image into global information and local detail information. S3. Image processing and information compression are performed using compressed sensing technology; S4. Multi-level ecological environment feature extraction is performed using an adaptive graph convolutional neural network. S5. By combining edge computing and automated feedback mechanisms, real-time sensor data and remote sensing image data are fused to optimize the model parameters.

[0006] Preferably, the remote sensing image data is acquired through at least one remote sensing platform, including satellites or drones, and the spatial resolution of the remote sensing image data does not exceed 1 meter.

[0007] Preferably, the time range includes environmental changes in different seasons and years, the ecological area includes cities, forests, wetlands, grasslands, agricultural land and water bodies, and the environmental conditions include sunny days, cloudy days, and rainy / snowy weather.

[0008] Preferably, the low-rank matrix decomposition is achieved through singular value decomposition, and the specific steps include: S2.1. Input Image Matrix ,in and These represent the number of rows and columns of the image, respectively. S2.2. Image Matrix Perform singular value decomposition to obtain ,in ,and Let be the rank of the low-rank matrix; S2.3. By selecting larger singular values ​​and retaining their corresponding singular vectors, the low-rank part is obtained. ,in , This is the result of the singular value decomposition of the low-rank part; S2.4. Noise Section Through calculation This means that the noise component is extracted from the original image by subtracting the low-rank component.

[0009] Preferably, the specific steps of S3 include: S3.1 Input Image Matrix ; S3.2. Image matrix Mapping to a low-dimensional space, through the mapping matrix of compressed sensing. Perform data compression, where S3.3. Solving the optimization problem The reconstructed image was obtained. ,in For measurement data, This is a regularization parameter that controls the reconstruction accuracy; S3.4. Repeatedly perform compressed sensing processing and reconstruction at multiple levels of the image to gradually extract the feature levels in the image.

[0010] Preferably, the specific steps of S3.2 include: Image matrix Mapped into a compressed data matrix We obtain the following through matrix multiplication: in, It is a sparse mapping matrix with a specific structure, the purpose of which is to achieve data compression by preserving important information in the image while reducing redundant information.

[0011] Preferably, S3.4 specifically includes: a. Perform preliminary compressed sensing processing on the remote sensing image data, dividing the remote sensing image data into different layers. Each layer contains different detailed information of the image data. The initial layer processes the lower frequency overall structure, while the higher layers focus on the detailed information of the image. b. By applying compressed sensing processing and reconstruction technology at multiple levels, multi-level features in the image are extracted step by step. After data compression and information extraction at each level, the optimization algorithm is used to reconstruct the image at each level and ensure the consistency of the features of each level with the overall image. c. The reconstruction results of each level will provide more detailed information for the next level. The reconstruction process is iterative, gradually approaching the final high-quality image reconstruction result.

[0012] Preferably, the edge computing includes: data preprocessing, real-time data stream processing, and local computing and data storage.

[0013] Preferably, the data fusion combines remote sensing image data with real-time sensor data through a deep learning model to achieve precise docking and fusion of the two data.

[0014] This invention provides a method for real-time assessment of the ecological environment using remote sensing imagery. It has the following beneficial effects:

[0015] This method for real-time assessment of the ecological environment using remote sensing imagery employs a noise suppression algorithm combining low-rank matrix factorization (LVDF) and singular value decomposition (SVD). This algorithm accurately removes noise from remote sensing images while preserving global information and local details. This effectively improves the quality of remote sensing images, providing a clear and interference-free data foundation for subsequent analysis. LVDF, performed through singular value decomposition, precisely separates noise from valid information in the images, ensuring the reliability and validity of the data even in high-noise environments.

