A Multi-Feature Adaptive Cloud Detection Method Based on Satellite Cloud Images

Abstract

This invention belongs to the field of image data processing technology, specifically relating to a multi-feature adaptive cloud detection method based on satellite cloud images. The method includes: acquiring and preprocessing satellite cloud images to obtain grayscale images; extracting image feature vectors from the grayscale images, wherein the image feature vectors include contrast, edge density, texture complexity, mean brightness, and mean gradient magnitude; matching detection methods based on contrast features, edge density features, and texture complexity features in the image feature vectors; generating cloud masks based on the detection methods, wherein when K-means clustering is used as the detection method, the brightest cluster is selected to generate the cloud mask; when an adaptive threshold segmentation method is used, candidate masks are generated based on a first threshold and a second threshold formed by the standard deviation and mean of the pixel values ​​of the grayscale image, and the optimal candidate mask is selected as the optimal mask; and using morphological opening and closing operations to remove isolated noise points and fill holes in the optimal mask to obtain the final mask.

Description

Technical Field

[0001] This invention belongs to the field of image data processing or generation technology, and specifically relates to a multi-feature adaptive cloud detection method based on satellite cloud images. Background Technology

[0002] Existing satellite cloud image detection methods still have significant shortcomings in practical applications. First, many methods rely on fixed thresholds or single detection techniques, lacking adaptability to complex scenes. In actual observations, cloud brightness varies significantly with solar altitude angle, surface reflectivity, and imaging conditions. Thin clouds are often difficult to distinguish from high-albedo features (such as snow surfaces and sandy areas), and under nighttime imaging conditions, the image signal-to-noise ratio is low, making false detections or false negatives more likely. This lack of robustness directly weakens the reliability of the detection results.

[0003] Meanwhile, cloud imagery varies greatly across different scenarios, with varying contrast, edge sharpness, and texture complexity across different regions and time periods. Single thresholding methods or fixed clustering parameters often fail to accommodate these differences, performing well in some scenarios but failing in others, leading to unstable overall detection results and a lack of universality and adaptability. Furthermore, most existing methods only output binary cloud masks, lacking quality control and result interpretation mechanisms. For example, they lack metrics such as cloud coverage, connectivity, and brightness differences to evaluate the effectiveness and reliability of the detection results, making it difficult for users to determine the credibility of the results. Especially in business application scenarios, if the detection results lack interpretable quality metrics, they will struggle to meet the needs of subsequent quantitative analysis and automated processing. Summary of the Invention

[0004] The present invention is proposed based on the above-mentioned needs of the prior art. The technical problem to be solved by the present invention is to provide a multi-feature adaptive cloud detection method based on satellite cloud images to improve detection efficiency and detection accuracy.

[0005] To address the above problems, the technical solution provided by this invention includes:

[0006] A multi-feature adaptive cloud detection method based on satellite cloud images is provided, including: acquiring and preprocessing satellite cloud images to obtain grayscale images; extracting image feature vectors from the grayscale images, wherein the image feature vectors include contrast, edge density, texture complexity, mean brightness, and mean gradient magnitude; and a matching detection method based on the contrast feature, edge density feature, and texture complexity feature in the image feature vectors, including: when the contrast is greater than a contrast threshold and the edge density is greater than an edge density threshold, using K-means clustering with K=2; when the contrast is less than or equal to the contrast threshold and the edge density is less than or equal to the edge density threshold, if... If the texture complexity is greater than the texture complexity threshold, an adaptive thresholding method is used; if the texture complexity is less than or equal to the texture complexity threshold, K-means clustering with K=3 is used. A cloud mask is generated based on the detection method. When K-means clustering is used as the detection method, the brightest cluster is selected to generate the cloud mask. When the adaptive thresholding method is used, candidate masks are generated based on the first threshold and the second threshold formed by the standard deviation and mean of the pixel values ​​of the grayscale image, and the best one is selected from the candidate masks as the optimal mask. Morphological opening and closing operations are used to remove isolated noise points and fill holes on the optimal mask to obtain the final mask.

