A multispectral cloud detection method based on sharpening module and gaussian mixture model

Abstract

The application discloses a multispectral cloud detection method based on a sharpening module and a Gaussian mixture model, and relates to the technical field of remote sensing image processing.The method comprises the following steps: acquiring a remote sensing image to be detected; inputting the remote sensing image to be detected into a pre-constructed cloud detection model trained, and obtaining an output cloud detection result; wherein the cloud detection model is constructed and trained by the following method: acquiring a remote sensing image dataset; pre-processing the remote sensing image dataset to obtain a cloud detection dataset; selecting features from the cloud detection dataset to obtain three clustering features; constructing a cloud detection model by using the cloud detection dataset and the clustering features, wherein the cloud detection model comprises a feature selection module, a thin cloud sharpening module, a GMM clustering module and a superposition module. The method is simple to operate, fast to run, reduces the complexity of the cloud detection model, and can realize fast and high-precision cloud detection for different cloud types.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology and relates to cloud detection in remote sensing images, specifically a multispectral cloud detection method based on a sharpening module and a Gaussian mixture model. Background Technology

[0002] With the rapid development of remote sensing technology, satellite remote sensing images are increasingly being used in various fields due to their advantages such as large coverage area, strong temporal validity, and good data geographic integration. However, more than 66% of the Earth's surface is covered by clouds, and the presence of a large number of clouds in optical remote sensing data affects the quality of remote sensing images, thereby reducing the data utilization rate. This also means that a large amount of ground information cannot be acquired by satellites, which in turn brings great difficulties to subsequent target detection and analysis tasks. Therefore, in the process of optical remote sensing image processing, accurate cloud classification is a first step required for further analysis of satellite images.

[0003] Current cloud detection methods can be broadly categorized into multispectral thresholding and machine learning-based methods. Multispectral thresholding utilizes the spectral differences between clouds and the underlying surface across different wavelengths, manually setting thresholds or feature selection rules to detect both. While this method is simple in structure and easy to implement, it requires extensive prior knowledge and the design of manually crafted features, resulting in significant resource consumption and insufficient detection accuracy, lacking universality. Furthermore, in complex situations, particularly when the ground is covered by ice, snow, or desert, or when the clouds are thin or fragmented, distinguishing between clouds and the underlying surface becomes difficult.

[0004] In recent years, with the development of machine learning, many scholars have introduced it into the field of cloud detection, resulting in machine learning algorithms such as Random Forest (RF), Support Vector Machine (SVM), and k-nearest neighbors (KNN). While existing cloud detection methods incorporate machine learning techniques, reducing reliance on human intervention and improving detection accuracy, these methods require sufficient training data to achieve reliable performance. Thin clouds are often semi-transparent, and their samples are sparse and scattered compared to thick clouds, consuming significant time and effort for manual annotation. Furthermore, there is no universally accepted definition for whether semi-transparent areas are thin clouds or underlying surfaces during manual annotation, which hinders feature learning by the classifier and easily leads to detection errors. Therefore, improving the accuracy of cloud detection in remote sensing images is a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0005] To address the problems of existing technologies, this invention proposes a multispectral cloud detection method based on a sharpening module and a Gaussian mixture model (GMM). As an unsupervised learning method, GMM eliminates the need for manual sample labeling, thus reducing pre-detection preparation. Furthermore, the GMM algorithm uses the reflectivity and brightness temperature of satellite multi-channel data as clustering features, effectively utilizing the spectral differences between clouds and the underlying surface for classification, thereby reducing detection errors caused by inaccurate labeling. This cloud detection algorithm selects corresponding feature schemes for different time periods through clustering feature analysis to achieve all-day cloud detection. Secondly, the Laplacian operator is used to enhance the spectral features of thin cloud regions, thereby improving the thin cloud detection accuracy of the GMM classifier.