[0016] This scheme utilizes compressed sensing technology to perform layered processing and information compression of image data. The application of compressed sensing not only reduces data storage requirements but also ensures effective capture of information at different scales by extracting multi-level features from the images layer by layer. Further optimization of feature extraction through adaptive graph convolutional neural networks enables the extraction of more detailed and accurate ecological and environmental features from remote sensing image data under different environmental conditions. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the process of realizing an invention; Figure 2 This is a schematic diagram illustrating the process of implementing noise suppression and low-rank matrix decomposition in the invention. Figure 3 This is a schematic diagram of the process for implementing an edge computing and automated feedback mechanism for an invention. Detailed Implementation

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

[0019] Example 1 like Figure 1-3 As shown, this embodiment of the invention provides a method for real-time evaluation of ecological environment status using remote sensing images, including: S1. Acquiring remote sensing image data, which covers data from different times, ecological regions, and environmental conditions. The remote sensing image data is acquired through at least one remote sensing platform, including satellites or drones. The spatial resolution of the remote sensing image data does not exceed 1 meter. The time range includes environmental changes in different seasons and years. The ecological regions include cities, forests, wetlands, grasslands, agricultural land, and water bodies. The environmental conditions include sunny days, cloudy days, and rainy / snowy weather.

[0020] S2. A noise suppression algorithm based on low-rank matrix factorization is adopted. This algorithm decomposes the low-rank portion of the image into its noise portion, effectively separating them using global and local detail information to achieve accurate noise removal. Low-rank matrix factorization is implemented through singular value decomposition, and the specific steps include:

[0021] S2.1. Input Image Matrix ,in and These represent the number of rows and columns of the image, respectively.

[0022] S2.2. Image Matrix Perform singular value decomposition to obtain ,in ,and Let be the rank of the low-rank matrix.

[0023] S2.3. By selecting larger singular values ​​and retaining their corresponding singular vectors, the low-rank part is obtained. ,in , This is the result of the singular value decomposition of the low-rank part.

[0024] S2.4. Noise Section Through calculation This means that the noise component is extracted from the original image by subtracting the low-rank component.

[0025] S3. Image layering and information compression are performed using compressed sensing technology. The specific steps of S3 include:

[0026] S3.1 Input Image Matrix .

[0027] S3.2. Image matrix Mapping to a low-dimensional space, through the mapping matrix of compressed sensing. Perform data compression, where The specific steps include: Image matrix Mapped into a compressed data matrix We obtain the following through matrix multiplication: in, It is a sparse mapping matrix with a specific structure, the purpose of which is to achieve data compression by preserving important information in the image while reducing redundant information.

[0028] S3.3. Solving the optimization problem The reconstructed image was obtained. ,in For measurement data, This is a regularization parameter that controls the reconstruction accuracy.

[0029] S3.4. Repeatedly perform compressed sensing processing and reconstruction at multiple levels of the image to gradually extract feature layers from the image, specifically including: a. Perform preliminary compressed sensing processing on the remote sensing image data, dividing the remote sensing image data into different layers. Each layer contains different detailed information of the image data. The initial layer processes the lower frequency overall structure, while the higher layers focus on the detailed information of the image.

[0030] b. By applying compressed sensing processing and reconstruction technology at multiple levels, multi-level features in the image are extracted step by step. After data compression and information extraction at each level, the optimization algorithm is used to reconstruct the image at each level and ensure the consistency of the features of each level with the overall image.

[0031] c. The reconstruction results at each level provide more detailed information for the next level. The reconstruction process is iterative, gradually approaching the final high-quality image reconstruction result.

[0032] S4. Multi-level ecological environment feature extraction is performed using an adaptive graph convolutional neural network.

[0033] S5. Combining edge computing and automated feedback mechanisms, real-time sensor data and remote sensing image data are fused to optimize model parameters. Edge computing includes data preprocessing, real-time data stream processing, and local computation and data storage. Data fusion uses a deep learning model to combine remote sensing image data and real-time sensor data, achieving precise docking and fusion of the two data sets.

[0034] Example 2 This implementation method employs a noise suppression algorithm based on low-rank matrix factorization, utilizing singular value decomposition to decompose image data, thereby achieving accurate noise removal. The specific implementation steps of this algorithm are as follows:

[0035] S2.2 Input Image Matrix image As the input matrix, where Indicates the number of rows in the image. This indicates the number of columns in the image. Assume the image size is... ,Right now:

[0036] Among them, matrix Each element This represents the brightness value of a pixel in an image.