[0007] Preferably, the method further includes determining the result of the final mask, including calculating the cloud coverage rate index, connectivity index, area rationality score and brightness matching index of the final mask respectively, and integrating the above indicators to form a comprehensive quality score.

[0008] Preferably, the first threshold and the second threshold are formed based on the standard deviation and average value of the grayscale image pixel values, and are expressed as follows: , ,in, The first threshold, The average value of grayscale image I. Let be the standard deviation of the grayscale image I, where the first threshold is a conservative threshold and the second threshold is an aggressive threshold.

[0009] Preferably, generating candidate masks based on the first threshold and the second threshold includes generating a first candidate mask and a second candidate mask based on the first threshold and the second threshold; calculating a first cloud coverage rate and a second cloud coverage rate using the first candidate mask and the second candidate mask respectively; and preferentially selecting a mask that makes the cloud coverage rate within a reasonable range as the preferred mask.

[0010] Preferably, the step of generating the first candidate mask and the second candidate mask based on the first threshold and the second threshold includes: when the pixel value in the grayscale image is greater than the first threshold, the first candidate mask at the corresponding pixel is set to 1, otherwise it is 0; when the pixel value in the grayscale image is greater than the second threshold, the second candidate mask at the corresponding pixel is set to 1, otherwise it is 0; expressed as: , ,in, As the first candidate mask, As the second candidate mask, Let be the pixel value in the i-th row and j-th column of the grayscale image.

[0011] Preferably, the calculation of the first cloud coverage rate and the second cloud coverage rate using the first candidate mask and the second candidate mask respectively is expressed as follows: , ,in, For the highest cloud coverage, For the second cloud coverage, The number of cloud pixels in the first candidate mask. H is the number of cloud pixels in the second candidate mask, H is the height of the grayscale image, and W is the width of the grayscale image.

[0012] Preferably, a mask that allows cloud coverage to fall within a reasonable range is selected as the preferred mask, where the reasonable range is 10% to 80%, expressed as: ,in, To optimize the mask, This is the index corresponding to the minimum value.

[0013] Preferably, the step of using morphological opening and closing operations to remove isolated noise points and fill holes in the optimal mask to obtain the final mask is expressed as follows: , Where M is the final mask, For closing operations, For expansion, For corrosion, The structural element is a 5×5 ellipse. For opening operation, It is a 3×3 ellipse structuring element.

[0014] Preferably, the cloud coverage rate index is calculated based on the final mask and is expressed as: , ,in, The cloud pixel ratio, Let be the pixel value of the final mask at position (i,j); The coverage score characterizes how close the coverage is to the ideal value, where the ideal coverage is 0.3. H is the grayscale image height, W is the grayscale image width, and max() is the maximum value function. Connectivity analysis is performed on the final mask to calculate the connected components and cloud areas, expressed as: , , , , The connected component labeling yields the component set. , the number of clouds ; This refers to the nth cloud segment; Let n be the pixel area of ​​the nth cloud. For average area, It is a function with maximum value. The area is scored for reasonableness; a continuity score is calculated based on the number of cloud clusters, expressed as: in, The upper limit is empirical, and continuity ∈ [0,1] represents the continuity score, with larger values ​​indicating greater continuity. The brightness matching degree is calculated based on the final mask and expressed as: , , , ,in, The average brightness of the cloud area; Represented as a conditional mean function; is the average brightness of the non-cloudy area; d is the difference between the two, and clip(x,0,1) means clipping x to [0,1]. ∈[0,1] represents the brightness matching degree, and the larger the value, the better the brightness separability.

[0015] Preferably, a comprehensive quality score is formed by integrating cloud coverage, connectivity, area rationality, and brightness matching indicators, as follows: , ,in, The coverage score represents how close the coverage rate is to the ideal value, with an ideal coverage rate of 0.3. ∈[0,1]; To score for the reasonableness of the area, ∈[0,1]; These are the corresponding weights for coverage score, area rationality score, continuity score, and brightness matching degree, with default values ​​of 0.3, 0.2, 0.2, and 0.3, respectively.