[0006] Technical solution: To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0007] In a first aspect, the present invention provides a multispectral cloud detection method based on a sharpening module and a Gaussian mixture model, comprising:

[0008] Acquire the remote sensing image to be detected;

[0009] The remote sensing image to be detected is input into a pre-built and trained cloud detection model to obtain the output cloud detection results.

[0010] The methods for constructing and training the cloud detection model include:

[0011] Step S1: Obtain the remote sensing image dataset;

[0012] Step S2: Preprocess the remote sensing image dataset to obtain the cloud detection dataset;

[0013] Step S3: Select three clustering features from the cloud detection dataset;

[0014] Step S4: Construct a cloud detection model using the cloud detection dataset and clustering features, wherein the cloud detection model includes a feature selection module, a thin cloud sharpening module, a GMM clustering module, and an overlay module.

[0015] In some embodiments, in step S2, the preprocessing includes a standardization process;

[0016] The standardization process for remote sensing images includes:

[0017]

[0018] In the formula, z is the value of the original remote sensing data after standardization, and X is the remote sensing data to be processed. mean X represents the average of the original data. stdThe standard deviation of the original data;

[0019] The preprocessing also includes using data augmentation techniques to enhance the remote sensing image dataset.

[0020] In some embodiments, step S3, selecting features from the cloud detection dataset to obtain three clustering features, including:

[0021] Statistical analysis of the spectral data of Himawari-8 was performed using probability density functions to generate probability density function distributions of cloud and underlying surface pixel reflectance, brightness temperature, and brightness-temperature difference. The sum of the reflectance of channels 3 and 4 of the AHI on the Himawari-8 was used as the basis for the analysis. 3+4 Brightness temperature BT of channel 15 15 And the brightness temperature difference (BTD) between channel 7 and channel 14 7-14 There are three types of clustering features; the selection of clustering features is completed based on the probability density function distribution of different clustering features of cloud and underlying surface pixels.

[0022] Furthermore, different clustering feature schemes were selected for cloud detection during the day and night: for remote sensing images acquired during the daytime, R... 3+4 Clustering features; the remote sensing images were acquired during nighttime and analyzed using Bit-Based Image Processing (BT). 15 and BTD 7-14 Clustering characteristics.

[0023] In some embodiments, the detection process for remote sensing images acquired during the daytime includes:

[0024] The feature selection module selects R features from remote sensing images acquired during the daytime. 3+4 Clustering features;

[0025] The thin cloud sharpening module sharpens remote sensing images acquired during the daytime to obtain enhanced images of thin cloud regions.

[0026] GMM clustering module, based on R 3+4 Clustering features: GMM clustering is performed on the remote sensing image to obtain the clustering results of the main cloud layer, and GMM clustering is performed on the enhanced image of the thin cloud region to obtain the clustering results of the thin cloud portion;

[0027] The overlay module overlays the clustering results of the main cloud layer and the thin cloud layer to obtain the cloud detection result.

[0028] In some embodiments, the detection process for remote sensing images acquired during nighttime includes:

[0029] The feature selection module selects BT features from remote sensing images acquired during nighttime periods. 15 and BTD 7-14 Clustering features;

[0030] GMM clustering module, based on BT 15 and BTD 7-14 Clustering features are used to perform GMM clustering on the remote sensing image to obtain cloud detection results.

[0031] In some embodiments, the thin cloud sharpening module uses the Laplacian operator to process the remote sensing image to enhance the thin cloud features, and then superimposes the image processed by the Laplacian operator with the original image to obtain an enhanced image of the thin cloud region.

[0032]

[0033] In the formula, g(x,y) represents the enhanced image of the thin cloud region, f(x,y) represents the pixel value of (x,y) in the remote sensing image, w(i,j) represents the sharpening convolution kernel, and ▽ 2 f is the Laplace operator;

[0034]

[0035] The Laplace operator is defined as follows:

[0036]

[0037] In some embodiments, the GMM clustering module uses a linear combination of multiple Gaussian functions to represent the probability density function of the data. The probability density function P(x|θ) is expressed as follows:

[0038]

[0039] In the formula, x is a vector X = {x1, x2, x3, ..., x}. n Given n sample values ​​X, which contain K categories, the weight of the j-th category is β. j The function f(x) is the probability density function of a single Gaussian model, μ j σ j Let be the mean and variance of the j-th class; the EM method is used to estimate the parameters θ = {β, μ, σ} of the GMM model. 2}, using R 3+4 The parameter K=7 was determined using the AIC and BIC criteria for the remote sensing image sample data.