[0037] S2.2 Perform singular value decomposition on the image matrix The image matrix is ​​processed using the Singular Value Decomposition (SVD) method. Decomposed into three matrices: in: It is a left singular vector matrix. It is a diagonal matrix that contains the singular values ​​of the image. It is a right singular vector matrix.

[0038] The order of images If it is 100, then and The dimension is (Or even smaller, depending on the actual rank value). Singular value decomposition provides the basis for low-rank approximation of the image and separation of noise information.

[0039] S2.3 Select the larger singular value and retain its corresponding singular vector. To extract the low-rank portion of the image, larger singular values ​​are selected, and their corresponding singular vectors are preserved. During singular value decomposition, the resulting singular values ​​are:

[0040] in These are singular values ​​sorted by size. (Before selection) Large singular values ​​(e.g.) The image is processed and its corresponding singular vectors are preserved. These larger singular values ​​and their corresponding singular vectors usually contain the main information of the image, while smaller singular values ​​usually represent noise.

[0041] Low-rank part It can be calculated using the following formula: in, and They are including the front A matrix of columns and rows, Is included before A diagonal matrix with singular values.

[0042] S2.4 Calculation of Noise Section Noise section This can be obtained by calculating the difference between the original image and the low-rank portion, i.e.: in, The original image matrix, This represents the low-rank part. This formula allows noise to be extracted from the image.

[0043] For example, the following matrix: Then the noise part for: This noise component refers to the unnecessary noise information in the image obtained through decomposition.

[0044] Example 3 This implementation uses compressed sensing technology to perform layered processing and information compression on remote sensing images, gradually extracting multi-layered features from the images to achieve data compression and accurate reconstruction. The following are the specific steps and example data for implementing this method:

[0045] S3.1 Input Image Matrix Input remote sensing image matrix ,in The number of rows in the image. This represents the number of columns in the image. The image size is [size missing]. The data can be represented as:

[0046] in, This represents the value of each pixel in the image; typically, these pixel values ​​are brightness values ​​or other image features.

[0047] S3.2 maps the image matrix to a lower-dimensional space. Mapping matrix through compressed sensing technology (in ) for image matrix Perform data compression. The specific steps are as follows:

[0048] 1. Image Matrix Mapped into a compressed data matrix : in, It is a sparse matrix with a specific structure designed to compress data by preserving the most important information in the image while removing redundant information.

[0049] 2. Mapping matrix Typically generated randomly, this ensures that the matrix can effectively perform sparse encoding and compression of the original data. The size is ,and The size is usually larger than Small, if the original image is You can choose The data is compressed to one-quarter of its original size.

[0050] S3.3 Obtains the reconstructed image by solving an optimization problem. Image data is recovered through an optimization problem, the following optimization problem is used: in: It is the data reconstruction error, representing the error from compressed data. and images reconstructed through mapping matrix The differences between them.

[0051] This is a regularization parameter that controls the reconstruction accuracy and sparsity.

[0052] yes The norm term is used to improve the sparsity of the reconstructed image and avoid excessive redundancy during the reconstruction process.

[0053] This optimization problem is solved using the standard gradient descent method to obtain the reconstructed image. .

[0054] S3.4 Repeatedly performs compressed sensing processing and reconstruction at multiple layers of the image. 1. Perform preliminary compressed sensing processing on the remote sensing image data: Remote sensing image data is divided into multiple layers, each representing different levels of detail within the image. Lower layers typically process lower-frequency information, representing the overall structure of the image. Higher layers process higher-frequency information, representing the details of the image.

[0055] For example, the image can be divided into three levels: Layer 1 (low-frequency part): Represents the overall structure of the image, such as large-scale terrain information.