[0016] Compared with existing technologies, this invention proposes a multi-feature adaptive cloud detection method based on satellite cloud images, which combines multi-feature driving, adaptive method selection, resolution perception, and quality judgment. This method analyzes the input normalized cloud image by calculating feature parameters such as contrast, edge density, texture complexity, mean brightness, and gradient magnitude. It can adaptively select appropriate detection methods under varying scene conditions and, combined with spatial resolution perception and quality judgment mechanisms, generate more stable, reliable, and interpretable cloud detection results. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings.

[0018] Figure 1 This is a flowchart illustrating the steps of a multi-feature adaptive cloud detection method based on satellite cloud images in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0020] In the description of the embodiments of the present invention, it should be noted that, unless otherwise explicitly specified and limited, the term "connected" should be interpreted broadly. For example, it can refer to a fixed connection, a detachable connection, or an integral connection; it can refer to a mechanical connection or an electrical connection; it can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in the present invention according to the specific circumstances.

[0021] Throughout the text, the terms “top,” “bottom,” “above,” “below,” and “on top” refer to the relative positions of components of the device, such as the relative positions of the top and bottom substrates within the device. It is understood that the device is multifunctional and independent of its spatial orientation.

[0022] To facilitate understanding of the embodiments of the present invention, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0023] Example 1

[0024] This embodiment provides a multi-feature adaptive cloud detection method based on satellite cloud images, such as... Figure 1 As shown.

[0025] The multi-feature adaptive cloud detection method based on satellite cloud images includes:

[0026] Acquire satellite cloud images and preprocess them to obtain grayscale images.

[0027] The preprocessing includes normalizing the satellite cloud image and mapping the grayscale values ​​of the normalized image to the [0,1] interval. Further, normalization can use the minimum and maximum values ​​of the entire image, or it can use percentile truncation (e.g., 1%–99%) to remove extreme value interference.

[0028] Among them, normalization can effectively eliminate dimensional differences and enhance the stability of feature calculation, providing stable input for subsequent feature calculation through the mapping of gray values.

[0029] Image features are extracted from grayscale images, including contrast, edge density, texture complexity, average brightness, and average gradient magnitude.

[0030] Contrast is calculated based on the standard deviation and average value of the pixel values ​​in the grayscale image.

[0031] Specifically, it is expressed as:

[0032]

[0033]

[0034] Where C represents contrast and I represents grayscale image. Let H be the grayscale value of the pixel in the i-th row and j-th column of the grayscale image, H be the height of the grayscale image, and W be the width of the grayscale image. The average value of the pixels in the grayscale image. The standard deviation of the pixel values ​​in the grayscale image. To prevent constants with zero denominators, take .

[0035] In noisy environments, this metric can be replaced by the median absolute deviation (MAD) to enhance robustness.

[0036] Canny edge detection is used to obtain the edges of grayscale images, and the edge density is calculated based on the total number of edge pixels and the total number of pixels in the image.

[0037] Specifically, the Canny edge detection operator is used to process the image. This operator identifies all object contours and texture boundaries in the image, outputting a binary image where black represents the background and white represents the edges. The number of white pixels in this binary image, i.e., edge pixels, is counted. By dividing the number of edge pixels by the total number of pixels in the image (H x W), a ratio between 0 and 1 is obtained, which is the edge density. A higher edge density indicates more lines and contours in the grayscale image, making the grayscale image appear potentially "sharper" or "complex."

[0038] Specifically, it is expressed as follows:

[0039]

[0040] Where E is the edge density. This is the binary result of Canny edge detection at pixel (x,y), where a value of 1 indicates an edge pixel.

[0041] The threshold parameter can be set fixedly or adaptively determined by the grayscale histogram quantile points to suit images with different brightness distributions.

[0042] The Laplacian operator is used to output the response map, and the variance of the response map is calculated to obtain the texture complexity.