[0040] In some embodiments, the overlay module is used to overlay the clustering results of the main cloud portion and the thin cloud portion to obtain the cloud detection result, as shown in the following formula:

[0041]

[0042] In the formula, W1 is the clustering result of the main cloud part of each sample point, W2 is the clustering result of the thin cloud part of each sample point, and W3 is the final cloud detection result of each sample point.

[0043] Secondly, the present invention provides a multispectral cloud detection device based on a sharpening module and a Gaussian mixture model, including a processor and a storage medium;

[0044] The storage medium is used to store instructions;

[0045] The processor is configured to operate according to the instructions to perform the steps of the method according to the first aspect.

[0046] Thirdly, the present invention provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0047] The beneficial effects of this invention are that, using GMM (Geometry Model), it can automatically learn the potential relationships between different categories and achieve classification without requiring a large number of training samples. This cloud detection method uses the reflectance and brightness temperature of satellite multi-channel data as clustering features, effectively utilizing the spectral differences between clouds and the underlying surface for classification, thereby reducing detection errors caused by inaccurate labeling. Furthermore, it selects corresponding feature schemes for different time periods to achieve all-day cloud detection. It also employs the Laplacian operator to enhance the spectral features of thin cloud regions, thereby improving the accuracy of thin cloud detection. This method is simple to operate, runs quickly, reduces the complexity of the cloud detection model, and can achieve fast and high-precision cloud detection for different cloud types. Attached Figure Description

[0048] Figure 1 This is a flowchart of the cloud detection method according to an embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram of the overall framework of the cloud detection method according to an embodiment of the present invention;

[0050] Figure 3 This is a schematic diagram of the image before and after sharpening in an embodiment of the present invention;

[0051] Figure 4 The R value of thin clouds before and after image sharpening in this embodiment of the invention. 3+4 Comparison diagram;

[0052] Figure 5 This is a schematic diagram of the cloud detection results in an embodiment of the present invention. Detailed Implementation

[0053] The present invention will be further described below with reference to the accompanying drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and should not be used to limit the scope of protection of the present invention.

[0054] In the description of this invention, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0055] Example 1:

[0056] A multispectral cloud detection method based on a sharpening module and a Gaussian mixture model includes:

[0057] Acquire the remote sensing image to be detected;

[0058] The remote sensing image to be detected is input into a pre-built and trained cloud detection model to obtain the output cloud detection results.

[0059] The methods for constructing and training the cloud detection model include:

[0060] Step S1: Obtain the remote sensing image dataset;

[0061] Step S2: Preprocess the remote sensing image dataset to obtain the cloud detection dataset;

[0062] Step S3: Select three clustering features from the cloud detection dataset;

[0063] Step S4: Construct a cloud detection model using the cloud detection dataset and clustering features, wherein the cloud detection model includes a feature selection module, a thin cloud sharpening module, a GMM clustering module, and an overlay module.

[0064] In some embodiments, the detection process for remote sensing images acquired during the daytime includes:

[0065] The feature selection module selects R features from remote sensing images acquired during the daytime. 3+4 Clustering features;

[0066] The thin cloud sharpening module sharpens remote sensing images acquired during the daytime to obtain enhanced images of thin cloud regions.

[0067] GMM clustering module, based on R 3+4Clustering features: GMM clustering is performed on the remote sensing image to obtain the clustering results of the main cloud layer, and GMM clustering is performed on the enhanced image of the thin cloud region to obtain the clustering results of the thin cloud portion;

[0068] The overlay module overlays the clustering results of the main cloud layer and the thin cloud layer to obtain the cloud detection result.