[0056] Layer 2 (Mid-frequency component): Represents mid-scale details of the image, such as regional variations.

[0057] Layer 3 (high-frequency part): Represents detailed information, such as local changes in small areas.

[0058] 2. Apply compressed sensing and reconstruction layer by layer: Compressed sensing processing and reconstruction techniques are applied to each layer of the image. Assume the first layer image matrix is... ,but:

[0059] Then, an optimization algorithm is used to solve the problem: After the reconstruction of the first layer is completed, it is used as the input for the second layer to process and gradually extract the features of each layer in the image.

[0060] Example 4 This implementation uses adaptive graph convolutional neural networks and edge computing technology to optimize the parameters of the ecological environment model by fusing real-time sensor data with remote sensing image data.

[0061] S4. Multi-level ecological environment feature extraction based on adaptive graph convolutional neural network 3. Input ecological and environmental data: The collected ecological and environmental data include remote sensing imagery and environmental sensor data, which are represented by the following matrices: Remote sensing image data: Assuming the image size is Each pixel contains multiple features (such as vegetation index, soil moisture, etc.). A matrix is ​​used... To represent remote sensing image data, where The feature dimension of the image.

[0062] Real-time sensor data: The data collected by the sensors is time-series data, such as temperature, humidity, and air pressure. Assuming there are 1000 data points, a matrix is ​​used. express.

[0063] 4. Graph Construction and Graph Convolution Operations: An adaptive graph convolutional neural network is used to process remote sensing image data and sensor data. First, based on the remote sensing image data... Constructing a graph structure:

[0064] Each node in the diagram represents a region in the image (e.g., each pixel or each small block of region).

[0065] The edges in the diagram represent the relationships between adjacent regions. For example, adjacent relationships can be constructed based on information such as the similarity of pixel values ​​and spatial location in the image.

[0066] Next, we apply graph convolution operations, as shown in the following formula: in: A is the adjacency matrix of the graph, representing the connection relationships between nodes.

[0067] D is the degree matrix, representing the number of connections for each node.

[0068] For the first Layer node features, initially For image data .

[0069] For the first Layer weight matrix.

[0070] For example, ReLU is an activation function.

[0071] By performing multi-layer graph convolution operations, hierarchical features in image data can be extracted step by step, and these features will reflect the ecological and environmental information in the image.

[0072] S5. Combining edge computing and automated feedback mechanisms to achieve data fusion and model optimization. 5. Edge computing component: Processing real-time sensor data through edge computing mainly includes the following aspects: Data preprocessing: The data obtained from the sensor is cleaned, denoised, and standardized. For example, suppose the sensor collects temperature and humidity data. After denoising and standardization, the processed data is obtained:

[0073] in, It is the average value of the sensor data. It is the standard deviation.

[0074] Real-time data stream processing: Edge computing nodes receive and process sensor data in real time. Assuming the data stream is updated every minute, the edge nodes process the data in real time and use models to predict the current state of the ecological environment.

[0075] Local computing and data storage: Preliminary computations (such as feature extraction and real-time analysis) are performed on edge devices, and some of the processing results are stored locally. This process can reduce transmission latency and bandwidth consumption through caching techniques.

[0076] 6. Data Fusion Section: Data fusion model: Deep learning models are used to fuse remote sensing imagery data and real-time sensor data to improve the model's prediction accuracy. Specifically, a fusion network is constructed:

[0077] in: This is remote sensing image data.

[0078] This is real-time sensor data.

[0079] The resulting predicted outputs include, for example, an ecological health index or land use type.

[0080] These are the parameters of the deep learning model.

[0081] The model improves prediction accuracy by fusing multi-level features between remote sensing imagery and sensor data, and can provide more accurate decision support, especially in complex environments.

[0082] Automated feedback mechanism: Edge computing nodes feed real-time processing results back to the model, dynamically adjusting model parameters through automated mechanisms. For example, when sensor data changes, the system automatically updates the model's weights to ensure the model adapts to environmental changes in real time. The update formula is as follows:

[0083] in: This is the learning rate.