[0043] The Laplacian operator is a filter that highlights rapidly changing areas in an image, and it is highly sensitive to details such as edges, speckles, and lines. In the response map output by the Laplacian operator, larger values ​​correspond to areas with rich detail. The variance is calculated to measure the dispersion of these values ​​in the Laplacian response map, thus revealing the overall texture of the grayscale image, i.e., its texture complexity: if the grayscale image has a complex texture and many details, the corresponding Laplacian response map will have many high-value points with dramatic changes, resulting in high texture complexity; conversely, if the grayscale image has a smooth texture, the values ​​in the Laplacian response map will be small and flat, resulting in relatively low texture complexity.

[0044] Specifically, texture complexity is expressed as:

[0045]

[0046] Where T is the texture complexity and Var(·) is the variance operator. (I) represents the output response map after processing the grayscale image I with the Laplacian operator. Let be the value of the pixel in the i-th row and j-th column of the Laplacian response graph. It is the mean of a Laplace response plot.

[0047] Furthermore, since the Laplacian operator is very sensitive to noise, the image is first Gaussian smoothed before calculation to filter out noise interference.

[0048] The average brightness value is calculated from the average grayscale value of all pixels in the grayscale image, directly reflecting the overall brightness of the grayscale image. The higher the value, the brighter the image, and the lower the value, the darker the image.

[0049] The average gradient magnitude is calculated based on the sum of the gradient magnitudes of all pixels and the total number of pixels in the image.

[0050] The Sobel operator is used to calculate the gradients of the grayscale image in the horizontal and vertical directions. For each pixel, the gradient magnitude is calculated based on the gradients in both directions. A larger gradient magnitude indicates a more drastic change in color or brightness, suggesting that the pixel is likely an edge. The average gradient magnitude is obtained by averaging the gradient magnitudes of all pixels.

[0051] Specifically, it is expressed as:

[0052]

[0053] Where G is the mean gradient magnitude. Indicates coordinates as The gradient of the pixels in the horizontal direction. Indicates coordinates as The gradient of a pixel in the vertical direction.

[0054] Furthermore, to ensure consistency across resolutions, the output of the Sobel operator is normalized.

[0055] The edge density E is represented by a binary value, and its result determines whether it is an edge. The mean gradient magnitude, on the other hand, characterizes the strength of the edge through its numerical value.

[0056] Finally, feature vectors of five dimensions of the grayscale image are obtained.

[0057] A matching detection method based on contrast features, edge density features, and texture complexity features in image feature vectors.

[0058] When the contrast is greater than the contrast threshold and the edge density is greater than the edge density threshold, K-means clustering is used, where K=3, to distinguish between thin clouds, thick clouds and the background.

[0059] When the contrast is less than or equal to the contrast threshold and the edge density is less than or equal to the edge density threshold, it is determined whether the texture complexity exceeds the texture complexity threshold. If it exceeds the threshold, an adaptive threshold segmentation method is used; if it is less than or equal to the threshold, K-means clustering is used, where K=2, thereby simplifying classification and enhancing stability.

[0060] Specifically, it is expressed as follows:

[0061]

[0062] in, The contrast threshold. The edge density threshold, This is the texture complexity threshold.

[0063] Cloud masks are generated based on detection methods. When K-means clustering is used as the detection method, the brightest cluster is selected to generate the cloud mask. When an adaptive threshold segmentation method is used, candidate masks are generated based on a first threshold and a second threshold formed by the standard deviation and mean of the pixel values ​​of the grayscale image, and the optimal mask is selected from the candidate masks as the cloud mask.

[0064] The first threshold and the second threshold are formed based on the standard deviation and mean of the pixel values ​​of the grayscale image, and are expressed as follows:

[0065]

[0066]

[0067] in, The first threshold, The average value of grayscale image I. Let be the standard deviation of the grayscale image I, where the first threshold is a conservative threshold and the second threshold is an aggressive threshold.

[0068] The candidate mask includes a first candidate mask generated based on a first threshold and a second candidate mask generated based on a second threshold. Specifically, it is expressed as follows:

[0069]

[0070]

[0071] in, As the first candidate mask, This is the second candidate mask.

[0072] The first cloud coverage rate and the second cloud coverage rate are calculated using the first candidate mask and the second candidate mask, respectively, as follows:

[0073]

[0074]

[0075] in, For the highest cloud coverage, For the second cloud coverage, The number of cloud pixels in the first candidate mask. This represents the number of cloud pixels in the second candidate mask.