[0069] In some embodiments, the detection process for remote sensing images acquired during nighttime includes:

[0070] The feature selection module selects BT features from remote sensing images acquired during nighttime periods. 15 and BTD 7-14 Clustering features;

[0071] GMM clustering module, based on BT 15 and BTD 7-14 Clustering features are used to perform GMM clustering on the remote sensing image to obtain cloud detection results.

[0072] In some embodiments, a multispectral cloud detection method based on a sharpening module and a Gaussian mixture model has the following flowchart: Figure 1 As shown. This cloud detection method uses cluster feature analysis to select corresponding feature schemes for different time periods to achieve all-day cloud detection. If it is nighttime, BT is directly used. 15 BTD 7-14 The raw data is subjected to GMM clustering to obtain detection results. If it is daytime, the detection process is divided into three parts: first, R... 3+4 The data is then sharpened and clustered to detect thin clouds. Simultaneously, the original data is clustered to detect thick clouds. Finally, the clustering results of thick and thin clouds are overlaid to obtain the final detection result.

[0073] The specific implementation steps are as follows:

[0074] Step S1: Obtain the remote sensing image dataset;

[0075] In some embodiments, the input remote sensing image is a multispectral image with cloud interference. The single input image is converted into a single dataset, and then the single dataset is combined into a total cloud detection dataset.

[0076] Step S2: Preprocess the remote sensing image dataset to obtain the cloud detection dataset; further, the preprocessing includes:

[0077] The original remote sensing images are standardized to obtain preprocessed images; data augmentation techniques are used to enhance the original cloud detection dataset.

[0078] The raw remote sensing data is standardized using the following formula:

[0079]

[0080] In the formula, z is the value of the original remote sensing data after standardization, and X is the original data to be processed. mean X represents the average of the original data. std This represents the standard deviation of the original data.

[0081] Step S3: Select three clustering features from the cloud detection dataset;

[0082] Statistical analysis of the spectral data of Himawari-8 was performed using the probability density function (PDF) to generate PDF distributions of cloud and underlying surface pixel reflectance, brightness temperature, and brightness-temperature difference. The sum of the reflectance of channel 3 and channel 4 of the AHI on the Himawari-8 (Ri) was used as the quantization factor. 3+4 Brightness temperature (BT) of channel 15 15 ) and the brightness temperature difference (BTD) between channel 7 and channel 14. 7-14 There are three clustering features. The selection of clustering features is completed based on the distribution of PDFs with different clustering features of cloud and underlying surface pixels.

[0083] Step S4: Construct a cloud detection model using the cloud detection dataset and clustering features. The cloud detection model includes a feature selection module, a thin cloud sharpening module, a GMM clustering module, and an overlay module. For example... Figure 2 The diagram shown is a schematic of the cloud detection model framework, which is divided into daytime and nighttime periods.

[0084] Step 4.1, Feature Selection Module: Although the R of the cloud and the underlying surface... 3+4 The PDFs exhibit significant distributional differences, making them suitable for separating clouds from the underlying surface. However, during nighttime, clear visible light images are difficult to capture using the AHI in the visible light band. Therefore, different clustering feature schemes are used for cloud detection during the day and night: during the daytime, R... 3+4 During nighttime hours, BT will be used. 15 and BTD 7-14 .

[0085] Step 4.2, Thin Cloud Sharpening Module: This module sharpens the pre-processed image (detected during daytime) to obtain an image with enhanced thin cloud regions. Thin clouds are easily misidentified as non-cloud pixels. To improve the algorithm's detection performance for thin clouds, convolution operations are used to sharpen the remote sensing image, enhancing the spectral characteristics of thin clouds. The formula is shown below:

[0086] g(x,y)=f(x,y)*w(i,j) (2)

[0087] In the formula, g(x,y) is the enhanced image of the thin cloud region, f(x,y) is the pixel value of (x,y) in the preprocessed image, and w(i,j) is the sharpening convolution kernel.