[0084] loss function Relative to parameters The gradient.

[0085] Each time sensor data is updated, the system automatically adjusts parameters and optimizes the ecological environment prediction model to adapt to the current environmental conditions.

[0086] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for real-time evaluation of the ecological environment using remote sensing imagery, characterized in that, include: S1. Acquire remote sensing image data, which covers data from different times, ecological regions, and environmental conditions; S2. A noise suppression algorithm based on low-rank matrix factorization is adopted, wherein the noise suppression algorithm decomposes the low-rank part of the image into the noise part; S3. Image processing and information compression are performed using compressed sensing technology; S4. Multi-level ecological environment feature extraction is performed using an adaptive graph convolutional neural network. S5. By combining edge computing and automated feedback mechanisms, real-time sensor data and remote sensing image data are fused to optimize the model parameters.

2. The method for real-time evaluation of ecological environment status using remote sensing imagery according to claim 1, characterized in that: The remote sensing image data is acquired through at least one remote sensing platform, including satellites or drones, and the spatial resolution of the remote sensing image data does not exceed 1 meter.

3. The method for real-time evaluation of ecological environment status using remote sensing images according to claim 1, characterized in that: The time frame includes environmental changes in different seasons and years; the ecological regions include cities, forests, wetlands, grasslands, agricultural land, and water bodies; and the environmental conditions include sunny days, cloudy days, and rainy / snowy weather.

4. The method for real-time evaluation of ecological environment status using remote sensing images according to claim 1, characterized in that: The low-rank matrix decomposition is achieved through singular value decomposition, and the specific steps include: S2.

1. Input Image Matrix ,in and These represent the number of rows and columns of the image, respectively. S2.

2. Image Matrix Perform singular value decomposition to obtain ,in ,and Let be the rank of the low-rank matrix; S2.

3. By selecting larger singular values ​​and retaining their corresponding singular vectors, the low-rank part is obtained. ,in , This is the result of the singular value decomposition of the low-rank part; S2.

4. Noise Section Through calculation This means that the noise component is extracted from the original image by subtracting the low-rank component.

5. A method for real-time evaluation of ecological environment status using remote sensing images according to claim 4, characterized in that: The specific steps of S3 include: S3.1 Input Image Matrix ; S3.

2. Image matrix Mapping to a low-dimensional space, through the mapping matrix of compressed sensing. Perform data compression, where S3.

3. Solving the optimization problem The reconstructed image was obtained. ,in For measurement data, This is a regularization parameter that controls the reconstruction accuracy; S3.

4. Repeatedly perform compressed sensing processing and reconstruction at multiple levels of the image to gradually extract the feature levels in the image.

6. A method for real-time evaluation of ecological environment status using remote sensing imagery according to claim 5, characterized in that: The specific steps in S3.2 include: Image matrix Mapped into a compressed data matrix We obtain the following through matrix multiplication: in, It is a sparse mapping matrix with a specific structure.

7. A method for real-time evaluation of ecological environment status using remote sensing images according to claim 6, characterized in that: S3.4 specifically includes: a. Perform preliminary compressed sensing processing on the remote sensing image data to divide the remote sensing image data into different layers, each layer containing different detailed information of the image data; b. By applying compressed sensing processing and reconstruction technology at multiple levels, multi-level features in the image are extracted step by step. After data compression and information extraction at each level, the optimization algorithm is used to reconstruct the image at each level and ensure the consistency of the features of each level with the overall image. c. The reconstruction results of each level will provide more detailed information for the next level, and the reconstruction process is iterative.

8. The method for real-time evaluation of ecological environment status using remote sensing imagery according to claim 1, characterized in that: The edge computing includes: data preprocessing, real-time data stream processing, and local computing and data storage.

9. A method for real-time evaluation of ecological environment status using remote sensing imagery according to claim 1, characterized in that: The data fusion combines remote sensing image data with real-time sensor data through a deep learning model.

Images