[0076] The preferred mask is one that allows the cloud coverage to be within a reasonable range. For example, a reasonable range is 10% to 80%. The specific selection process is as follows:

[0077]

[0078] Among them, is the preferred mask, is the index corresponding to the minimum value.

[0079] The pixels with value 1 in the first candidate mask are divided into two categories: those that only belong to the first candidate mask and do not belong to the second candidate mask; those that belong to both the first candidate mask and the second candidate mask. When T2 < I(i,j) ≤ T1, it is 1 only in M1; when I(i,j) > T2, it is 1 in both M1 and M2. That is, the cloud region of M2 is completely contained within the cloud region of M1, and the cloud coverage rate R1 of M1 is always greater than or equal to the cloud coverage rate R2 of M2. The overlapping region (I(i,j) > T2) corresponds to the brighter cloud pixels in the image, and the difference region (T2 < I ≤ T1) corresponds to the medium-brightness cloud pixels, which are only marked as clouds in M1.

[0080] The obtained preferred mask may contain noise, such as misdetected cloud points; there may be holes inside the cloud, such as undetected cloud points, especially in thin cloud regions; the cloud boundary may be rough or broken. To improve the quality of the preferred mask, morphological opening and closing operations are used to remove isolated noise points and fill holes in the optimal mask, and the size of the structural element can be obtained by mapping from the physical size according to the sensor spatial resolution. Further, the structural element is elliptical, which can produce a smoother and isotropic (uniform in all directions) effect, and is more in line with the shape of cloud clusters in nature.

[0081] The morphological opening and closing operations are specifically expressed as:

[0082]

[0083]

[0084] Among them, M is the final mask, is the closing operation; is the dilation; is the erosion; is the structural element (ellipse, 5×5); is the opening operation; is the structural element (ellipse, 3×3). The final mask is a binary image matrix, and the element value is 0 or 1, where 1 represents cloud pixels and 0 represents non-cloud pixels.

[0085] Judging the result of the final mask includes calculating the cloud coverage rate index, connectivity index, and brightness matching degree index of the final mask respectively, and fusing the above indexes to form a comprehensive quality score.

[0086] Calculating the cloud coverage rate based on the final mask is specifically expressed as:

[0087]

[0088]

[0089] in, The cloud pixel ratio, This represents the pixel value of the final mask at position (i,j), specifically taking the value 0 or 1, where 1 represents a cloud pixel and 0 represents a non-cloud pixel. The coverage score represents how close the coverage is to the ideal value, which is 0.3.

[0090] Perform connected component analysis on the final mask, calculate the connected component and cloud area, and express it as follows:

[0091]

[0092]

[0093]

[0094] Among them, the connected component labeling yields the component set. , the number of clouds ; For the nth cloud segment; Let n be the pixel area of ​​the nth cloud. For average area, It is a function with maximum value. The area is scored as reasonable.

[0095] The continuity score is calculated based on the number of cloud clusters. Fewer connected components indicate a more continuous cloud cluster, specifically expressed as follows:

[0096]

[0097]

[0098] in, This is the upper limit based on experience, and can be adjusted according to the actual situation; continuity∈[0,1] represents the continuity score, and the larger the value, the more continuous it is.

[0099] The brightness matching degree is calculated based on the final mask, specifically expressed as follows:

[0100]

[0101]

[0102]

[0103]

[0104] in, The average brightness of the cloud area; Represented as a conditional mean function; it calculates the arithmetic mean of a set of pixels that satisfy a given condition. is the average brightness of the non-cloudy area; d is the difference between the two. clip(x,0,1) means clipping x to [0,1]. ∈[0,1] represents the brightness matching degree, and the larger the value, the better the brightness separation between clouds and non-clouds.