[0088] In remote sensing images, the reflectance of the thin cloud region between thick clouds and the underlying surface changes rapidly, representing a high-frequency component of the image. The Laplacian operator, as a high-pass filter, can effectively enhance high-frequency signals in the image. Therefore, the Laplacian operator is chosen to enhance the thin cloud features. The definition of the Laplacian operator is as follows:

[0089]

[0090] By differencing the second derivatives of the Laplace operator in the x and y directions respectively, we can obtain:

[0091] ▽ 2 f(x,y)=f(x+1,y)+f(x-1,y)+f(x,y+1)+f(x,y-1)-4f(x,y) (4)

[0092] In the formula, ▽ 2 Here, f(x+1,y) represents the pixel value at (x+1,y) in the preprocessed image, f(x-1,y) represents the pixel value at (x-1,y) in the preprocessed image, f(x,y+1) represents the pixel value at (x,y+1) in the preprocessed image, and f(x,y-1) represents the pixel value at (x,y-1) in the preprocessed image.

[0093] The above formula is converted into a Laplacian convolution kernel, as shown below:

[0094]

[0095] After processing with the Laplacian operator, the spectral features of thin clouds are enhanced, but the thick clouds, which represent low-frequency signals, are removed. Therefore, the image processed by the Laplacian operator needs to be superimposed on the original image, as shown in the following formula:

[0096]

[0097] The resulting sharpened image can highlight the details of thin clouds while preserving the thick cloud portions of the original image. The sharpening convolution kernel used in this paper is derived from equation (6):

[0098]

[0099] Step 4.3, GMM clustering module: The feature selection module selects three clustering features and performs GMM clustering on the preprocessed image and the thin cloud region enhancement image respectively to obtain the clustering results of the main cloud part and the thin cloud part.

[0100] Gaussian Mixture Models (GMMs) can represent the probability density function of data through a linear combination of multiple Gaussian functions, and can simulate distributions of arbitrary shapes given sufficient samples. Therefore, GMMs can accurately describe the reflectivity and brightness temperature probability distributions of clouds and underlying surfaces. The formula for its probability density function is as follows:

[0101]

[0102] In the formula, x is a vector X = {x1, x2, x3, ..., x}. n Given n sample values ​​X, which contain K categories, the weight of the j-th category is β. j The function f(x) is the probability density function of a single Gaussian model, μ j σ j Let be the mean and variance of the j-th class.

[0103] Among them, the Expectation-Maximization (EM) method is selected for GMM parameter estimation.

[0104] Study area R 3+4 BT 15 BTD 7-14 The set is X = {x1, x2, x3, ..., x} n}, x i It is extracted from set X, where the elements are independent and conform to GMM. From equation (8), we know that the GMM density function contains the parameter set θ = {β, μ, σ}. 2 Let θ be the set of parameters. For a single Gaussian model, the parameters θ can be estimated using the maximum likelihood method, whose likelihood function is given by the probability density function:

[0105]

[0106] Expectation-maximization (EM) is a commonly used method for maximum likelihood estimation. It is suitable for data with hidden unknown variables and can accurately estimate the parameters of the optimal probability model. The specific steps are as follows:

[0107] Step 1: Based on prior knowledge, set the initial parameter set for each category of samples (i.e., hidden variables) as θ = {β, μ, σ}. 2}

[0108] Step 2: Calculate the probability that the data was generated by a Gaussian distribution:

[0109]

[0110] Step 3: Update μ and σ of each Gaussian model 2 :

[0111]

[0112]

[0113] Step 4: Update the β of each Gaussian model:

[0114]

[0115] Repeat steps two through four until the maximum likelihood value L(θ) of the parameter set stabilizes, at which point the algorithm is complete.

[0116] Among them, the setting of the K value before clustering in the GMM algorithm.