[0105] A comprehensive quality score is formed by integrating cloud coverage, connectivity, area rationality, and brightness matching metrics. Specifically, it is expressed as follows:

[0106]

[0107]

[0108] in, The coverage score represents how close the coverage rate is to the ideal value, with an ideal coverage rate of 0.3. ∈[0,1]; Score the area reasonableness (ideal average area is 1000 pixels). ∈[0,1]; These are the corresponding weights for coverage score, area rationality score, continuity score, and brightness matching degree, with default values ​​of 0.3, 0.2, 0.2, and 0.3, respectively.

[0109] This embodiment, based on the joint analysis of multiple features such as contrast, edge density, and texture complexity, can adaptively switch between binary clustering, ternary clustering, and adaptive threshold segmentation methods for different cloud map scenarios, thereby overcoming the problem of insufficient robustness of traditional fixed thresholds or single methods under complex conditions. This framework ensures that the algorithm maintains high detection stability even in scenarios with varying brightness, complex backgrounds, and low signal-to-noise nighttime scenes.

[0110] In the clustering process, this embodiment further introduces spatial coordinates and local gradient magnitudes as pixel features, enabling the algorithm to have a stronger ability to identify thin clouds and cloud boundaries. Through this feature enhancement design, cloud edges can be preserved more completely, the transition between cloud areas and non-cloud areas is smoother, and the phenomenon of boundary breaks and missed detection of thin clouds is significantly reduced.

[0111] This embodiment maintains consistent detection standards across different satellite platforms and observation resolutions, ensuring the transferability and applicability of detection results across platforms and mission scenarios.

[0112] Furthermore, this embodiment calculates multi-dimensional indicators such as cloud coverage, connectivity, and brightness matching degree while generating the cloud mask, and integrates them into a comprehensive quality score. Combined with a validity judgment threshold, an interpretable result evaluation and self-monitoring mechanism is constructed. This not only improves the credibility of the detection results but also provides users with intuitive quality judgment criteria, facilitating automated screening and monitoring in business applications.

[0113] In summary, the proposed multi-feature adaptive cloud detection method based on satellite cloud imagery, characterized by "multi-feature driven, method-adaptive, quality-interpretable, and resolution-aware," can stably output high-quality cloud masks in complex and ever-changing satellite cloud imagery scenarios, and achieves engineering-level self-monitoring capabilities through quality control. This method can be seamlessly integrated with subsequent optical flow estimation and cloud image extrapolation modules, thereby significantly improving the overall accuracy and reliability of the forecast.

[0114] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-feature adaptive cloud detection method based on satellite cloud images, characterized in that, include: Acquire and preprocess satellite cloud images to obtain grayscale images; Extract the image feature vector of the grayscale image, wherein the image feature vector includes contrast, edge density, texture complexity, average brightness, and average gradient magnitude; The matching and detection method based on contrast features, edge density features, and texture complexity features in image feature vectors includes: when the contrast is greater than a contrast threshold and the edge density is greater than an edge density threshold, K-means clustering with K=3 is used; when the contrast is less than or equal to the contrast threshold and the edge density is less than or equal to the edge density threshold, if the texture complexity is greater than the texture complexity threshold, an adaptive threshold segmentation method is used; if the texture complexity is less than or equal to the texture complexity threshold, K-means clustering with K=2 is used. Cloud masks are generated based on detection methods. When K-means clustering is used as the detection method, the brightest cluster is selected to generate the cloud mask. When using the adaptive thresholding method, candidate masks are generated based on a first threshold and a second threshold formed by the standard deviation and mean of the pixel values ​​of the grayscale image, and the optimal one is selected from the candidate masks as the optimal mask. The step of generating candidate masks based on a first threshold and a second threshold includes generating a first candidate mask and a second candidate mask based on the first threshold and the second threshold. The first cloud coverage rate and the second cloud coverage rate are calculated using the first candidate mask and the second candidate mask, respectively. Prioritize the selection of masks that allow cloud coverage to be within a reasonable range as the preferred masks; Morphological opening and closing operations are used to remove isolated noise points and fill holes in the optimal mask to obtain the final mask; The determination of the final mask result includes calculating the cloud coverage rate index, connectivity index, area rationality score and brightness matching index of the final mask respectively, and combining the above indicators to form a comprehensive quality score.

2. The multi-feature adaptive cloud detection method based on satellite cloud images according to claim 1, characterized in that, The first threshold and the second threshold are formed based on the standard deviation and mean of the pixel values ​​of the grayscale image, and are expressed as follows: , in, The first threshold, The average value of grayscale image I. Let be the standard deviation of the grayscale image I, where the first threshold is a conservative threshold and the second threshold is an aggressive threshold.

3. The multi-feature adaptive cloud detection method based on satellite cloud images according to claim 1, characterized in that, The generation of the first candidate mask and the second candidate mask based on the first threshold and the second threshold includes, When a pixel value in a grayscale image is greater than the first threshold, the first candidate mask at the corresponding pixel is set to 1; otherwise, it is set to 0. When a pixel value in a grayscale image is greater than the second threshold, the second candidate mask at the corresponding pixel is set to 1; otherwise, it is set to 0. This is represented as: in, Let be the first candidate mask corresponding to the pixel value in the i-th row and j-th column of the grayscale image. This is the second candidate mask corresponding to the pixel value in the i-th row and j-th column of the grayscale image. Let be the pixel value in the i-th row and j-th column of the grayscale image.

4. The multi-feature adaptive cloud detection method based on satellite cloud images according to claim 3, characterized in that, The calculation of the first cloud coverage rate and the second cloud coverage rate using the first candidate mask and the second candidate mask respectively is expressed as follows: in, For the highest cloud coverage, For the second cloud coverage, The number of cloud pixels in the first candidate mask. H is the number of cloud pixels in the second candidate mask, H is the height of the grayscale image, and W is the width of the grayscale image.

5. The multi-feature adaptive cloud detection method based on satellite cloud images according to claim 4, characterized in that, The preferred mask is one that allows the cloud coverage to be within a reasonable range, which is 10% to 80%, expressed as follows: in, To optimize the mask, As the first candidate mask, As the second candidate mask, This is the index corresponding to the minimum value.

6. The multi-feature adaptive cloud detection method based on satellite cloud images according to claim 1, characterized in that, The method of using morphological opening and closing operations to remove isolated noise points and fill holes in the optimal mask to obtain the final mask is expressed as follows: Where M is the final mask, To optimize the mask, For closing operations, For expansion, For corrosion, The structural element is a 5×5 ellipse. For opening operation, It is a 3×3 ellipse structuring element.

7. The multi-feature adaptive cloud detection method based on satellite cloud images according to claim 1, characterized in that, The cloud coverage rate metric is calculated based on the final mask and is expressed as follows: in, The cloud pixel ratio, For the final mask at position Pixel value at; The coverage score represents how close the coverage is to the ideal value, where the ideal coverage is 0.

3. H is the height of the grayscale image, W is the width of the grayscale image, and max() is the maximum value function. Perform connected component analysis on the final mask, calculate the connected component and cloud area, and express it as follows: Among them, the connected component labeling yields the component set. , the number of clouds ; This refers to the nth cloud segment; Let n be the pixel area of ​​the nth cloud. For average area, It is a function with maximum value. Score for reasonableness of area; The continuous score is calculated based on the number of cloud clusters and is expressed as follows: in, The upper limit of experience. It represents continuous fractions; the larger the value, the more continuous the fraction. The brightness matching degree is calculated based on the final mask and is expressed as follows: in, The average brightness of the cloud area; Represented as a conditional mean function; The average brightness value for non-cloudy areas; The difference between the two is This means limiting x to [0,1]. This represents the brightness matching degree; a higher value indicates better brightness separability.

8. The multi-feature adaptive cloud detection method based on satellite cloud images according to claim 7, characterized in that, A comprehensive quality score is formed by integrating cloud coverage, connectivity, area rationality, and brightness matching metrics, and is expressed as follows: in, The coverage score represents how close the coverage rate is to the ideal value, with an ideal coverage rate of 0.

3. ; To score for the reasonableness of the area, ; These are the corresponding weights for coverage score, area rationality score, continuity score, and brightness matching degree, with default values ​​of 0.3, 0.2, 0.2, and 0.3, respectively.

Images