[0117] Determining the value of parameter K to ensure that the mixture model is both accurate and concise is the primary goal of model optimization. This invention uses the AIC and BIC criteria from information statistics to determine parameter K, and their expressions are as follows:

[0118]

[0119] In the formula: m is the number of parameters in the mixture model; n is the number of samples; lnL(x|K, β, μ, σ) 2 Let V be the maximum likelihood value. AIC V BIC The minimum value is used as the criterion for determining the optimal model.

[0120] First, select 4 groups of R. 3+4 Using remote sensing images as sample data, and setting the K value between 1 and 14, V was calculated for different K values. AIC V BIC It can be seen that as K increases, V AIC V BIC The overall trend is downward, and the downward trend gradually stabilizes when K>8. To further determine the value of K, K was successively set to 5, 6, 7, and 8, and the GMM cloud detection performance under different K values ​​was compared to determine the appropriate K value. As the K value increases, the number of missed cloud pixels gradually decreases. When K=7, the algorithm's detection results for both thick and thin clouds are relatively consistent with the actual cloud coverage. In addition, the difference between K=7 and 8 in the detection results of cloud edge details is minimal. Therefore, the choice of K=7 in this invention can ensure that the hybrid model is both accurate and concise.

[0121] After clustering, all subcategories are divided into two main categories: cloud and non-cloud. Cloud is labeled as 1 and non-cloud is labeled as 0.

[0122] Step 4.4, Overlay Module: The clustering results of the main cloud layer and the thin cloud layer are overlaid to obtain the cloud detection result, as shown in the following formula:

[0123]

[0124] In the formula, W1 is the clustering result of each sample point in the preprocessed image, W2 is the clustering result of each sample point in the thin cloud region enhancement image, and W3 is the final cloud detection result for each sample point.

[0125] Using the specific method described in Example 1, a comparison was made of the sharpening effects before and after sharpening of a collected optical remote sensing image, such as... Figure 3 As shown, it is evident that the thin clouds in the remote sensing image become much clearer after sharpening. Furthermore, the R-values ​​of the thin clouds in the image were extracted. 3+4 Perform PDF comparison, such as Figure 4 As shown in the image, sharpening reduces the R-value of thin clouds. 3+4 The significant increase makes it easier to separate clouds from the underlying surface, which will effectively reduce the algorithm's missed detection of thin cloud pixels.

[0126] Using the cloud detection model in Example 1, Himawari-8 images containing both thick and thin clouds from different regions and time phases were selected for cloud detection. The detection results are as follows: Figure 5 As shown in the diagram, the truth labels obtained through visual interpretation are represented by the yellow portion, while the detection results are represented by the white portion. The comparison demonstrates that the cloud detection method provided by this invention can accurately identify thick and thin clouds in images, exhibiting robust performance and possessing significant application value in the field of remote sensing image research.

[0127] Example 2

[0128] Secondly, this embodiment provides a multispectral cloud detection device based on a sharpening module and a Gaussian mixture model, including a processor and a storage medium;

[0129] The storage medium is used to store instructions;

[0130] The processor is configured to operate according to the instructions to perform the steps of the method according to Embodiment 1.

[0131] Example 3

[0132] Thirdly, this embodiment provides a storage medium on which a computer program is stored, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0133] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0134] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0135] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0136] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0137] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A multispectral cloud detection method based on a sharpening module and a Gaussian mixture model, characterized in that, include: Acquire the remote sensing image to be detected; The remote sensing image to be detected is input into a pre-built and trained cloud detection model to obtain the output cloud detection results. The methods for constructing and training the cloud detection model include: Step S1: Obtain the remote sensing image dataset; Step S2: Preprocess the remote sensing image dataset to obtain the cloud detection dataset; Step S3: Select features from the cloud detection dataset to obtain three clustering features, including: statistical analysis of the spectral data of Himawari-8 using probability density functions to form probability density function distributions of cloud and underlying surface pixel reflectance, brightness temperature, and brightness-temperature difference; and using the sum of reflectance R of channels 3 and 4 of the AHI on Himawari-8. 3+4 Brightness temperature BT of channel 15 15 And the brightness temperature difference (BTD) between channel 7 and channel 14 7-14 Three clustering features are used; the selection of clustering features is completed based on the probability density function distribution of different clustering features of cloud and underlying surface pixels. Step S4: Construct a cloud detection model using the cloud detection dataset and clustering features, wherein the cloud detection model includes a feature selection module, a thin cloud sharpening module, a GMM clustering module, and an overlay module; Different clustering feature schemes were used for cloud detection during the day and night: the remote sensing images were acquired during the daytime period, and R... 3+4 Clustering features; the remote sensing images were acquired during nighttime and analyzed using Bit-Based Image Processing (BT). 15 and BTD 7-14 Clustering features; The detection process for remote sensing images acquired during the daytime includes: a feature selection module, which selects R features from the remote sensing images acquired during the daytime. 3+4 Clustering features; a thin cloud sharpening module, which sharpens remote sensing images acquired during the daytime to obtain enhanced images of thin cloud regions; a GMM clustering module, based on R... 3+4 Clustering features: GMM clustering is performed on the remote sensing image to obtain the clustering results of the main cloud layer, and GMM clustering is performed on the enhanced image of the thin cloud region to obtain the clustering results of the thin cloud portion; Overlay module: the clustering results of the main cloud layer and the thin cloud portion are overlaid to obtain the cloud detection result; The detection process for remote sensing images acquired during nighttime includes: a feature selection module, which selects BT features from the remote sensing images acquired during nighttime. 15 and BTD 7-14 Clustering features; GMM clustering module, based on BT 15 and BTD 7-14 Clustering features are used to perform GMM clustering on the remote sensing image to obtain cloud detection results.

2. The multispectral cloud detection method based on a sharpening module and a Gaussian mixture model according to claim 1, characterized in that, In step S2, the preprocessing includes standardization processing; The standardization process for remote sensing images includes: In the formula, z is the value of the original remote sensing data after standardization, and X is the remote sensing data to be processed. mean X represents the average of the original data. std The standard deviation of the original data; The preprocessing also includes using data augmentation techniques to enhance the remote sensing image dataset.

3. The multispectral cloud detection method based on a sharpening module and a Gaussian mixture model according to claim 1, characterized in that, The thin cloud sharpening module uses the Laplacian operator to process remote sensing images to enhance thin cloud features. The image processed by the Laplacian operator is then superimposed on the original image to obtain an enhanced image of the thin cloud region. In the formula, g(x,y) represents the enhanced image of the thin cloud region, f(x,y) represents the pixel value of (x,y) in the remote sensing image, and w(i,j) represents the sharpening convolution kernel. For the Laplace operator; The Laplace operator is defined as follows:

4. The multispectral cloud detection method based on a sharpening module and a Gaussian mixture model according to claim 1, characterized in that, The Gaussian Mixture Module (GMM) uses a linear combination of multiple Gaussian functions to represent the probability density function of the data. The probability density function P(x|θ) is expressed as follows: In the formula, x is a vector X = {x1, x2, x3, ..., x}. n Given n sample values ​​X, which contain K categories, the weight of the j-th category is β. j The function f(x) is the probability density function of a single Gaussian model, μ j σ j Let be the mean and variance of the j-th class; the EM method is used to estimate the parameters θ = {β, μ, σ} of the GMM model. 2 }, using R 3+4 The parameter K=7 was determined using the AIC and BIC criteria for the remote sensing image sample data.

5. The multispectral cloud detection method based on a sharpening module and a Gaussian mixture model according to claim 1, characterized in that, Overlay module: Used to overlay the clustering results of the main cloud layer and the thin cloud layer to obtain the cloud detection result, as shown in the following formula: In the formula, W1 is the clustering result of the main cloud part of each sample point, W2 is the clustering result of the thin cloud part of each sample point, and W3 is the final cloud detection result of each sample point.

6. A multispectral cloud detection device based on a sharpening module and a Gaussian mixture model, characterized in that, Including processor and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1 to 5.

Images